Month: September 2025

Why Power is Critical for Data Centers and their Hyperscaler Customers

Posted on by io-fund

Hyperscalers are spending hundreds of billions of dollars annually on AI data center capex, from physical data center space, GPUs and servers, hardware and networking. With these substantial sums flowing towards GPUs that are now being refreshed on an annual cadence, the impetus for hyperscalers, neoclouds and other cloud providers turns to how quickly these GPUs can be energized and deployed, to maximize the period of returns before the next generation comes online. 

If a company like Microsoft buys tens of billions of Nvidia’s Blackwell GPUs, the longer the massive investment in GPUs waits for power, the more delayed that revenue and profits become. In turn, this plays into market share as competitors who can energize GPUs faster will have a critical head start over those that are waiting for power. This is simple in concept, yet the lack of power having vast consequences cannot be overstated if you combine the sheer size of investments being made in AI alongside fierce, heightened competition.

AI is a spending race, but this means it is at the core, a power race. It does not matter if a hyperscaler spends tens of billions more on capex if it cannot secure the power to stand up new data center infrastructure to then deploy those GPUs immediately. The AI market is officially moving from being compute constrained to being power constrained, and this shift is important for I/O Fund members to prepare for.

We were among the first research companies to cover this topic in June of 2024, many quarters before the problem became well-known. We furthered this by investing early in a Bitcoin miner and one of the year’s highest-performing AI energy stocks. When we say we work hard to be early to trends for the benefit of our Members, we mean exactly that.

Given that we are soon approaching the moment when AI becomes (painfully) power constrained, we want to revisit this trend by examining how hyperscalers and others are powering new data centers. Below, we also look at the methods that can provide power the quickest, along with insights into location and costs, and more. As you can imagine, this analysis will help inform additional stock ideas as we position for 2026 and beyond.

Why Power is Critical, and Why it Will Continue to Be

More than one year ago, we first discussed how quickly power consumption was increasing with new GPUs in the analysis AI Power Consumption: Rapidly Becoming Mission-Critical. This trend is set to continue with Nvidia pushing towards an ultimate goal of super-sized 1MW server racks, or 8x more than GB200 racks. 

Nvidia’s Blackwell lineup already brings a significant increase in power consumption, nearly double the H200’s 70 kW at 120 kW for the GB200 NVL72 and 140 kW for the upcoming GB300 racks.  

Beyond Blackwell, Nvidia’s future design lineup shows continual increases in power consumption. Its Rubin generation is expected to boost thermal design power (TDP) by 50% over Blackwell at up to 180 kW per rack, with the upgraded Vera Rubin then doubling this to 360 kW per rack by 2027. In its largest configuration, the Vera Rubin NVL576, dubbed the ‘Kyber’ rack, could draw as much as 600 kW (0.6 MW), or 5x that of the GB200 NVL72 in just a two-year design timeframe. These figures do not include networking, interconnects, cooling and other hardware, which will further boost power draw per rack.

This rapid increase in power consumption per GPU generation is critical, as existing infrastructure is simply unable to meet these escalating power demands. For example, Applied Digital pointed out that nearly 70% of current data centers “contain racks requiring between four and nine kW of power, and less than two percent of data centers have racks with greater than 50kW.” For comparison, Super Micro’s GB200 NVL72 SuperCluster requires 132kW, while the upcoming Kyber rack could more than quadruple that to 600kW. Because Blackwell-based servers are now 15x to 30x the power density, cooling and power delivery strategies have to be redesigned, as liquid cooling now becomes a necessity.  

This sharp rise in power density means current infrastructure may be unable to transition from 4-9kW racks to >130kW racks without incurring significant retrofitting costs, while building new infrastructure bypasses that hurdle and allows for optimization for high-powered racks.

The push to 600kW racks over the next few years means this is not a transient problem, rather it is one that the industry will continue to face, meaning continuous new construction may be needed to handle surging power demand. For example, Vantage’s upcoming 1.4 GW campus in Texas for Oracle is designed for ultra-high density racks up to 250kW, yet this will not be enough power to able to host NVL576 racks in just two to three years’ time. Additionally, a former Microsoft Azure AI executive reportedly said he estimated that the “requirements in terms of power for the data center would probably at least double every three years and maybe exponentially so over a period of time,” further reinforcing this.

How Much Does 1 GW Cost?

At this point, we can reasonably estimate costs to build 1 GW of new AI data center capacity, though ultimately, this will depend upon location (given some of the disparities in construction costs regionally), as well as hardware, as newest gen-GPUs will require more advanced ancillary equipment and cooling tech than older, less power generations (such as Hopper).

For example, GPUs and necessary ancillary hardware including InfiniBand/Ethernet, cooling and other equipment is likely in the range of $18 million to $24 million per MW for high-end GPUs such as Nvidia’s B200s.

IREN disclosed that it purchased ~4,200 B200s and ancillary equipment for $193 million, which calculates out to the mid ~$21 million per MW range based on total power draw of 1.93kW and a 1.1 PUE. For AMD’s Instinct MI350X GPU, which consumes ~1kw before ancillary equipment, costs could be towards the lower end of range, given its pricing at ~$25,000 per GPU, versus ~$30,000 to $40,000 for the B200. Pricing and power needs for custom silicon are much more opaque, though it is likely that these chips come at a greater discount compared to Nvidia and AMD’s leading GPUs.

Translating this to GW-scale, IREN’s 50MW facility can host >20K B200 GPUs, which, based on its purchase pricing, translates out to $21.6 billion per GW. For its planned 2GW Sweetwater campus, IREN claims it can support ~600K GB300 GPUs, which, at ~$80,000 per GPU, would cost $48 billion, or $24 billion per GW.

Cushman & Wakefield estimates new data center construction costs per MW in the range of $9-15 million across key markets, averaging $11-13 million. This aligns with estimates from CBRE for $10-14 million per MW, though costs can reach $16-20 million per MW in certain cases (or higher).  This is simply for the ‘powered shell’, or the core building that has grid connection and has power in place, but has not been outfitted with racks or servers per tenant specifications.

Cushman & Wakefield estimates new data center construction costs per MW in the range of $9-15 million across key markets, averaging $11-13 million.

Source: Cushman & WakefieldCushman & Wakefield

So, assuming construction costs of ~$12 to $14 million per MW, total costs per MW for new facilities, GPUs and ancillary equipment are estimated between $30 to $38 million per MW. In total, this projects to roughly $30 to $38 billion per GW to build a data center from the ground up.In total, this projects to roughly $30 to $38 billion per GW to build a data center from the ground up. This is backed by Microsoft’s early 2025 announcement for ~$80 billion in spending for AI data centers in fiscal 2025 corresponding to >2 GW of new capacity additions.

Putting This in Terms of Capex

While capex is ultimately one of the more important figures for AI investors to track, it’s necessary to put in perspective how much capacity this capex correlates to, considering the tight grip power has over the industry.

Consider that the largest tech firms – Microsoft, Amazon, Alphabet, Meta and Oracle – are on track to likely spend upwards of $380 billion this year, up more than 50% YoY, and closer to $500 billion in 2026, with the majority going towards data centers. This puts total spending in 2025 and 2026 potentially as high as $880 billion, not even including CoreWeave, Nebius, xAI and others building out capacity.

Running off this calculation that each GW of new capacity could cost between $30 to $38 billion from the ground up, for the powered shell, GPUs and related hardware, we can reasonably estimate how many GW that Big Tech could bring online based on capex spend.

Assuming ~70% of capex goes directly towards data center capacity, given that hyperscalers and neoclouds alike continue to talk about supply constraints and strength of demand, this projects Big Tech could bring ~7-9 GW online with 2025 capex of $380 billion. For 2026, this would project ~9-12 GW of new capacity coming online, or cumulative total of ~16-21 GW.

Now, assuming closer to ~85% of capex goes towards new data center capacity, as Microsoft, Amazon and Alphabet now have a new cloud contender to deal with in Oracle (who also must build capacity rapidly to meet nearly half a trillion in RPO), this projects much larger buildouts. For 2025, this would estimate ~8.5-11 GW of new capacity and another 11-14 GW in 2026, or cumulative needs of nearly 20-25 GW. At midpoint, this would be ~4 GW higher than the prior assumption.

Forecasts All Point to Surging Capacity & Demand Growth

As we had covered in our free newsletter, Nuclear Power Emerging as a Clean AI Data Center Energy Source, data center power demand is expected to grow at an accelerated clip through the end of the decade and beyond, with more powerful GPUs and surging growth in inference two main drivers.

For example, Boston Consulting Group forecasts 45 GW of growth in global data center power demand in just three years, from 82 GW in 2025 to 127 GW by 2028. This represents an acceleration from a 12% CAGR from 2020 to 2023, to a 16% CAGR from 2023 to 2028.

Chart showing acceleration in global data center power demand from 71 GW in 2024 to 127 GW by 2028, driven by generative AI and inference.

On the other hand, McKinsey projects data center capacity will rise ~2.5x to 219 GW by 2030, up from 82 GW in 2025, with AI contributing 70% of that demand. This corresponds to total capacity growth of 137 GW over the next five years, with 112 GW coming from AI.

Goldman Sachs estimated global data center power usage at 55 GW in early 2025, far below BCG’s 82 GW figure. However, GS projects power usage to reach 84 GW in 2027 and increase further to 122 GW by 2030, corresponding to total growth of just 67 GW. However, considering 2025 and 2026 capex spend could support >20 GW of new capacity, this forecast may understate the pace of capacity growth.

This is quite the wide range of projected capacity growth over the next three to five years. But, more importantly, what level of capex does this require? Given the prior calculations for each GW to cost between $30 to $38 billion from the ground up (not accounting for future generation chips), building out 67 to 112 GW by 2030 could necessitate anywhere between $2 trillion to $4.3 trillion in capex over the next five years. McKinsey estimates that spending could reach as much as $6.7 trillion through 2030, which may support up to ~176 GW of new capacity.

At current projections, Big Tech is expected to spend nearly $1.2 trillion on capex from 2024 through 2026, meaning that these projections are well in the realm of possibility based on current spending trends.

Data Center Spending Up 30% YoY

The data center market in the US remains heavily constrained as high levels of demand are outstripping surging supply, even as data center construction reached $40 billion annualized in June, up 30% YoY and a new record. Primary market supply, including key regions such as Northern Virginia, Atlanta and Dallas-Fort Worth, rose 17.6% from H2 2024 and 43.4% YoY to a record 8.16 GW in 1H 2025, per CBRE.  

Also, we see the highest amount of new data center construction in markets offering the lowest build costs. This suggests hyperscalers and other providers are seeking the cheapest paths for new capacity while aiming to bypass long connection wait times with behind-the-meter or off-grid sources, such as on-site gas turbines.

CBRE noted that Northern Virginia, Atlanta, San Antonio and Dallas-Forth Worth were the four markets with the highest amount new construction in the first half of 2025. The four combined for 4.86 GW of total capacity under construction, or nearly 82% of total new construction activity across primary and secondary markets. Interestingly, Seattle, which has one of the longer times to power at ~48 months, versus 36 months for Dallas, has just 9.5 MW under construction, while Chicago, with some of the highest construction costs, only has 244 MW under construction.

More on Location – Latency & Climate are Factors to Consider

There are more nuances about location beyond time to power and construction costs that factor into site selection, attractiveness and ultimately where hyperscalers, neoclouds or even miners choose to build. Two primary factors include latency and climate.

The map below shows data center presence by city, with major markets in dark blue, emerging markets with more abundant power in blue, and secondary markets in gray. Many of the recent headline-grabbing builds can easily be placed into either primary or emerging market locations.

Source: McKinseyMcKinsey

The reason that these new, larger builds are often located in primary and emerging markets is not simply because of strong existing infrastructure or more power availability, but also for proximity to key cities with low-latency. For example, TeraWulf’s New York site offers sub-7ms latency to New York and <8ms to Boston, while Galaxy’s Helios data center in Texas offers <15ms latency to Dallas. Research from Applied Digital found that Stargate’s Abilene site and Northern Virginia both have <80ms latency to major cities across the US, with 100ms feeling ‘instantaneous’ to users.

However, data center vacancies dropped to a record low 1.6%, signaling strong demand from hyperscaler and AI customers, who continue to lock up supply quickly. CBRE added that nearly three-quarters of under-construction capacity of 5.25 GW was already pre-leased by hyperscalers and other providers aiming to secure capacity amid land and power constraints.

Industry Executives See Power as a Primary Constraint

Commentary from executives at hyperscalers, neoclouds, Bitcoin miners, colocation providers and commercial real estate firms all point to power as a key constraint (and consideration) facing the market this year and next:

CBRE said in its H1 2025 Data Center Report that “power availability and infrastructure delivery timelines remained the most decisive factors shaping site selection, leasing activity and pricing across all major U.S. markets.”

Equinix executives stated that “the amount of power we need isn't sitting around on the grid. And so we are planning, and I think most people in the room that are doing data center development are ensuring you have clear line of sight to that power before you take down any land or plan any data center capacity.” Execs also noted that the “reality of it is there [are] constraints in the marketplace, whether that's power availability in the key metros where we're looking to operate…”

A survey by Bloom Energy of 44 hyperscaler and colocation developers found that availability of power was the number one consideration for new site selection, with 84% of respondents placing that in the top 3 with an average rating of 7.8 out of 10.

TeraWulf CEO Paul Prager said he believes there is “a good argument that the market might even be tighter in 2026 than in 2025 given ongoing power constraints and rising hyperscaler CapEx.”

Amazon CEO Andy Jassy said that “you see some of the constraints and they kind of exist in multiple places, [but] the single biggest constraint is power.” Microsoft CEO Satya Nadella said Microsoft needs “power in specific places so that we can either lease or build at the pace at which we want.”

Google Cloud’s Thomas Kurian explained that “as you get these more powerful chips, they also take a lot more power. And power is, in many cases, a short resource.” Arm’s CEO Rene Haas has said that without improvements in efficiency, "by the end of the decade, AI data centers could consume as much as 20% to 25% of U.S. power requirements. Today that’s probably 4% or less."

Types of Data Center Builds, and Where Hyperscalers are Currently Going for Power

There are four common types of hyperscale/AI data center builds that are prevalent in the market: greenfield, build-to-suit, colocation, and brownfield. Each offers a different set of pros and cons as it relates to time to power, customization ability, and cost.

  • Greenfield: Refers to a hyperscaler or CSP owning the land, power and building the data center facility and infrastructure from the ground up. Greenfield builds offer the highest degree of customization ability over every aspect of the facility from power delivery to rack placement, though it comes at a much higher cost and often with the longest timelines to completion due to permitting, site selection, and grid connection.
  • Build-to-suit: Refers to a developer owning the land and securing the power for the facility, and constructing the data center tailored to the needs of the hyperscaler or  CSP buying the capacity, typically via long-term leases. Build-to-suit data centers offer hyperscalers design flexibility without taking on the higher capex needs of a greenfield build.
  • Colocation: Refers to when a hyperscaler or CSP rents capacity (racks, power and cooling infrastructure) from a provider, offering a path to meet quick capacity needs though with no input on facility design.
  • Brownfield: Refers to retrofitting existing infrastructure to meet hyperscaler/AI needs, such as what Bitcoin miners are pursuing with existing mining infrastructure. Brownfield builds are often cheaper and faster than greenfield, but can be limited in terms of power and space by what is present with the existing infrastructure.

Meta’s 5GW Hyperion Campus

Meta is undertaking some of the industry's largest data center projects to support its AI superintelligence quest, with its greenfield Prometheus data center in Ohio expected to be the first 1GW campus to come online in 2026. This is followed by its Hyperion campus in Louisiana, expected to have an initial capacity of 2GW before scaling to 5GW over several years.

Meta’s Hyperion campus is expected to cost $50 billion (~$10 million/MW at full scale), with the social media giant securing $29 billion in financing from PIMCO and Blue Owl to fund the project. The facility’s power requirements are immense, equivalent to approximately 4 million homes.

To power this, Entergy is constructing three new combined-cycle gas turbines coming online in late 2028 to provide an initial 2.3 GW of power, while building new substations and installing new transmission lines. Entergy also may add an additional 2 GW of solar power to support the expansion of the campus towards 5 GW. Meta is said to be covering the $3.2 billion cost of the turbines for the first 15 years, while also pledging to bring 1.5GW of solar and battery power to the grid.  

Amazon Strikes $18B Nuclear Deal with Talen

Amazon made a splash with the largest ever nuclear power PPA in history, purchasing 1.92 GW of nuclear power from Talen Energy to support colocated AWS data centers in Pennsylvania. Under the deal, Talen will ramp to full volume no later than 2032, with the deal extending through 2042 with options for further extension.

The two had initially attempted to go with a ‘behind-the-meter’ deal where Amazon would purchase power directly from Talen and bypass the grid, though FERC had blocked this in late 2024 on concerns about grid reliability and upwards pressure on consumer rates. Under the new $18 billion contract, the colocated power agreement “will transition to a ‘front-of-the-meter’ arrangement after the completion of transmission reconfigurations expected in the spring of 2026,” after which the plant will provide power to PJM’s grid with Talen acting as the supplier to Amazon and PPL responsible for transmission and delivery.

More broadly speaking, Amazon has signed utility-scale solar and wind deals globally, while also supplementing data center sites with on-site solar to augment grid power. Amazon is also said to be exploring fuel cell and gas turbine use at facilities to have more direct control over power.

Microsoft Adds More than 2GW of New Capacity

At the start of 2025, Microsoft disclosed that it was planning to spend roughly $80 billion through the end of its fiscal year in June $80 billion on AI-enabled data centers, to help ease capacity constraints and meet strong demand. In July, CEO Satya Nadella announced that Microsoft “ stood up more than 2 gigawatts of new capacity over the past 12 months alone,” marking a rather aggressive capacity expansion considering the company was said to have ~5GW at its disposal in early 2024.

Microsoft is continuing to build out its data center footprint, announcing a $4 billion additional investment in Wisconsin to house “hundreds of thousands” of Nvidia’s GPUs in a new facility, joining a $3.3 billion data center announced last year. Aligning with Nadella’s comments, Microsoft is also committing to leasing new capacity, with a mega build-to-suit deal with Nebius in New Jersey and a $6.2 billion colocation deal with Nscale and Aker in Norway.

Nebius’ new data center in New Jersey is being constructed by DataOne, who said in March that it would deliver the first phase of the data center in 20 weeks via a behind-the-meter solution. In Norway, Aker says that the new five-year deployment beginning in 2026 is powered by secured grid capacity and 100% renewable energy.

Alphabet Procuring Clean Energy to Support Data Centers

Alphabet is progressing towards its 24/7 Carbon-Free Energy by 2030 target, where each data center is backed 24/7 by clean energy. To support this, Alphabet said in its 2025 Sustainability Report that it procured 8 GW of clean energy primarily via long-term PPAs, that, once operational, “could generate nearly four times more electricity than our incremental load growth from 2023 to 2024.”

These include solar, wind and battery storage, as well as future investments for advanced geothermal or small nuclear reactors. The company said that in 2024, these PPAs brought 2.5 GW of clean energy to the grid to support its data centers.

Oracle Signs Deal with Bloom Energy for On-Site Power, Backs 1.4GW Natural Gas Data Center

Oracle has made a handful of different moves on the power side, signing a deal with Bloom Energy for near-immediate fuel cell deployment while also backing Vantage’s new 1.4 GW gas-powered West Texas data center.

Bloom is working to deploy its fuel cell tech at select Oracle Cloud Infrastructure (OCI) data centers in the US, with deployments expected to occur through late July to late October 2025. However, neither Oracle nor Bloom confirmed the scope, size or value of these deployments for on-site power generation.

Oracle is backing Vantage’s upcoming 1.4 GW data center, which is expected to see the first of ten buildings go live in the second half of 2026, built to handle next-gen ultra-high-density racks up to 250kW (versus ~130kW for Blackwell). Per Bloomberg, Oracle is set to spend more than $1 billion annually to power the campus with gas generators rather than waiting for a utility connection.

Crusoe Adds Natural Gas Turbines to Power Data Centers

Crusoe, developer of Stargate’s Abilene data center, has partnered with investment firm Engine No.1 to access 4.5 GW of power from seven of GE Vernova’s natural gas turbines that Engine No. 1 and Chevron’s joint venture purchased earlier this year.

These turbines are expected to bypass the grid and provide power directly to Crusoe’s data center campuses, with energy supply likely in place by 2027. However, Crusoe did not disclose whether this power would be directed to Stargate’s Abilene data center, as it is reportedly in discussions with multiple hyperscalers about where this power may be deployed.

xAI Tapping Gas Turbines for Colossus

xAI is powering its Colossus supercomputer via gas turbines, having more than doubled its number of turbines from 15 to 35 in April this year. The gas turbines have a combined capacity of ~422MW, per the Southern Environmental Law Center (SELC), though the group alleges that only 15 of these turbines are permitted. In May, the SELC also noted that xAI was aiming to add between 40 and 90 more turbines for its second Colossus data center in Memphis, raising concerns about pollution and health risks to nearby civilians.

Meeting Future Hyperscaler Power Needs

The most pressing question is, where does the industry go from here for data center power? Future hyperscaler needs continue to grow, with Amazon, Microsoft, Alphabet and Oracle combining for more than $1 trillion in RPO that will (hopefully) convert to revenue, while the broader industry could see anywhere between 67 to 112 GW (or more) of growth through 2030.

Utilities Expect Power Delivery Far Behind Hyperscaler Expectations

There exists a significant disconnect between when hyperscale and colocation developers expect to have site power, and when utilities expect to be able to deliver said power, according to research from Bloom Energy from April. Therefore, connecting new data centers to the grid in quick fashion may not be the most feasible option for hyperscalers looking to deploy gigawatts of capacity quickly, and instead, alternative power sources may be in higher demand.

For example, across the board, developers are expecting to have power delivered by 2027 on average, with most regions seeing expectations as early as late 2025. This is likely driven by consistent strong demand for AI infrastructure services, as new capacity will allow hyperscalers to meet more demand and drive more revenue.

Yet, utilities do not expect to deliver power in most of these primary and secondary markets until 2028, at the earliest, with Austin/San Antonio seeing one of the longest timelines at mid-2029.

Source: Bloom EnergyBloom Energy

This is supported by research from TD Cowen regarding grid connection timelines for new data centers, which span anywhere from 36 months to 48 months in these markets.

Grid connection timelines for new data centers in major markets span anywhere from 36 months to 48 months.

TD estimates connection timelines in Chicago at ~36 months and San Antonio at ~42 months, aligning with responses from Bloom’s survey. There has also been discussion regarding even longer timelines; in 2024, Bloomberg reported that utility Dominion Energy said >100MW data centers in Virginia were facing up to seven year wait times for new connection hookups.  

Primary Market Grids at Risk of Shortfalls

Many primary markets like Northern Virginia, Texas, and Chicago are not necessarily the best equipped to handle surging data center demand, as the power grid in these regions is at elevated risk of supply shortfalls during extreme conditions.

For example, PJM’s grid will be at elevated risk from 2026 onwards, along with ERCOT in Texas, whereas the upper Midwest (MISO) is already at elevated risk. For example, ERCOT projects peak net loads may outpace generation capacity as soon as 2026. Thus, interconnection delays for its grid (and MISO) could stretch to up to 5 years to allow for more generation capacity to come online to avoid further stress.

Source: NERCNERC

This means that building out gigawatts of new capacity in at risk regions may place more emphasis on behind-the-meter deals, on-site generation to minimize strain on the grid, or adding back-up power sources to allow for shifting off the grid when needed.

For example, Oracle is said to be paying ~$1 billion annually for gas generators to power Vantage’s upcoming 1.4 GW data center in West Texas instead of waiting for the grid to be ready, or an extra 2.5% of its operating expenses each year for a single site. Microsoft likely selected Nebius in New Jersey for its ability to deliver hundreds of MW of capacity in ~12 months by going behind-the-meter. xAI stood up its Colossus cluster with 100K GPUs in just 4 months with gas turbines.

Where Does the Power Come From?

There are multiple different ways that hyperscalers, neoclouds and developers can get power to data centers to meet upcoming demand growth over the next few years, each offering its own benefits and drawbacks.

Grid interconnection: This is when data centers connect to the power grid under standard service, providing access to flexible power needs with no additional capex and a wide range of power generation options, including renewables. However, grid interconnection requests are often the longest time to power, ranging from three to seven years for hyperscale data centers in most key markets.

Behind-the-meter: BTM refers to when data centers connect directly to the power source and bypass the grid (meter), which can offer significant time advantage with stand-up times often in the range of several months to a year, along with cost savings from buying power direct versus at retail price. It also gives data centers more control over the power as well as a lower risk for disruption from grid outages. BTM deals can be sourced from multiple different power sources, such as solar, wind or nuclear.

On-site power generation: With on-site power, data centers will install their own power source within the facility grounds, also offering a relatively quicker time to power of a few months to over a year. On-site power can come in many forms, such as Bloom’s fuel cells, natural gas turbines or generators such as those from GE Vernova or Caterpillar, and in the 2030s and beyond, potentially small modular nuclear reactors. Bloom Energy’s survey found that 38% of data centers expect some form of on-site power by 2030, up from 13% last year.

  • Natural gas turbines/generators: NG is a widely available fuel source with a broad pipeline in the US, offering continuous power to data centers. Turbines can come in a range of sizes and be easily deployed, such as Caterpillar subsidiary Solar’s SMT-130 turbines that xAI is using, or GE Vernova’s LM2500XPRESS that Crusoe is using, scaling up to 1GW capacity. Notably, NG turbines could help meet substantial future demand, as GE Vernova is expanding manufacturing in South Carolina to be able to ship 20 GW worth in 2027. Large (>225MW) turbines are reportedly sold out over the next three years.
  • Fuel cells: Similar to NG, fuel cells can be quickly deployed (in as little as three months per Bloom and Oracle’s deal), and provide continuous power for operations. Due to be a relatively newer tech, FCs can come at a higher cost than NG, but without the related emissions. Bloom is planning to double its FC manufacturing capacity to 2GW in 2026 to meet rising on-site power demand.
  • Small modular reactors: SMRs are drawing more interest for future demand needs, as commercialization at scale is not likely until 2030 or beyond. Google is working with Kairos to bring 0.5 GW of SMR capacity online from 2030 through 2035, while Oklo and NuScale are progressing with commercialization plans and a long-term combined ~20 GW backlog.

Retrofitting existing infrastructure, ie. Bitcoin mining: This leverages existing infrastructure with secured power to the building, offering quick delivery times as short as a few weeks to a year, depending on cooling, flooring or other upgrades needed. While this method can offer quick time to power for >100MW sizes with low latency, low electricity costs and cooling expertise, miners are rather capital constrained and may be unable to build out capacity beyond what is currently in their pipelines. Miners have been attracting substantial deal activity, primarily from neoclouds, from an ability to deliver larger chunks of power quickly, with capex costs well below greenfield builds.

Where Hyperscalers May Go for Power Needs

Power is becoming one of the largest constraints for hyperscalers, neoclouds and developers, as surging power consumption with each GPU generation is necessitating new infrastructure to handle these increasingly power dense racks. The industry is racing to deploy hundreds of billions of dollars’ worth of AI servers and related hardware before the next refresh cycle to drive growth and maximize ROI.

As a reminder, Big Tech capex implies potentially more than 20 GW of new capacity will come online this year and next, while industry forecasts suggest global demand could rise between 67 to 112 GW by 2030.Traditional grid interconnections face several years’ worth of delays and cannot keep pace with how quickly hyperscalers want to stand up new data centers, putting the emphasis on alternative strategies to secure power.

Gas generators and turbines are emerging as a popular choice and likely will remain popular with tens of GW of manufacturing capacity coming online in 18 months along with readily available fuel. Fuel cells can also help meet near-immediate needs with rapid deployment timelines, though capacity is limited to only a portion of expected demand growth over the long run. Bitcoin miners have also found a role in meeting near-term demand, yet availability capacity is thinning out quickly following multiple long-term deals.  

The I/O Fund is conducting deep-dive research on the energy sector as part of our ongoing focus on AI infrastructure. Our team is evaluating leading energy stocks that could play a pivotal role in powering the next wave of data center and AI growth. The results will be featured in our Top 10 New Ideas report, which will be delivered exclusively to Discovery Members by mid-October. Learn more here. Top 10 New Ideas report, which will be delivered exclusively to Discovery Members by mid-October. Learn more here.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

Meta Stock Emerges as a Strong Mag 7 AI Leader

Posted on by io-fund

The AI frenzy has investors fixated on revenue growth as proof of returns on AI spending that can be as high as $100 billion per year, depending on the company. Yet, Meta is proving that a stronger signal of AI strength may be found further down the income statement — in the bottom line.  

Without much fanfare, Meta put up a solid earnings report this past quarter as ad impressions rebounded to 11% up from 5% growth; this is a critical metric for the company as management must prove they are reaping the rewards of large capex spending. However, it’s also clear the large capex spend is impacting the bottom line as the company beat by 22% for growth of 38.4%. When adding back the $0.52 tailwind from increasing the useful life of servers, EPS beat by 11.2%. 

There is evidence that both the top line and bottom line can continue to improve as Meta quietly asserts its AI strength come 2026. This quarter, average revenue per person (ARPP) is showing initial signs of bottoming with an important uptick YoY that was absent last quarter when ARPP declined YoY. Secondly, Meta has indicated their internal AI operations will result in lower headcount come 2026 as AI reaches the capabilities of a mid-level software engineer. Combined with a potential inflection point in their ad business, that indicates strong double digit EPS growth will continue in 2026. 

Accelerating Growth in Core Business – Largest Beat in Nearly 4 years

Revenue for the quarter came in at $47.5 billion, beating consensus by $3 billion. While a 5% top-line beat may seem modest at first glance, this was Meta’s largest revenue surprise in 15 quarters. When removing one-time adjustments, EPS beat by 11%, or nearly double the beat seen on the top line. 

Chart showing Meta revenue growth accelerating from 16% in Q1 FY25 to 21.6% in Q2 FY25, with projected growth above 20% in Q3 FY25.

Meta’s year-over-year revenue growth shows acceleration from 16% in Q1 FY25 to 21.6% in Q2 FY25, with projections for continued growth above 20% into Q3 FY25.

This quarter’s performance stands in sharp contrast to Q1, when revenue grew a modest 16% YoY, with just a $1 billion beat, and ad impressions were up a muted 5%. In other words, Q2 wasn’t just another incremental improvement, it was a potential inflection point that reset expectations for the back half of the year. It may have also signaled that Meta’s core advertising engine could renew its upward trajectory from the impact of its AI investments. 

Looking ahead, Q3 guidance calls for revenue between $47.5 billion and $50.5 billion, implying another quarter of ~21% YoY growth at the midpoint. If the Company can continue to deliver back-to-back quarters of >20% growth, this should put the narrative of slowing growth in the rearview.  

Sequential Improvement in Margins, up 5 points YoY 

Revenue grew 16% YoY, the slowest rate since Q2 2023 as Meta lapped tough comps. Ad impressions growth decelerated to 5%, with all regions slowing sharply. 

Chart showing Meta’s operating margin expanding from 38% in Q2 FY24 to 43% in Q2 FY25, reflecting stronger profitability trends.

Meta’s operating margin expanded five points year-over-year, rising from 38% in Q2 FY24 to 43% in Q2 FY25, highlighting stronger profitability trends.

Ad impressions rebounded to 11% YoY, more than doubling sequentially. APAC led the way with +16% impressions growth, while US & Canada improved significantly, climbing to +9% from just +4% in Q1. Ad pricing remained firm at 9% YoY, a slight deceleration from Q1 but notable given the acceleration in impressions.  

AI at the Core of Ad Re-Acceleration 

The key driver behind this resurgence is Meta’s aggressive deployment of AI to improve ad efficiency and user engagement. Management highlighted recent upgrades to its ad recommendation system, which now leverages more signals and longer context windows to drive higher performance.  

These improvements had tangible effects in Q2. Ad conversions increased ~5% on Instagram and ~3% on Facebook, reflecting the system’s ability to better match advertisers with the right audiences. Time spent also improved meaningfully, rising 5% on Facebook and 6% on Instagram, which expands available inventory.  

Graphic illustrating Meta’s Q2 2025 ad impressions rising significantly quarter over quarter, highlighting strong advertising performance.

Meta Q2 2025 Ad Impressions Surge: Significant Quarter-over-Quarter Growth Highlights Strong Performance

This dynamic showed up clearly in Meta’s performance metrics. The rebound in ad impressions from +5% in Q1 to +11% in Q2 was the sharpest sequential improvement over a year, driven by strength across regions, particularly in APAC. Despite this surge in volume, pricing held firm, increasing 9% YoY, just a modest deceleration from the prior quarter’s 10% growth. This stability indicates advertisers are seeing higher value per impression, thanks in large part to AI-driven performance gains. 

ARPP also benefitted from these trends, climbing nearly $2 year over year to $13.65, just shy of Q4’s record $14.25. This suggests advertisers aren’t just buying more impressions, they’re paying more for better-performing ones. 

At the center of this success is Advantage+, Meta’s flagship AI ad platform. Advantage+ automates campaign targeting, budget allocation, and creative generation, providing advertisers with a powerful easy-to-use tool that integrates generative AI directly into Meta’s ad ecosystem. The results speak for themselves: Advantage+ is now operating at a $20 billion annual run rate, up 70% YoY. 

Advertisers using Advantage+ report up to 22% improvements in returns on ad spend (ROAS), and adoption continues to climb, with more than 4 million advertisers now using at least one generative AI creative tool.  

According to the most recent earnings call, management stated why they are seeing such rapid growth: "studies show that for every dollar spent with our AI-enabled Advantage+ products, advertisers generate on average $4.52 in revenue for their businesses.” 

Why Meta’s AI Story Stands Apart 

Sign up for free to receive the following information: 

  • The one simple reason that Meta could see a stronger 2026 than currently anticipated 
  • Whether Meta’s high capex of $100B run rate is a boon for future growth or a concern for investors to watch out for 
  • Looking beyond revenue, what line item we are watching for weakness in 2026 and what needs to happen to offset this

Overall, Meta theoretically has a quick time to market for AI improvements to make an immediate impact on its existing monetization engine; the ad platform. This difference makes Meta’s capex cycle less speculative in the immediate term. While many peers front-loaded their AI spending over the past 12-18 months without a clear timeline for payback, Meta is scaling infrastructure after achieving product-market fit, with clear visibility into ROI. 

Meta also has an unrivaled distribution advantage. Roughly 700 million monthly users already engage with embedded Meta AI features inside Facebook, Instagram, and WhatsApp. More recently, the company announced they reached 1 billion users across their AI services following the launch of the Meta AI app last April 

With 3.4 billion total daily active people across its Family of Apps, Meta can distribute new AI features at scale. Unlike standalone AI apps that must drive user acquisition, Meta can compress the adoption curve dramatically by deploying AI directly into platforms where billions of people already spend time. 

Building the Infrastructure to Match 

Mark Zuckerberg confirmed that the Company is building Prometheus, which will be the first 1+ GW AI Accelerator cluster, slated to come online in 2026. Capex is ramping rapidly to meet these goals. For FY25, guidance now calls for $66-72 billion, up from roughly $40 billion in 2024. 

Chart showing Meta’s FY25 capital expenditure guidance rising to $69 billion, surprising investors with the scale of its continued spending surge.

Meta FY25 Capex Guidance Hits $69B: Market Surprised by Continued Spending Surge

Management also provided visibility into $100 billion in annual capex by 2026, reflecting Meta’s commitment to building out training and inference capacity at unprecedented scale. Unlike earlier cycles, where spending was tied to long-lived assets or speculative projects like the metaverse, this wave of investment is highly focused on shorter-lived assets such as GPUs, networking gear, and data center hardware that directly power ads, inference and generative AI workloads. 

Funding From Strength, Not Dilution

One of Meta’s biggest advantages is its ability to self-fund this ambitious AI buildout. The company generated $25.6B in operating cash flow during Q2, representing a robust 53.8% margin. Free Cash Flow came in at $8.6B, down from $10.3B in Q1 as capex surged, but still massive relative to nearly any other company in tech. 

Cash balance declined to $47.1B, reflecting the Company’s $15 billion Scale AI acquisition, while debt remained steady at $28.8B. Management emphasized that it plans to finance the bulk of its capex internally, while also exploring co-development partnerships for certain mega-projects. This flexibility distinguishes Meta from smaller AI infrastructure players that rely heavily on capital markets to fund growth. 

Margins Stable Today, but 2026 a Watch Item 

Despite the surge in spending, Meta’s margins held strong in Q2. Gross margin held steady at 82.1%, up 80 basis points year over year. Operating margin improved to 43.1%, rising 170 basis points sequentially and 500 basis points YoY, while net margin came in at 38.6%. 

Management did caution, however, that 2026 expenses will grow faster than revenue as depreciation, energy costs, and data center operating expenses rise with the scale of new infrastructure. If revenue lands near $230 billion in 2026, above current consensus of $215 billion, operating margins could dip toward 34-36%, down from ~39-40% today. Even so, Meta’s high cash generation and potential partnerships for data center development give it multiple levers to manage this growth without compromising financial stability.  

Conclusion: 

Meta’s Q2 could quietly become the company’s inflection point after a weak Q1. Impressions rebounded, pricing held firm, and ARPP was up $2 YoY, driving the Company’s largest top-line surprise in nearly four years. Guidance for Q3 sets up another quarter of >20% growth, cementing the case that Meta’s core business is back in growth mode.  

Notably, Meta’s AI spend is already paying off. Its investments are directly linked to revenue growth through its ad platform, creating a virtuous cycle that reinforces itself with every incremental improvement in AI performance. While 2026 expenses will pressure margins in the near term, Meta is entering this next phase from a position of extraordinary strength: a core business growing north of 20%, is cash flow positive even after $100 billion in capex, and offers an AI roadmap that offers quicker monetization.

Following 22 trade alerts between March-April this year, we have a handful of positions up over 80%, one position up over 100% and two entries up over 300% in 2025.  Our cumulative returns of 210% over a five-year period would place us as #2 if we were a hedge fund and #5 if we were an ETF. Learn more here

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in META at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

IREN: GPU Fleet Doubled to 23K, AI Cloud ARR Guide Raised to $500M 

Posted on by io-fund

The company IREN was formally known as Iris Energy, yet changed its name to reflect the stock ticker last year. There are a few key things that separates IREN from other Bitcoin miners. The first is that IREN is still a Bitcoin miner whereas others have retrofitted their Bitcoin operations for the data center entirely or plan to very soon as the operations were not profitable. In contrast, IREN is able to turn a profit from Bitcoin mining and plans to use that cash to help fund its AI data center expansions. IREN is able to turn the profit due to selling the Bitcoin it mines yet also due to operational efficiencies that result in lower power costs. The last part is interesting, to say the least, given the company not only has 3GW of power yet has demonstrated they can offer a PUE of 1.1 to 1.2 on air cooled sites, and around 1.4 PUE for the incoming liquid cooled sites like Sweetwater.  

Secondly, IREN can offer a hybrid mix of both colocation and AI cloud services, which is essentially bare metal servers without a hypervisor or virtual layer. This is attractive to hyperscalers as virtualization can lead to performance loss. For the AI cloud, IREN functions similar to CoreWeave as the GPUs are leased “as a service.” For colocation, IREN provides the facility, power and cooling for customers to deploy and manage their own hardware. Although IREN is open to a mix of both colocation and cloud deals as it diversifies how they fund operations and raise capital, there’s no doubt that cloud deals are what could separate this stock from its Bitcoin mining peers. 

IREN Doubles GPU Fleet to 23K with $674M Nvidia and AMD GPU Purchase 

At the start of the week, IREN announced that it had purchased an additional ~12.4K GPUs for $674 million, more than doubling its GPU fleet to ~23K. This order included approximately 7.1K Nvidia B300s, 4.2K Nvidia B200s and 1.1K AMD MI350X GPUs, scheduled to be delivered over the coming months. Notably, this is the first time that IREN has purchased AMD’s GPUs, as it is currently focused primarily on buying Blackwell GPUs, which account for >85% of its fleet. 

This builds upon rapid expansion of its GPU fleet in late August, where IREN purchased ~4.2K B200s for $193 million, followed by the $168 million purchase of 1.2K air-cooled B300s and 1.2K liquid-cooled GB300s just three days later. 

Currently, IREN’s fleet includes: 

• 1.9k Nvidia H100s & H200s 

• 19.1k Nvidia B200s & B300s 

• 1.2k Nvidia GB300s  

• 1.1k AMD MI350Xs 

In total, IREN’s fleet cost ~$1.24 billion, including ancillary equipment such as InfiniBand, servers, cabling, and licensing costs. This puts the company’s all-in cost per GPU at just above $53,000, though notably the Blackwell Ultra generation carries a significant premium to the B200s. IREN’s recent 2.4K purchase of B300 and GB300s cost ~$70,000 per GPU ($60,000 for the B300 and $80,000 for the GB300), while its 4.2K B200 purchase cost just under $46,000 per GPU.  

Financing future GPU purchases will be paramount for IREN’s expansion plans, as the company has turned to high-single digit interest rate, 24 to 36 month lease financing for recent purchases. Depending on the how its latest $674 million purchase is financed, these leases already could add up to $100 million in quarterly expenses tied to these GPUs.  

With the latest purchase, IREN has likely maxed out its capacity at Prince George, which it has previously said can host >20K GPUs. Across its British Columbia data centers, IREN can support >60K GPUs, which would likely require an additional $3B+ in funding depending on GPU mix. Comments from management hint at expanding its GB300 fleet, which is currently the most expensive GPU with an average all-in cost of ~$80,000:  

“Construction is well underway on a new 10-megawatt liquid cooled data center at Prince George, designed to support more than 4,500 Blackwell GB300 GPUs. … Beyond Prince George, Mackenzie and Canal Flats. Our data center campuses in each of these locations create an even larger opportunity with powered shells, existing and designed to the same architecture as Prince George, these sites offer a straightforward and replicable pathway to more than 60,000 GB300s. Horizon 1 and our broader portfolio of data center sites in Texas opens up a further path to continued AI cloud growth.” 

At Horizon 1, IREN says that it can host ~19K GB300 GPUs at 50 MW IT load, which, based on prices calculated above, would cost the company upwards of $1.5 billion. Funding this and fully outfitting its British Columbia sites for 100% AI cloud capacity would likely cost $5 billion or more, or ~10x IREN’s most-recently reported cash holdings. This also does not account for its 2GW Sweetwater campus, which IREN says can support >600K GB300s. 

Analysts are expecting IREN to quickly scale its fleet through 2026, with Roth Capital projecting IREN to reach a ~112K GPU fleet by year-end 2026. This ~90K increase in GPU fleet could require IREN to take on more than $6 billion in debt, per Roth’s calculations. This would represent nearly 60% its current valuation and likely cost ~$600 million quarterly, which would not be covered by revenue nor cash flows.  

New AI Cloud ARR Guidance of >$500M 

In accordance with its doubled GPU fleet, IREN unveiled a new AI Cloud annualized revenue (ARR) guidance, now targeting >$500 million in ARR by Q1 2026, more than double its guidance from August for $200 million to $250 million by year-end 2025. The prior $200-$250 million guide was based on a fleet size of 10.9K GPUs, which management expects to be “progressively commissioned over the coming months.”  

This is a rather ambitious target, as this would represent nearly 20x growth from June’s $26 million annualized run rate, or quarterly revenue of just $7 million. July and August have also shown minimal growth in AI cloud ARR, at ~$28 million and ~$29 million, respectively, compared to $26 million in both May and June.  

Management says that they are observing demand for multi-thousand Blackwell GPU clusters, which should help revenue ramp quickly, yet the lack of a ramp suggests that clusters of that size have not yet been delivered. Additionally, the first B200 tranche “was immediately contracted on a multiyear basis” upon commissioning, highlighting that demand for the new generation remains high with customers ready to contract chips immediately once they are available. 

On a quarterly view, IREN is essentially targeting a strong ramp in the December quarter extending into the March quarter, considering July and August revenue mean Q3 is approximately flat QoQ assuming similar run rate in September.    

IREN also disclosed that the >$500 million ARR target is “an illustrative run-rate measure of potential revenue based on a ~23k GPU deployment and internal company assumptions regarding utilization and GPU pricing.” IREN warns that this figure “is not fully contracted, there can be no assurance that it will be achieved, and actual revenue may differ materially.” The ARR target also assumes on-time delivery and commissioning of GPUs, and any delays could quickly and easily derail this target, considering the pace of growth needed over the next six months. 

IREN says it expects delivery of its most recent 12.4K order from Sept 22 to be also delivered over the coming months, and much of its target likely hinges on this tranche as it accounts for more than half of its total fleet. Any delays in delivery towards the start of 2026 could threaten this ARR guide. Management hinted that they are “pre-contracting ahead of delivery”, providing additional visibility into future growth, but that does not span across its entire order book, limiting a high degree of confidence in the near-term.  

IREN Offers Cheaper Energy than Industry Average 

If IREN can maintain lower power costs compared to other Miners, then that could potentially provide a leg up in terms of signed deals. For example, IREN sees a profit of about $50,000 per Bitcoin mined at current prices for a hardware gross margin of about 70%. The company has stated their Childress facility sees energy rates of $0.033 kWh. In the most recent earnings call, the company stated they saw $0.035 kWh in Q4. 

According to an analysis by Nlyte.com the industry average PUE is 1.58 with companies like Google reaching 1.10. The analysis also points toward rates of $0.0616/kWh in regions like Iowa up to $0.2496/kWh in regions like Rhode Island. This helps to illustrate how IREN stands apart as the company is seeing energy rates 50% lower than cheap regions. 

Regarding PUE, this fluctuates depending on air cooled versus liquid cooled, and may be impacted as power demands rise with new generations of GPUs. However, as of now, the company sees PUEs as low as 1.1 for air cooled and anticipates PUEs of 1.4 for its Sweetwater facility. All of the above is lower than the industry average. 

“So as you mentioned, across the BC sites, we're operating at a PUE of 1.1. That's on an air cooled basis. Once we install the liquid-cooled facilities there, we expect that to be operating on an average slightly higher than that, but still well under 1.2 PUE across the year. At Childress, the Horizon 1 liquid cooled installation, the number that you mentioned is much closer to a peak PUE number, although we actually expect it to be less than 1.4 and the average PUE over the year to be around 1.2.” 

IREN Mixes Colocation with Cloud for a Hybrid Approach 

There are pros and cons to colocation versus cloud deals for IREN. Quite a bit of time was spent going through how the company plans to approach its hybrid strategy, given it’s rare to see a Bitcoin miner offer GPUs-as-a-service. It makes sense to try for Cloud deals as they come with a 97% margin, yet analysts were detailed in their questions how successful a Miner will be in pivoting to operate as a neocloud is an unknown and is the only attempt, thus far (the obvious question being, if there’s demand for this, why wouldn’t all Miners pursue this route). 

As pointed out in the earnings call, the following are the major differences between the two from a funding standpoint: 

  • Colocation offers longer-dated contracts. For IREN, the contracts would be between 5-20 years yet takes around 7 years to see the capital absorbed. It leads to lower cash flows. The benefits to colocation is the infrastructure costs are provided for, which are 2-3X higher than the GPU costs. 
  • Cloud deals are shorter contracts with strong margins, yet not many companies are in a position to offer cloud deals. Most Bitcoin miners operate deep in the red whereas IREN has profitable mining operations and controls the infrastructure “end-to-end.” There was mention on the call of a 3-year payback on these deals: “As of today, we find a 3-year payback on data center and GPU infrastructure pretty compelling, particularly when Anthony is lining up 100% GPU financing at single-digit interest rates.” 

The takeaway from the earnings call is that IREN would, of course, prefer to do AI cloud deals yet it remains to be seen if this will be attractive to key customers. As of now, most hyperscalers and neocloud customers would prefer the colocation option.  

Should IREN be able to find a strong market for its AI cloud services, then the stock could do quite well. However, this is the first attempt (and perhaps only attempt) across the Miners to double up as a neocloud. Typically, pivots are difficult to pull off and colocation may be the more popular choice on the customer side (i.e., we can flip this and say … why would a hyperscaler or neocloud pay IREN a higher amount in three years on the cloud side if it can spread out a similar amount over seven years with a contract term of up to 20 years? What's the benefit to the customer to go cloud?) 

It’s my contention that CoreWeave is in a league of its own, offering early access to hundreds of thousands of Nvidia’s GPUs. The higher utilization rates that CRWV offers is an advantage whereas Bitcoin miners have not factored in FLOPs for training models, which can have a much larger impact on output than a simple metric like lower energy rates. As we covered in the past, CoreWeave’s specialty is in optimizing memory bandwidth, improving communication between GPUs, clearing data input bottlenecks, and other ways in which to fix batch size, enable faster data loading, and/or better ways to balance the compute. In other words, what CoreWeave does is not easily replicated. 

Therefore offering an AI cloud is one thing, yet it’s another matter entirely to offer enough software optimizations to justify recurring revenue that will result in 2X higher costs than colocation (if we assume cloud is a 3-year return on capital versus 7-years).  

Fluidstack and Poolside are the primary customers for IREN which indicates they simply want to move quickly and are willing to overpay. Should cloud contracts continue to accumulate, it would be important to understand if there is an exit clause. 

As a reminder, Fluidstack is the neocloud that Google put up a 50% guarantee for on the lease with Terawulf. This caused WULF’s stock to surge. Depending on how Fluidstack funds a specific lease, it can be viewed as an attractive customer. Here is what management stated on the call regarding their AI cloud offering: 

“Because we own and operate the full end-to-end stack, we are able to deliver superior customer service, tighter control over efficiency, uptime and service quality translating directly into a better customer experience for our customers. We are leading our service with a bare metal service because it gives sophisticated developers, cloud providers and hyperscalers what they want most, direct access to compute and the flexibility to bring their own orchestration.” 

Financials 

Top-Line Performance: Growth on Two Fronts 

IREN reported FY25 revenue of $501.0 million, up 168% year-over-year from $187.2 million, underscoring one of the fastest growth rates in the sector. This surge is driven by two different engines: the traditional Bitcoin mining business, which contributed $484.6M, and the emerging AI Cloud Segment, which posted $16.4 million, nearly five times its FY24 contribution. This represents IREN’s first full fiscal year with AI cloud as a meaningful contributor, validating management’s strategy of diversifying beyond the volatile crypto cycle into high-performance computing and AI hosting. 

  • FY25: $501.0M vs $187.2M in FY24 (+168% YoY). 
  • Q4’25: $187.3M vs $56.8M in Q4 FY24 (+229% YoY) and $113.6M in Q3 FY25 (+65% QoQ) 

The fourth quarter alone delivered $187.3 million, up 229% from the $56.8 million earned in the year-ago period and up 65% sequentially from Q3’s $113.6 million. This kind of sequential growth is rarely seen outside of hypergrowth SaaS, let alone in a miner. The bulk of Q4 revenue came from Bitcoin at $180.3 million, with AI Cloud adding $7.0 million.

As seen in the charts above and below, analysts expect further continuation of this trend in significant top-line growth. Quarterly estimates remain strong through FY26 with all quarters exhibiting greater than 80% YoY growth. Annually, this equates to an FY26 growth rate of 109%, with an additional 49.3% growth in FY27 before pausing in FY28.  

Segment Breakdown: Still Bitcoin-Heavy, but AI Cloud Gains Relevance 

To be clear, Bitcoin mining still dominates the financials, accounting for 97% of FY25 revenue. Revenue rose 163% year-over-year as hash rate climbed to 50 EH/S, even as net electricity costs per coin mined rose to $25,642 post-halving (vs. $18,127 in the prior year). Favorable spot prices offset that inflation, keeping the economics attractive. 

The eye-catcher is AI Cloud. The segment remains small, at just 3% of total revenue, yet it is growing rapidly. FY25’s $16.4 million compares with $3.1 million last year, and Management is already telegraphing a path to $200 – $250 million annualized run-rate assuming 10,900 GPU’s are deployed by December 2025. Q4’s $7.0 million contribution shows that ramping is starting to appear in the reported numbers. This segment’s strategic importance outweighs its current financial contribution, as it diversifies revenue away from Bitcoin and positions IREN as a credible infrastructure partner for AI workloads. 

Top Line Growth Drives Margin Expansion  

Gross profit (excl. depreciation) for FY25 was $342.0 million, translating to a 68.3% margin, up nearly 15 percentage points from FY24’s 53.5%. Combining 167% top line growth with this 15 pp of margin expansion leads to Gross Profit increase of 241.7% YoY. This is the textbook definition of operational leverage. In Q4, the margin reached 71.8%, further reflecting benefits from scaling mining operations and disciplined power cost management. 

Operating Income turned positive in FY25 at $17.3 million or a 3.5% margin, a sharp swing from FY24’s ($25.2M) loss and (14.6%) margin. Q4 Operating Income was $20.6 million (11.0% margin), another milestone, as operating profit is now sustainable rather than a one-off. 

Adjusted EBITDA tells the operational efficiency story most clearly. For FY25, IREN delivered $269.7 million, a 54% margin, up nearly 5x from FY24’s $54.4m. In Q4 alone, adjusted EBITDA reached $123.0 million, or 65.7% of revenue. 

GAAP Net Income was strong but requires an asterisk. FY25 bottom line came in at $86.9 million, swinging from ($28.9) million loss last year with diluted EPS of $0.39. Q4’s $176.9 million profit, an absurd 94% margin) was artificially boosted by $775 million in unrealized gains on financial instruments and a $9.1 million gain on liability extinguishment. Strip those out and the true earnings power looks more like the $20M operating income figure. 

FY25 Key Metrics: 

  • Gross Profit (ex-D&A): $342.0M, margin of 68.3%, up from 53.5% in FY25 
  • Operating Income: $17.3M, margin of 3.5% vs. ($27.2M) or -14.6% in FY24. 
  • Ad. EBITDA: $269.7M, margin of 54%, up 5x YoY from $54.4M, a 29% margin 
  • Net Income: $86.9M vs ($28.9M) in FY24 

Q4’25 Key Metrics: 

  • Gross Profit (ex D&A): $134.4M, margin of 71.8%, up from 71.0% in Q3’25. 
  • Operating Income: $20.6M, margin of 11.0% vs. $29.1M or 20.1% in Q3’25. 
  • Adj. EBITDA: $123.0m, margin of 65.7% margin 
  • Net Income: $176.9M, a 94% margin  

Overall, it’s important to remember that IREN’s profitability is extremely sensitive to BItcoin’s current trading price. The profitabilty discussed above is due to operational efficiencies on the mining operations combined with Bitcoin trading above $100K and not from scaling the AI data center infrastructure  or cloud services (yet).  

EPS remains volatile due to GAAP mark-to-market gains 

EPS flipped positive in FY25, with diluted EPS of $0.39 versus a ($0.29) loss in FY24, marking the Company’s first full-year profit on a per-share basis. However, Management cautions that GAAP EPS is heavily influenced by fair-value accounting marks. Looking forward, analysts see GAAP profitability through FY26 as the Company expands its AI Cloud offerings efficiently at scale. 

Operating Cash Flow Turns Positive amidst GPU-fleet buildout 

Operating cash flow was a bright spot in FY25, at $245.9 million, up from just $52.2 million the prior year. That represents 48.9% of revenue flowing through to operating cash, a healthy conversion rate given the non-cash noise in reported earnings. 

Despite healthy operating cash flow trends, free cash flow was negative in FY25 as capex spend was immense. IREN spent $1.38 billion on PP&E, consisting mainly of GPUs and data center expansion, more than doubling FY24’s spend of $692 million. The result was free cash flow of ($1.13B), a -226% FCF margin. In other words, every dollar of operating cash flow generated was more than outspent on buildout. 

Financing flows more than filled this gap with $1.30 billion raised via converts, equity, and leases. Net-net, IREN ended FY25 with $160 million more cash than it started with, despite billion-dollar capex outlays. After year-end, the Company raised another $253.5 million via ATM equity sales and finalized a lease program that funds GPUs entirely, with fixed monthly payments of ~$2.8M and a buyout option at 18% of cost after 36 months. This strategy shifts capital intensity away from cash up front, preserving liquidity while enabling AI Cloud scaling.  

Key Metrics: 

  • Cash and cash equivalents: $564.6M, up from $404.6M in FY24. 
  • PP&E: $1.38B, mainly GPU purchases and data center buildouts. 
  • Debt: $962.8M, convertible notes outstanding (non-current) 
  • Equity: $1.82B, reflects retained earnings and capital raises.  
  • Shares outstanding: 258.1M. 

Liquidity / Solvency Comparison vs. Peers 

At fiscal year-end, IREN held $564.6 million in cash against $962.8 million in convertible debt, equating to cash-to-debt ratio of .59x. Put differently, the Company had roughly 60 cents of cash for every dollar of debt outstanding.  

For context, this is a stronger liquidity position than many mining peers, who often carry higher net leverage and rely more heavily on dilutive equity raises. The ratio underscores that IREN is not currently overextended. Its sizeable cash cushion provides flexibility to meet near-term obligations, fund working capital, and invest in ongoing GPU deployments. 

However, the ratio also illustrates the reality of IREN’s capital intensity. With a $1.38 billion in FY25 Capex and another multi-billion-dollar investment cycle ahead, cash on hand can quickly become consumed unless offset by financing inflows. Management has already leaned on convertible notes, ATM equity, and hardware lease financing to balance the scales. 

The sustainability of IREN’s expansion will depend on a handful of factors worth tracking. To fund operations, IREN must keep it’s cash-to-debt ratio stable through disciplined liquidity management. Further, IREN must avoid excessive reliance on equity dilution, which could weaken per-share economics even if absolute liquidity remains healthy. Beyond liquidity concerns, the Company needs to execute a successful ramp AI Cloud revenues to justify the spend. If IREN can scale AI Cloud revenues and maintain current unit economics (~93+% gross margin), these cash flows would provide a healthy recurring buffer against heavy capex. 

In short, IREN’s 0.59x cash-to-debt ratio highlights both balance sheet strength and exposure. As discussed above, the Company has meaningful liquidity today, but the scale of its expansion means this ratio will be a key metric to monitor as it pursues GPU deployments and new data center builds through FY26. 

Conclusion: 

If IREN can prove they can pull off charging recurring revenue for its AI cloud services, then the stock is one to watch. We are at a critical juncture for cloud deals as analysts are expecting IREN’s revenue to decelerate in H2 2026. Therefore, any cloud deals that beef up current analyst expectations can help to strengthen this narrative.  

Right now, we prefer to stay as close to the hyperscaler deals as possible when evaluating Bitcoin Miners. The reason for this is that it solves the pain point of having a company with deep pockets back-stop the leases, which in turn, improves creditworthiness and credit terms. As many of you are aware, our ethos is to participate in the upside while protecting to the downside. We want the best of both worlds, and in a highly speculative momentum play like Bitcoin Miners pivoting to AI data center infrastructure, the primary goal is to reduce risk.  

We are watching IREN closely and would buy on a clear breakout only. If we were to buy, we’d closely adhere to all stops.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Every Thursday at 4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock and crypto entries and exits. Beth Kindig offers weekly deep dives including lesser-known cryptocurrencies and AI stocks, plus the team offers trade alerts. The I/O Fund team is one of the only audited portfolios available to individual investors. If you’d like to subscribe to the Advanced Market Signals plan, email us at premium@io-fund.compremium@io-fund.com.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

Bitcoin Miner Riot: Anticipation Builds for a Data Center Deal

Posted on by io-fund

The saying that “good things come to those who wait,” is what Riot Platforms is banking on. The company is taking its time in preparing its two major sites for AI data center customers. Meanwhile, Bitcoin mining peers seem to be rushing toward deals, and their stocks have been popping off over the past few weeks as a result. 

When asked in the earnings call what the hold up is, Riot explained they are looking to maximize the deal terms as they have one of the most attractive sites in the country (an analyst’s words regarding the Corsicana site, not theirs). Corsicana is especially attractive as it’s located near the Tier 1 market of Dallas and has ample power and water for liquid cooling. The Rockdale site is the second priority for securing a lease and will be used for Bitcoin mining in the interim. 

Overall, Riot has stated they have “over 1.7 gigawatts of power immediately available near major markets.” Riot’s balance sheet also stands out with $2.4 billion in liquidity due to having $330 million in cash and 19,000 Bitcoin.  

However, what investors are up against is navigating near-term volatility while a deal announcement could be another quarter (or two) away. This is because Riot is going through a master site design this quarter (ending now in Q3), which helps to start the discussions for a deal. It was stated the site design is what would kick off discussions with customers: “And if you're talking about engaging with serious counterparties, this is the type of information that they [want] you to come to the table with in order to advance discussions substantially.” 

While Riot is currently in the final stages of designing the data center and working to secure a lease, it may be Q4 or later before a firm deal is announced. On the revenue side, if a deal is done in Q4, it will likely be more geared toward mid-year 2026 or the second half based on current timing for substation development at Corsicana. 

In the opening remarks, Riot explained they are in no rush: “To be clear, we are not pursuing a so-called pivot into AI HPC initiatives with a view of doing a "deal.” Rather, we have added a new data center development capability, which we will apply to as much of our power portfolio as possible and which will transform our company in the years to come.”

Riot’s 8-Step Process for AI Data Center Pivot 

Riot has outlined its eight step process behind its AI data center pivot, with the company now focusing fully on the sixth, seventh and eight steps: 

  1. Engage consultants: Riot tapped data center industry consultant Altman Solon in January to explore feasibility of AI/HPC pivot at Corsicana 
  2. Add experience to board: Riot added three new board members in February with relevant data center experience 
  3. Engage financial advisors: Riot tapped Evercore as a financial advisors and Northland as a co-placement agent to lead discussions with possible AI customers 
  4. Continue infrastructure development: Pertains to the development of the 600MW substation at Corsicana 
  5. Build internal expertise: Riot stated it is making key hires to build out its internal data center expertise 
  6. Complete basis of design: Riot is leveraging new hires and external consultants to best design data centers that fit its available infrastructure 
  7. Secure lease with tenant: Riot says it is “actively progressing” towards securing a lease with a “high-quality” tenant and collaborating on final design iterations 
  8. Build future pipeline: Pertains to further expanding its AI data center footprint in the future with powered land acquisitions and development 

While Riot is currently in the final stages of designing the data center and working to secure a lease, it may be Q4 before a firm deal is announced. On the revenue side, it will likely be more geared toward mid-year 2026 or the second half based on current timing for substation development at Corsicana.  

Riot’s Corsicana Data Center Can Offer 600MW for AI Data Centers 

Riot is working to expand its Corsicana data center, purchasing 355 acres of land in May and an additional 238 acres in July, taking its total acreage to 858 and allowing it to “accommodate various data center designs and development plans.”  

Corsicana has 1 GW of power capacity, with 400 MW currently operational and earmarked for Bitcoin mining, with the additional 600MW of power capacity for AI/HPC hosting is expected to be ready in early 2026. The facility also has two fiber providers on-site, offering ample connectivity and low latency to major hubs Dallas and Austin. Riot also disclosed that its all-in power costs at Corsicana (for BTC mining) were ~$0.035/kWh, matching IREN with industry-low power prices. 

The entire 1GW is expected to be available sometime over the next 4-5 quarters: “So first, for some of the critical infrastructure that's needed to build this capacity, we have already secured. I'm referring to the 600-megawatt substation that's being built that's expanding the site to 1 gigawatt. We have already procured that equipment. That equipment is already arriving, and that is going to take our Corsicana site to 1 gigawatt in 2026.” 

What’s unique about Riot is they already have approval for power to both sites, whereas newer sites are subject to legislation like Texas Senate Bill 6, which requires Bitcoin miners to cover some of the costs of expanding the grid. There are also curtailments to where miners/HPC operators have to reduce usage when the grid is maxed out and are required to have backup power. The company stated they will be able to power the pivot to AI without any further steps: “This power isn't pending certain steps happening. This power is coming in the next 6 months and scaling up from there.” 

In Q4 FY24, Riot estimated it would spend ~$65 million in capex on developing the 600MW substation and acquiring other long-lead items required to energize this phase. Through the first half of FY25, Riot spent $51.3 million in capex related to the substation, and projected an additional $17.9 million in spending through Q4, or ~$69.2 million in total capex, a 6.5% overrun. Riot’s Q4 acquisition of E4A likely plays a critical role as the company is engaged in substation equipment, transformer assembly and power plant servicing. Additionally, the mid-year land purchases, water lines and other site costs are expected to total approximately $77 million this year.  

The company is evaluating more expansion opportunities for AI/HPC, and assessing feasibility for developing further power capacity, but has not made any strides outside of Corsicana yet. Combined with its 700MW Rockdale facility, Riot can offer 1.7GW of power in close proximity to Tier 1 Dallas and also Austin, though Corsicana’s first 400MW phase and Rockdale are presently fully outfitted for Bitcoin mining and will be retrofitted in phases when management believes the timing is right.  

How Much Could Corsicana be Worth? 

Riot is well aware of the opportunities in pivoting to AI data center hosting, more so now considering the flurry of AI deals that have been signed with miners recently. As we pointed out in our Advanced analysis in January, Marathon Digital and Riot Platforms: Leveraged Bitcoin Bets, Riot’s management had said in Q3 2024 that there is “notable sense of urgency for power access in 2025 with AI HPC companies willing to pay a premium for timely access at attractive sites.” Riot also stated at the time that it was in preliminary discussions with AI HPC firms over some level of capacity, and it would see if there are deals to monetize capacity at a better rate than mining.  

As of now, Riot has yet to sign an AI hosting deal, though analysts from Needham expect the company to be in advanced discussions by Q4 and sign a lease agreement as early as Q1 2026. This would be a few quarters behind some of the earlier movers in the industry, but this hinges on its Corsicana facility, with its AI side still under development. Notably, Riot likely will not transition its entire footprint over to AI hosting, stating in its 10-Q that is in developing a “scalable data center platform designed to allocate a portion of our power capacity toward supporting AI/HPC workloads.”  

Despite the lack of a deal, allocating 600 MW of power capacity to AI (assuming ~440 MW of critical IT load as Needham estimates) allows a reasonable projection for what Corsicana could be worth under similar terms as other miners.  

Assuming a deal structured around $1.8 million in average annual revenue per MW of critical IT load, Corsicana’s AI data center side could fetch a deal worth ~$8 billion over ten years, or ~$792 million in average annual revenue. Other miners have outlined targeted net operating income or adjusted EBITDA margins on these deals of >85%, suggesting that Corsicana could deliver adjusted EBITDA or net operating income of ~$673 million annually on average. As a reminder, these are average annual figures, meaning that revenue in the initial stages will begin at a much smaller size as capacity ramps and end much larger than these stated figures.  

For Riot’s combined Corsicana and Rockdale facilities offering 1.7GW of available power, and assuming both are fully converted to AI with a 1.3 PUE for ~1.3GW of critical IT load, the two could be worth more than $23 billion for a 10-year deal structured at similar terms, or average annual revenue of $2.34 billion.   

Bitcoin Mining Expansion Remains on Track 

On the flip side, Riot is still committed to expanding its Bitcoin mining fleet and hash rate, even though the Corsicana pivot reduced its 2025 deployed hash rate target by 8 EH/s.  

In Q2, Riot updated its year-end hash rate target to 40 EH/s, up from 38.4 EH/s, likely driven by its recent addition of 125 MW of capacity for Rockdale by acquiring additional assets from Rhodium and closing pending litigations. This represents ~26% YoY growth in Riot’s deployed hash rate, aligning with Riot’s target to maintain ~4% of the global network, currently at ~984 EH/s. By Q1 2026, Riot is aiming to bring its hash rate up to 45 EH/s, what would represent its largest sequential expansion in hash rate since Q3 2024 as it works to double its power in Kentucky to 127 MW.  

Rockdale will remain a Bitcoin mining operation until Corsicana finds a customer. Riot positioned it as a strength to keep Rockdale as Bitcoin mining operations for now, stating: “So you can think of our strategy as using Bitcoin mining at sites like Rockdale to monetize that power to ensure that no power stranded and wasted, turning that into meaningful cash flows for the company and then ultimately looking to transition that capacity to data center leases when the time is right.” 

Financials 

Bitcoin Mining Remains RIOT’s Core Revenue Driver 

Riot’s Q2’25 results reaffirm its position as one of the largest self-mining operators in the world. The Company reported total revenue of $153 million, more than doubling from $70 million in the year ago quarter. Most of this revenue came from Bitcoin Mining segment, which contributed $140.9 million, up 152% year-over-year. The drivers here are straightforward: Riot produced significantly more Bitcoin, at higher prices. On top of that, Bitcoin itself averaged roughly $98.8k during the quarter compared to $66k last year. The combination of higher volume and higher pricing powered a sharp rise in top-line contribution. 

Average operating hash rate increased from 11.3 EH/s last quarter to 31.7 EH/s this quarter, while deployed hash rate reached 35.4 EH/s by quarter-end. Think of hash rate as Riot’s “factory horsepower”. The more horsepower it has online, the larger its share of Bitcoin production – but it also burns more energy which makes power contracts and efficiency just as critical. 

Engineering revenue, though a fraction of mining, also provided a small boost. The segment generated $10.6 million in revenues, up modestly from $9.6 million last year. This now includes contributions from December 2024 acquisition of E4A Solutions, which expanded Riot’s capabilities in custom electrical and fabrication work. Engineering also carries $22.7 million of contract liabilities, suggesting a healthy pipeline of projects yet to be recognized. 

“Other’” Revenue totaled just $1.5 million, largely residual income from legacy hosting contracts that have since been terminated. Importantly, Riot no longer reports hosting a standalone segment, underscoring the Company’s shift to pure self-mining plus a smaller engineering team. 

Looking forward, consensus expects continued topline growth through FY25 and into FY26, albeit at a decelerating pace. Seen in the chart above, analysts expect 99.6% and 30.47% YoY growth in Q3 and Q4 to round out FY25. Estimates moderate looking further into FY26 with growth drifting off to mid-teens (17.6%) by Q4’26. 

For investors looking at AI upside, there are no contracts signed and no AI revenue recognized in Q2. The AI narrative remains aspirational, not actual. Compared to peers like Applied Digital and Terawulf, both having already signed multi-year, multi-billion dollar hosting agreements with hyperscalers, Riot is behind on the diversification curve. At this point, AI should be viewed as upside optionality embedded in Riot’s power infrastructure, rather than a near-term driver of financial results.  

Key Metrics: 

  • Total Revenue: $153.0M (+119% YoY) 
  • Bitcoin Mining: $140.90 (+152% YoY) 
  • Engineering: $10.6M (+10% YoY) 
  • Other: $1.5M (hosting residual) 
  • Avg. BTC price: $98.8k (vs ~$66.6k YoY) 

Gross Margin Expansion Driven by Healthy Mining Economics, Operating and Net Margins Inflated by Crypto Gains 

Mining economics remained healthy. Cost of revenue in the Bitcoin Mining segment of $78.2 million implying mining gross profit of $62.7M and margin of 44.5%. Power accounted for the majority of costs at $62.2M, 79% of total segment costs. Other items included $5.7M in taxes and fees, $4.7M in compensation, $4.1M in miscellaneous operating expenses, and $1.5M in insurance.  

All-in power costs rose from 2.1 cents /kWh last year to 3.5 this quarter as Riot added Corsicana and Kentucky to its portfolio. Even with this increase, Riot remains one of the lowest-cost operators in the sector. Rockdale’s 345 MW PPA remains a key differentiator, locking in fixed-price power through 2030 with the ability to resell curtailed load back into ERCOT during peak demand. 

On a GAAP basis, Riot’s margins were eye-popping. Operating income was $216.1 million, translating to an operating margin of 141% compared to negative 167% last year. Net income of $219.5 million equated to a net margin of 143%, compared to negative 121% a year ago. As always, context matters. Riot’s operating and net margins reflect the heavy influence of new Bitcoin accounting rules.  

Prior to 2025, Bitcoin and other digital assets were classified as “indefinite-lived intangible assets” under US GAAP, meaning that Company’s had to carry them at cost on the balance sheet and test annually for impairment. If BTC prices dropped below carrying value, an impairment charge was booked. If BTC prices recovered later, you could not mark the assets back up, leading to asymmetric accounting – lots of impairments, but no mark-ups. 

Effective January 2025, the Financial Accounting Standards Board (FASB) implemented a new rule which requires digital assets like Bitcoin to be carried at fair value, with changes flowing through net income. This means that each quarter, Company ‘s must revalue their BTC to market prices as of balance sheet data, causing non-cash gains / losses that distort operating income and GAAP figures.  

In Q2, this resulted in a $262.8 million “Change in fair value of Bitcoin” gain flowing directly through the income statement. This single item more than explains the swing to profitability. Without it, Riot’s core operating performance looks much thinner. Further, operating profitability was also pressured by Depreciation and amortization ($83.2M), stock-based compensation ($59.7M), and unrealized losses on securities reached ($57.1M). 

GAAP EPS came in at $0.65 basic and $0.58 diluted, versus ($0.32) loss in Q2’24. Looking forward, analysts do not expect annual improvement toward profitability in FY25 or FY26, as EPS estimates are ($0.35) and ($0.48) in FY25 and FY26 respectively. GAAP EPS is not a reliable measure of Riot’s underlying profitability. Investors should instead focus on cash operating margins and Adjusted EBITDA to understand the true economics of the mining business. 

Key Metrics: 

  • Mining gross profit: $62.7M 
  • Mining gross margin: 44.5% 
  • Operating income: $216.1M 
  • Operating margin: 141% (vs – 167 YoY) 
  • Net Income: $219.5M 
  • Net Margin: 143% (vs –121 YoY) 
  • Basic EPS $0.65 
  • Diluted EPS: $0.58 

Cash Flow and Balance Sheet 

Despite reporting net income, Riot’s operating cash flow was sharply negative at ($231.3 million), compared to ($122.1 million) in the prior quarter and ($42.5 million) in the prior year. That equates to an OCF margin of (151%). That disconnect again stems from the way Bitcoin revenue is recognized. Riot books revenue at the fair value of BTC when mined, but no cash is received unless those coins are sold. In Q2, Riot mined about $283.7 million worth of BTC, inflating revenue and net income, but none of that translated into cash inflow.  

Working capital shifts added further drag. Accounts Payable fell by ~$78M, deferred revenue declined ~$34M, and accrued liabilities decreased ~$12M, all of which consumed cash, resulting in negative OCF even as GAAP profits surged. 

On the balance sheet, Riot remains well capitalized but increasingly levered. As of June 30, 2025, the Company held $255.4 million in cash, plus $74.9 million in restricted cash. Bitcoin holdings amounted to $1.71 billion unrestricted and $353.7 million restricted, giving Riot one of the largest corporate Bitcoin treasuries globally. Total assets stood at $4.29 billion, offset by $838.4 million in debt and equity of $3.30 billion.   

While liquidity is ample, leverage has crept higher as Riot continues to finance its large-scale buildouts in Corsicana and Kentucky. ATM equity issuances also remain part of the funding toolkit, with share count rising in the quarter form new issuances and an acquisition-related issuance. The trade-off here is clear: Riot has the liquidity to scale, but at the costs of higher leverage and dilution.  

Capital intensity remains high. Riot has $83.5 million in remaining commitments to MicroBT for miner purchases, expected to be paid in 2025. During the first half of the year, the Company placed $86.4 million in deposits on new equipment and added $93.2 million of property and equipment. These figures highlight the scale of ongoing investment required to maintain Riot’s leadership in hash rate capacity: 

Key metrics: 

  • Cash and cash equivalents: $255.4M 
  • Restricted Cash: $74.9M 
  • Bitcoin Holdings: $1.71B Unrestricted + $353.7M restricted 
  • Debt: $838.4M (current $252.6M, non-current $585.8M) 
  • Equity: $3.30B 
  • Miner purchase commitments: $83.5M (MicroBT) 
  • Deposits on equipment (H1’25): $86.4M 
  • Property & Equipment additions (H1’25): $93.2M 

Litigation Risks 

Litigation is a non-trivial overhang. Riot is facing a lawsuit seeking nearly $500 million from GMO, a Japanese IT and mining company, related to termination colocation agreement. GMO alleges that Riot wrongfully terminated the agreement and is seeking damages. Riot disclosed the case it’s in legal proceeding footnote but stated it cannot reasonably estimate the outcome or probability of loss at this time. 

This claim is significant: ~$500M represents almost 15% of total assets ($4.29B) and is more than double the Company’s quarterly income. While Riot has a strong equity base and deep liquidity, an adverse ruling would be material. Notably, peers like APLD or IREN don’t face comparable litigation exposure, making Riot’s risk profile a bit heavier by comparison. Investors should treat this as a tail risk that could swing sentiment if unfavorable developments occur.  

Bulls vs Bears 

The Bull Case: 

  • Scale and Cost Advantage: Riot operates one of the largest fleets in North America with over 35 EH/s deployed and maintains a competitive power cost of ~$3.5 cents per kWh, keeping it at the low end of the industry cost curve. 
  • Balance Sheet Optionality: With $255 million in cash and $1.7 bitcoin of Bitcoin holdings, Riot has liquidity unmatched by most peers, providing downside cushion and upside leverage to BTC prices. 
  • AI / HPC Upside: While not yet monetized, Riot is positioning its multi-gigawatt power platform for future AI/HPC hosting, offering optionality that could re-rate the stock if contracts materialize. 

The Bear Case: 

  • Accounting Noise: GAAP earnings are dominated by Bitcoin fair-value revaluations, making reported profitability a less useful metric that is volatile and disconnected from underlying cash economics. 
  • Litigation and Leverage: The $496 GMO lawsuit and rising debt levels ($838M) create potential balance sheet and legal overhangs, especially if BTC retraces. 
  • Lagging AI Execution: Unlike peers APLD and WULF, RIOT has yet to sign meaningful AI/HPC deals, leaving it more exposed to Bitcoin cycles with less diversification. 

Conclusion: 

Analysts have called Riot’s Corsicana data center site one of the most attractive in the country, as it’s located near the Tier 1 market of Dallas with ample power and water for liquid cooling. Riot is taking its time in preparing the 600 MW site for AI data center customers, and it may be Q4 or later before a firm deal is announced. This timing dynamic, possibly months behind other mining peers who have already secured deals, opens the door for potential near-term volatility.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Every Thursday at 4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock and crypto entries and exits. Beth Kindig offers weekly deep dives including lesser-known cryptocurrencies and AI stocks, plus the team offers trade alerts. The I/O Fund team is one of the only audited portfolios available to individual investors. If you’d like to subscribe to the Advanced Market Signals plan, email us at premium@io-fund.compremium@io-fund.com.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

NuScale Signs Deal worth up to 6 GW of Capacity with ENTRA1

Posted on by io-fund

NuScale has operated for nearly two decades as the company was founded in 2007. That’s key as the company is ahead of competitors as the only small modular reactor (SMR) technology to receive two approvals from the Nuclear Regulatory Commission (NRC). The company also has an established manufacturing supply chain and has invested $2 billion over the years for plant licensing and operations. In other words, NuScale has a significant head start.  

We’ve covered the company in the past in the analysis, “NuScale Power: Recent NRC Approval Paves Path to Commercialization” at the time stating the company had a path to doubling its modules in production to 12. In the earnings call, management discussed Doosan, their supply chain partner, having the capability to produce 20 modules. However, an important caveat is that it could take until 2030 until the modules are in the ground: “Anybody says they can get to 2030 COD without these long lead items, I don't see it as a possibility. We're still focused on getting modules in the ground by 2030 time frame.” The reference to COD means “commercial operation date.”  

We also pointed out in that analysis the partnership with ENTRA1, where NuScale acts as more of an OEM by supplying ENTRA1, which in turn, supplies the power directly to customers. Most recently, ENTRA1 signed a deal with Tennessee Valley Authority for up to 6 GW of nuclear power in “the largest U.S. small modular reactor deployment program to date.”  

There are significant risks with this stock as its most recent quarter reported $8.1 million in revenue despite a $12 billion market cap. That puts its sales valuation at a whopping 127 forward PS. The risks peak when the company enters the second half of next year as there will be an increase in opex at the same time as an increase in capex. However, per the rough math, a 6GW deal could be worth as much as $36 billion, thus, it’s reasonable to see the market starting to move into stocks that can supply power by the gigawatt. 

Below, we discuss the post-earnings update on NuScale’s prospects as well as the updated financials and a few key points to consider on this high risk stock. Notably, technicals are in the driver’s seat as the fundamentals do not offer a quality profile, rather it’s a potential momentum play on the AI data center energy theme. We plan to adhere to technical stops to protect the downside, as it could be considerable if the stock market cools off on AI.  

NuScale Signs Milestone Agreement with ENTRA1, Supporting 6GW Capacity 

In late August, NuScale signed a 20-year partnership milestone agreement (PMA) with key commercialization partner ENTRA1, to support ENTRA1’s recent landmark agreement with the Tennessee Valley Authority (TVA) for new nuclear generation capacity of up to 6GW across the TVA’s footprint in six new plants. This is the largest SMR program to date in the US, though TVA offered no firm timelines on when it expects to break ground on the plants or when it expects operations to commence. 

The PMA, lasting through 2045, will see NuScale provide 72 of its 77MWe modules for ENTRA1’s future projects along with a $35-55 million contribution per module, paid out as certain milestones occur: 

  1. 15% of the $35-55 million contribution is due when ENTRA1 signs a non-binding agreement, memorandum of understanding, letter of intent, or framework agreement related to the development of a power project. This 15% contribution applies to 72 modules. (Milestone 1) 
  2. 35% will be due upon the signing of a binding power purchase agreement, energy off-take agreement or the deployment/delivery of one or more SMRs into a power project. The 35% contribution applies to a max of 48 modules at a given time. (Milestone 2) 
  3. 50% will be due upon signing of an OEM manufacturing agreement or other documentation related to the purchase or deployment of SMRs. (Milestone 3) 

The deal provides NuScale with reduced risk related to module deployment, as payments scale based on progress with the largest sums (85%) due upon and after binding deals are secured. The payment structure also incentivizes ENTRA1 to secure deals and progress towards deployment of NuScale’s SMRs across different power projects. Based on the structure of the milestone payments (with ENTRA1 not submitting any Milestone 2 invoices this year with a cap on invoice amounts in 2027), this suggests the two are eyeing the first binding agreements sometime in 2026 to 2027. 

There are a few other contingencies and notable terms within the deal. While pricing per module has been redacted for confidentiality, the agreement will see payments adjust annually, by the greater of either 5% or the CPI Index. Module pricing can be adjusted annually as well, but milestone payments will then be renegotiated. While this accounts for increases in costs, it also could present a larger risk to NuScale as milestone payments could rise substantially over the course of the deal. Additionally, the agreement can be terminated by either party in writing if the other party can no longer pay its debts or enters bankruptcy proceedings.  

As we discussed in our prior Discovery analysis, working with ENTRA1 frees NuScale up from capital-intensive plant development, but also limits its involvement post-deployment. This is expected to offer more flexibility for customers, where they can either assume full transfer of ownership of the plants from ENTRA1, serve as an operator, or simply buy the power under long-term power purchase agreements.   

Analysts have recently raised concerns about the structure of NuScale’s and ENTRA1 partnership, with RBC Capital saying that shifting the financial and development risk to ENTRA “only transfers uncertainty, as ENTRA1 still needs to line up funding and buyers” for NuScale’s modules.  

Rough Math on 6GW 

In the past, we had outlined that 1.85GW in the Standard Power deal would be worth roughly $6B per GW. If we assume the ENTRA1 deal has a similar rate, the deal could be worth around $36B in full.  

 “NuScale has not been upfront about module pricing, though its original 2018 estimate with UAMPS for a 600 MWe, 12-module plant was ~$4.2 billion, or ~$350 million per 50 MWe module, before rising to $6.2 billion for an uprated 720 MWe plant in 2023, or ~$515 million per module.  This suggests that a ~1 GW VOYGR-12 plant could be worth more than $6 billion, though these estimates are based off older figures which included surging raw material costs post-pandemic.” 

The 6 GW PMA does include milestone payments, which, at these estimated module costs, could represent 10-15% of the module price; therefore, the overall revenue opportunity in the long run could be ~85% to 90% of the deal value.  

The Path to Funding Capex May be Alleviated by 6GW Deal with ENTRA1 

In the most recent earnings call, a primary focus was how the company plans to handle higher opex intersecting with higher capex. For timing, H2 2026 was specifically called out as a customer will need to absorb “over a couple of hundred million dollars of CapEx.” The analyst went so far as to call it “carpe diem time.” 

“So is it fair to say that this is carpe diem time that the second half of this year is a critical time to get signed, sealed and delivered customer commitment or 2 that can start shouldering some of this so you don't have to do it all on balance sheet.” 

Management stated the way they plan to manage the costs is to wait until they secure a deal before they start to build inventory. Notably, this conversation occurred before the 6GW deal was announced, yet applies in terms of how management plans to walk a tight rope between building inventory and delivering inventory – which could take a few years if we take at face value the comment that they won’t be putting modules in the ground until 2030. 

Here was the discussion on the incoming higher opex in anticipation of new commercial orders:  

“Yes, Marc, I appreciate you asking that because we didn't want to indicate to our analysts and to the markets that we do plan an increase in OpEx for Q3 and additional increase for Q4. […]  that is in line with our efforts to continue to develop 12 modules and develop our supply chain and invest in the commercialization of NuScale. So it's not — there is an intent to build more than 12 modules right now […] So we've maintained discipline over — I think it's been 6 quarters where we've held OpEx to plus or minus 5% or so. And now we're engaging a very focused and very methodical increase in OpEx in order to engage the supply chain and just get ready for the commercial contracts, which we're anticipating.” 

All things considered, chances are favorable that a hyperscaler or other deep-pocketed AI company shows up by H2 2026 to fund the capex via ENTRA1. We’ve certainly seen a string of deals between hyperscalers and energy companies, a few examples from the long list include Microsoft/Brookfield, Microsoft/Constellation, Google/Brookfield, Google/Intersect, Amazon/Talen, Amazon/Dominion, etc. We’ve seen demand ramp for alternative forms of power as well, such as Bitcoin miner and hydrogen fuel cells. Therefore, I would not be surprised if there is an update to this regard on how the economics of the 6GW benefits NuScale in the upcoming earnings call. 

As noted in the paragraph above, NuScale has more favorable terms than other SMRs as they supply ENTRA1 who deals with the more complex side of selling modules direct and powering the modules in the field. Therefore, NuScale has one customer, and ENTRA1 takes care of the hyperscaler relationships as well as the more costly aspects of a Power Purchase Agreement. This could help the stock reach quicker profitability compared to peers.  

Here is how management explained this: “We're not out here trying to develop technology, develop power plants, against PPAs. That's a huge pull. We're a tech company. We develop technology. Technology is for a NuScale power module. NuScale power module is built. We're kind of an OEM reseller — sorry, we're an OEM seller of a piece of equipment. We outsource that equipment. We outsourced the production. We deliver it to an ENTRA1 power plant. ENTRA1 puts the power front and then sell the power.” 

Double-Clicking on NRC Approvals and Manufacturing Experience 

In addition to having two NRC approvals, NuScale is also differentiated by its manufacturing experience and deep supply chain partnerships. To some extent, the manufacturing experience may provide a more important head start than NRC approval.  

Here is what management stated: “And so there's a 2-part dance here, regulatory licensing, but also you got to get the darn things built. And that takes a herculean effort, and we've been engaged in that for years now.” 

According to previous earnings calls, management stated it takes a minimum of 4 years of operation to secure NRC approval, however recent legislation may have lowered these requirements. There is a lot of red tape involved with energy sector legislation, but the current information shows NRC approval can happen in a little as 18 months as the United States seeks to accelerate nuclear deployment.  

“This Order directs the NRC to complete rulemakings within 18 months to comprehensively revise its regulations and guidance documents, with a focus on balancing safety concerns with the benefits of nuclear energy for our economy and national security. The revisions will include: Establishing fixed deadlines for evaluation and approval of licenses, including an 18-month deadline for construction and operation of new reactors and a 12-month deadline for continued operation of an existing reactor.” You can read more here. 

We covered the NRC deal in our previous write-up stating NuScale recently received NRC approval for its 77 MWe US460 design, which was based in part off its 2023 NRC certified US600 50MWe design. The approval means that the 77 MWe design meets rigorous safety standards and is now approved for use in the US without further review, valid for 15 years. The design can also be referenced by developers and other companies in COLA or construction permits. 

Revenue Inflects Off a Low Base 

NuScale reported Q2 revenue of $8.01 million, a sequential decline from $13.38 million in Q1 but a sharp acceleration from just $0.97 million in the prior-year-quarter. Management noted that the sequential pullback in revenue and gross margin reflects project timing. The YoY ramp underscores that NuScale is moving past the “de minimis revenue” phase into more consistent contract recognition.  

In terms of revenue quality, customer concentration is significant, as related-party Flour drove 92% of Q2 revenue and 69% year to date. As of quarter end, Fluor was also owed ~$2.7 million. This level of customer concentration will be a key risk to monitor as revenues continue to build. From a segment perspective, majority of revenues are driven by Power plant/NPM related services and are broken out as follows: 

  • Power Plant/NPM related services: $7.435M 
  • Energy Exploration Centers: $0.551M 
  • Other: $0.068M 

Looking ahead, consensus calls for $11.3 million in Q3 and $12.1 million in Q4, which would bring FY25 revenue to $47.1 million, representing 27% YoY growth. Analysts model a step-function in FY26 to $152.2 million (+223% YoY) and then $320.6 million in FY27 (+110% YoY) as early deployments scale and VOYGR platforms moves closer to commercial reality.  

Key Revenue Metrics: 

  • Q2’25: $8.01M, down 41% QoQ from $13.38M, up 725% YoY from $0.97M. 
  • Q3’25 (est.): $11.28M 
  • Q4’25 (est.): $12.11M 
  • FY25 (est.): $47.11M 
  • FY26 (est.): $152.52M 
  • FY27 (est.): $320.61M 

While the near-term revenue base remains small, the Street is clearly modeling a steep growth curve beginning in FY26. 

Volatile Margins Reflect Scaling Pains 

Margins remain volatile, heavily distorted by stock-based compensation, which at $5.23 million equaled 64% of revenue in Q2. Gross profit came in at $1.78 million, a 22.1% margin, down from $7.0 million and 52.4% in Q1 but ahead of just $0.12 million (12.1%) a year ago. The sequential pullback highlights lumpiness tied to project timing, though the YoY improvement signals early operating leverage. 

At the operating line, losses widened to $43.1 million versus $35.3 million in Q1 and $41.0 million in the prior year quarter. Operating margin of (534%) deteriorated sequentially from (264%) but remains far improved against last year’s extreme (4,330%). 

Net Loss of ($37.6) million compared to ($30.4) million in Q1 and ($74.4) million last year. On margin, this equated to (467%), highlighting the scale of losses relative to revenue. Importantly, NuScale’s reported results continue to be weighed by high R&D and G&A spend as the company invests to bring SMRs to market. 

Key Margin Metrics: 

  • Gross profit $1.78M, down from $7.00M in Q1’25, up from $.12M in Q2’24. 
  • Gross margin 22.10%, down from 52.35% in Q1’25, up from 12.10% in Q2’24. 
  • Operating loss of $43.08M, down from ($35.33M) in Q1’25, roughly in line with ($41.02M) loss in Q2’24. 
  • Operating margin of (534%), down from (264.10%) in Q1’25, up from (4330.0%) in Q2’25. 
  • Net loss of ($37.61M), down from ($30.40M) in Q1’25 but up from ($74.44M) loss in Q2’24. 
  • Net Margin of (466.97%), down from (224.70%) in Q1’25 but up from (7700%) in Q2’25. 

EPS 

GAAP EPS came in at ($0.13), missing consensus of ($0.11) by ~ 2 cents (a 16% miss). Analysts do not expect NuScale to turn GAAP profitable within the next several years, but modeled losses are narrowing: FY25 ($0.49), FY26 ($0.42), and FY27 ($0.32) on an adjusted basis. The trajectory implies that as revenue scales post-2026, operating leverage should begin to chip away at persistent deficits. 

Key EPS Metrics: 

  • GAAP EPS of ($0.13) actual, behind estimates of ($0.11), miss of 16.07%. 

Cash Burn Accelerates Amidst Buildout but Runway Remains 

Cash burn accelerated this quarter, with operating cash outflow of ($33.3 million), more than Q1’s ($22.8 million) but lower than ($36.0 million) in the Q2’24. As you can see below, this cash burn reduced NuScale’s estimates cash runway from ~21 quarters in Q1 to ~18 in Q2. Operating cash flow margin of (415%) versus (170%) in Q1 reiterates how early NuScale is in monetization relative to ongoing expense, as the company is pouring resources into financing future growth. 

NuScale ended the quarter with $420.5 million in cash, down from $521.4 million in Q1 but up significantly $130.9 million a year ago. The year over year increase in cash is driven equity funding; on August 11, NuScale entered into an at-the-market share sales agreement with multiple investment banks, allowing the company to raise up to $500 million. NuScale says that it plans to use any capital raised for general corporate purposes, including building out SMR projects or R&D. The current ATM program could represent up to ~4% dilution if fully executed.  

Importantly, NuScale carries no debt, preserving balance sheet flexibility even as cash burn accelerates. Deferred revenue of $0.34 million was down sequentially but remains above prior-year levels, reflecting the early stage of backlog conversion. 

A notable datapoint: institutional ownership climbed meaningfully, from 54.5% in June to 66.1% by late September, with shares held increasing from 72.7 million to 88.5 million. That kind of step-up in institutional participation is often a vote of confidence in the longer-term story despite near-term losses. 

Balance Sheet Metrics: 

  • Operating Cash Burn of $33.3M, up from $22.79M in Q1’25 and up from $36.03M in Q2’24. 
  • Operating Cash Flow Margin of (415.73%), down from (170.3%) in Q1’25 but up from (3730.0%) in Q2’24. 
  • Cash $420.47M, down from $521.4M in Q1’25 but up from $130.9M in Q2’24. 
  • Debt burden remains at zero. 
  • Deferred revenue of $0.34M, down from $0.92M in Q1’25, up from $0.08M in Q2’24. 

Bulls vs Bears 

NuScale’s Q2 print reinforces both sides of the investment debate. On the bull side, the company is finally demonstrating tangible revenue traction, with >700% YoY growth, a healthy cash balance of $420 million, and a rapidly expanding base of institutional support. Street estimates call for a steep step-function in revenue beginning in FY26, implying that NuScale’s VOYGR platform could scale into a meaningful clean energy contributor just as global demand for carbon-free baseload power accelerates. 

On the bear side, the near-term financial picture remains challenged. Q2 saw cash burn swell to $56 million, SBC distorted margins with expense equal to nearly two-thirds of revenue, and GAAP EPS missed expectations by a wide margin. Profitability is not in sight through FY27, and execution risk around deployment timelines remains high. 

Taken together, NuScale is a name straddling promise and pain: the long-term narrative is underpinned by regulatory approvals, strong partnerships, and institutional sponsorship, while the short-term is marked by heavy losses and cash outflows. For investors, the debate boils down to conviction in the SMR deployment curve. If NuScale executes, the modeled revenue inflection in FY26 – FY27 could mark the beginning of real operating leverage: if not, dilution and on-going burn may dominate the story. That divergence is the essence of the NuScale investment case. 

Conclusion 

NuScale’s key commercialization partner ENTRA1 has signed the largest SMR deployment deal to date at 6GW across six plants in TVA’s footprint. While the deal could be worth up to $36 billion based off prior estimates per GW, NuScale remains inherently high-risk as operations may not commence until 2030. Additionally, the company trades at an elevated 127 forward revenue multiple, while opex and capex will increase in the second half of next year, raising pressure on funding. As stated in the intro, NuScale is a potential momentum play on the AI data center energy theme, and we plan to adhere to technical stops to protect the downside, as it could be considerable if the stock market cools off on AI.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Every Thursday at 4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock and crypto entries and exits. Beth Kindig offers weekly deep dives including lesser-known cryptocurrencies and AI stocks, plus the team offers trade alerts. The I/O Fund team is one of the only audited portfolios available to individual investors. If you’d like to subscribe to the Advanced Market Signals plan, email us at premium@io-fund.com4:30 pm Eastern, the I/O Fund team holds a webinar for premium members to discuss how to navigate the broad market, as well as various stock and crypto entries and exits. Beth Kindig offers weekly deep dives including lesser-known cryptocurrencies and AI stocks, plus the team offers trade alerts. The I/O Fund team is one of the only audited portfolios available to individual investors. If you’d like to subscribe to the Advanced Market Signals plan, email us at premium@io-fund.compremium@io-fund.com.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

Galaxy Bridges Crypto and Asset Management with AI Data Centers 

Posted on by io-fund

Galaxy is a crypto infrastructure company that is branching out to power AI data centers. Unlike the other Bitcoin miners we have covered recently, Galaxy is deeply entrenched in the crypto industry and will continue its crypto operations while expanding to build an AI campus in Texas called “Helios.” 

On the crypto side, Galaxy offers trading, crypto wallet, custody staking and tokenization. They also offer investment products including a treasury for Solana and recently handled one of the largest trades in Bitcoin’s history at $9 billion. They provide infrastructure for popular Layer-1 crypto company Solana, including being a node validator and helping both Solana and Ethereum to offer liquid staking to institutions. In total across all of the funds and investment products, Galaxy has $9 billion AuM.  

When it comes to data center capacity among the Miners, Galaxy is pursuing one of the largest with a planned 3.5GW of data center capacity. However, management stated they will not see a big impact on revenue until 12-24 months from now: “I do think, to highlight what Chris said, this next 12 months is really important because we go from putting out a lot of money to having an asset that starts spitting out cash. And once 12 months from now, 24 months from now, Helios becomes a big cash generator for us, it allows a lot more flexibility in how we deploy capital.” 

There will be an initial impact on revenue come Q1 2026. CoreWeave is the only customer for Galaxy at this time with the company beginning to recognize revenue at the start of next year: “As a reminder, in the Data Center segment, we do not expect to report financial results until Q1 of 2026 when we begin recognizing revenue from CoreWeave under Phase 1 of our lease agreement.” 

There are puts and takes to Galaxy having a successful, core operation in crypto while also expanding into AI energy. The company only recently began trading on the Nasdaq after being on the Toronto Exchange (TSX) for about five years. The operating margin of 3.4% is better than other Miners due to crypto infrastructure and investment services that offset any losses from mining. However, this may be inconsequential over time given data center deals are high margin. One could argue the lack of focus may make it difficult to compete in two highly competitive markets.  

As discussed further in the call, Galaxy will become more attractive once it diversifies beyond CoreWeave and/or once we are nearer to more substantive delivery of the Helios campus.  

Below, we look at what separates Galaxy and a few items we’d like to see before initiating a position.  

Helios Campus is Massive, and is a Top 5 or Top 10 Data Center Site 

Galaxy’s Helios offers 3.5GW of power with management stating this likely qualifies them for a Top 5 data center site: “We executed and bought another 1 gigawatt of potential capacity around the Helios site. That would make that a 3.5-gigawatt site, got to be one of the top 5 data centers in the world if we get that fully built out.” 

Here is the timing for how that’s expected to come online: 

  • CoreWeave is committed to the 800MW currently approved at the site. This is considered Phase 1, Phase 2 and Phase 3.  
  • Of this Phase 1 is expected to deliver 133MW “throughout the first half of 2026.” 
  • The company is moving toward breaking ground on Phase 2 
  • Timeline is not clear on the additional GWs yet but it was stated there would be an acceleration of sorts on ERCOT approval to where the additional 1GW recently added would likely be approved on the timeline of the 1.7GW that has been under approval. Here is what was stated: “So I would think about the additional 1 gigawatt of interconnect request at the land we are acquiring to be somewhat on par with our existing 1.7 gigawatts in the backlog. And so think about 1.7 gigawatts going to 2.7 gigawatts in terms of timeline.” 

When you combine the fact that management has stated Helios will become a big generator of cash in 12 to 24 months from now, it’s likely they are referring to the 1.7GW to 2.7GW being built out and energized. 

One thing that stood out in Galaxy’s commentary was (at this time) they do not have other customers in the pipeline beyond CoreWeave, stating “I think our partnership with CoreWeave is going to take up the vast majority of our attention over the next few years.” 

Funding Remains the Hurdle  

By now, you may be noticing a theme which is that real estate is the easy part, and even access to GPUs is easing, yet powering the data centers and funding these massive projects remain the bigger hurdles.  

Galaxy ended Q2 with $2.6 billion in equity capital, up more than $700 million sequentially. This increase was fueled by the $480 million primary capital raise in May, appreciation in digital assets and balance sheet investments, and a $292 million one-time accounting adjustment from the corporate reorganization. This equity raise is allocated for Phase 1 Helios (133 MW) and project debt financing is nearly complete. 

While Phase 1 is essentially de-risked from a funding standpoint, it represents just a fraction of Helios’ full 3.5GW potential. At management’s estimated capex of $11-13 million per MW, Phase 1 requires roughly $1.5 billion of total spend. Scaling Phases 2 and 3 to reach the multi-gigawatt buildout will demand tens of billions in additional capital, well beyond the cash and balance sheet equity Galaxy has today. Management has flagged private equity partnerships and refinancing strategies as potential levers, but the implied cost of capital (~10-11%) underscores the financing challenges ahead.  

The takeaway is that the Phase 1 financing is locked in, but the enormity of capital required for subsequent phases remains a gating factor for realizing Helios’ 3.5GW vision. 

Galaxy has secured funding only for Phase 1 thus far, leaving the far majority of the 3.5GW open in terms of how to raise the capital to expand the Helios site.  

Here is what was shared on the earnings call: “As a reminder, the equity portion of Phase 1 has already been funded through our existing equity capital. Once we have secured the project level debt financing, we will have the capital necessary to fund the anticipated CapEx for Phase 1 of approximately $11 million to $13 million per megawatt. For Phase 2, we are still finalizing the design and engineering specifications, but expect the total project CapEx to be slightly higher than the Phase 1 on a per-megawatt basis. We have already commenced work on project level debt financing for the Phase 2 project. Throughout the Phase 1 financing processes, we've established strong relationships with a wide range of banks and private credit managers who are active in the space, and I have confidence in our ability to secure debt financing for Phase 2 in the coming months.” 

The company called out potential private equity deals as one source of capital, with the goal of refinancing mid-project as terms become more attractive. With that said, the rates the company is seeing now were stated to be around 10%, which is quite steep. 

“Yes. James, so I think our expectation on where we land on Phase 1 is in line with what we've articulated in the past. It will come out at a sub-10% stream rate. But when you take into account upfront fees and potential breakage, depending on when you assume we'd have a refinancing event or not, we'll likely end up in the 10% to 11% expected yield in terms of cost of capital, even as credit spreads are tightening in real time.” 

Brief Overview of the Crypto Operations 

Galaxy’s crypto offerings are multi-faceted as the company is deeply entrenched in the crypto industry. Primarily, of the many different crypto products and asset management solutions the company offers, the tokenization of assets could be the most lucrative. Galaxy is one of the largest infrastructure companies by staking weight on the Solana platform, where the tokenization of assets is expected to do quite well. Solana accounts for 46% of on-chain trading volumes compared to Ethereum’s 22%. 

According to the earnings call, Galaxy saw the strongest month in the company's history in July for its digital assets business. Part of this was representing a large transaction worth 80,000 Bitcoin for a value of $9 billion. 

  • Crypto trading desk that offers both spot trading and derivatives. The company is said to have over 1400 institutional partners that trade on its platform. 
  • It’s OTC derivatives platform saw $20 billion in volume last year. 
  • The company is also a large crypto lender with $1.1B in loans in Q2 2025. 
  • The company recently integrated with Fireblocks, a digital asset security platform with 2,000 institutions to help broaden access to Galaxy’s staking solutions. 
  • Galaxy asset management oversees $4.7B across ETFs and other alternative/venture funds. In the case of asset management, it was stated that Galaxy makes a 1% fee. 
  • Six days ago, it was announced that Galaxy will provide the asset management and treasury service for a $1.65B Solana treasury. A primary goal of the treasury is to help integrate Solana into traditional finance markets. It’s expected that Galaxy will provide treasury services for additional assets, as well. 
  • The company was an early investor in Solana and recently tokenized Galaxy’s stock on the Solana blockchain.   
  • Galaxy is also partnered with Invesco as they launched a joint spot Bitcoin ETF in 2023-2024 and have filed for a Solana spot ETF. 

Financials Overview: 

Galaxy Digital is evolving into a hybrid model: part crypto financial services platform, part AI / HPC infrastructure developer. Unlike some of the other miners (APLD, WULF, IREN) that have pivoted hard into AI data center hosting, Galaxy is choosing to maintain its crypto roots while simultaneously building one of the largest AI campuses in the world. This dual-track approach creates both unique optionality and added complexity. This shows up most clearly in the financials, where Galaxy’s reported revenue is inflated by crypto trading inflows, unlike peers whose top lines are directly tied to contracted hosting revenue. 

Revenue Optics vs. Economic Reality 

Galaxy’s revenue optics differ sharply from peers, and it is critical to distinguish between reported revenue from its assets under management (AuM). Reported Q2’25 revenue was $9.1 billion, (-30% QoQ, flat YoY), but this figure primarily reflects grossed-up digital asset sales that are subsequently offset by transaction expenses. By contrast, Reported AuM of roughly $9 billion reflects capital managed or staked on behalf of clients across asset management products, staking platforms, and funds.  

While fee income from this pool was just $17 million in Q2, AuM remains an important context metric because it shows the scale of Galaxy’s crypto franchise and potential fee base as markets expand. For peer context, APLD, WULF and IREN have no comparable AuM base, as their reported revenue reflects directly contracted hosting or mining economics.  

As Management cautioned, “top line is overshadowed by gross principal trading, investors should focus on Adjusted Gross Profit (AGP) instead. On that basis, Q2 AGP of $299 million marked a swing from ($204) million in Q1’25. Given the hybrid model, Galaxy further bifurcates AGP at the segment level: 

  • Digital Assets: $71 million, up 10% QoQ 
  • Treasury & Corporate: $228 million, driven by market-to-market gains on crypto holdings and balance sheet investments. 
  • Data Centers: No revenue yet: Phase 1 lease with Coreweave begins in Q1’26 

Where Galaxy diverges from peers is that once Helios begins contributing, the step-change in recurring revenue could be massive: 

  • Company guidance: For the first 393 MW (Phase I + II), Galaxy disclosed ~$900 million in average annual revenue over the 15-year term, with >$700 million expected in the first full year of energization. 
  • Implied economics: $900M / 393 MW = ~$2.3 million revenue per MW per year. 
  • Scaling up: If the 526 MW of contracted / committed IT load (Phases I – III) is built, this would imply ~$1.2 billion in annual revenue at similar economics. If the full 800 MW of gross approved power were contracted on similar terms, that could reach ~$1.8 billion per year.  

Contrast this with: 

  • APLD: Q4’25 revenue of $38 million was entirely from hosting, backed by $11 billion in contracted CoreWeave HPC lease (~15 years of ~$730 million per year, which scales to ~49x FY25 revenue) 
  • WULF: Began recognizing AI lease revenue is Q3’25 from its Core42 partnership, with early revenue flow already diversifying away from BTC mining. 

The takeaway here is that Galaxy’s reported revenue is inflated by trading flows, but its economic revenue base ($299 million) is far smaller and more volatile. Peers like APLD, WULF, and IREN are already stacking up contracted, recurring HPC revenue, whereas Galaxy’s step-change won’t show up until 2026. When it does, back-of-the-envelope math suggests $900 million to $1.2 billion annually from CoreWeave alone, making it one of the largest HPC lease arrangements in the space. 

Margins Inflect but Volatility Remains 

Operating income improved to $166 million vs. a ($392) million loss in Q1. By comparison, APLD and WULF are still reporting large operating losses as they front-load HPC build outs. These peers are loss-making today but will structurally expand margins as long-term leases begin flowing. Galaxy is already GAAP profitable, but primarily due to market-sensitive mark-to-market gains. 

Net income landed at $30.7 million, compared to losses of ($295 million) QoQ and ($126 million). Its worth noting that these margins remain razor thin: GAAP net margin of 0.3% versus APLD’s recurring hosting margins in the 20 – 30% range. Drivers of the swing: 

  • $135 million digital asset gains and $195 million investment gains, with Bitcoin up 30% QoQ and Ethereum up 36% QoQ 
  • Offset by $127 million impairments and a $125 million loss on derivative notes. 

While peers will see margin expansion as AI leases ramp, Galaxy’s profitability remains tied to crypto market cycles until Helios revenue turns on.  

EPS Turns Positive on Market Gains 

GAAP EPS was positive in Q2’25. Management introduced Adjusted EBITDA as a clearer profitability lens, arguing this better reflects operating trends in Digital Asset and Treasury, removing derivatives noise. Like APLD, Galaxy’s non-GAAP metrics show progress. However, investors should be aware – Galaxy’s adjusted figures are still market sensitive and not contracted. 

Balance Sheet and Cash Flow 

Galaxy’s balance sheet is both its strength and its complexity, as it is large in scale but more volatile than peers due to crypto exposure. As of Q2’25, Galaxy held $1.18 billion in cash / stablecoins and $2.0 billion in net digital assets, giving it more self-funding capacity than miners reliant on continuous external financing.  

Q2’25 Balance Sheet Highlights: 

  • Total Assets: $9.08B, up 43% QoQ from $6.34B 
  • Total Equity: $2.62B, up 38% QoQ from $1.90B 
  • Cash & Stablecoins: $1.20B, flat QoQ  
  • Balance Sheet Net Digital Assets: $1.28B, up 40% QoQ from $908M 
  • Balance Sheet ventures, Fund, and Other Investments: $718M, up 15% QoQ from $623M. 
  • Net Income of $30.7M vs. ($295M) in Q1’25. 
  • Adjusted EBITDA of $211M vs. ($290M) in Q1’25. 

These figures underscore Galaxy’s liquidity advantage today, though the reliance on digital assets and investments make them inherently more sensitive than contracted revenue streams. The crux here is that crypto assets create valuation swings that peers with cleaner hosting-focused balance sheets don’t face to the same degree.  

Peer Comparison: 

  • APLD: $120.9M cash, $688.2M debt (0.18x cash/debt, improving to 0.57x post-quarter raise). 
  • WULF: Similar leverage profile to APLD, funded via staged equity and project-level partnerships for its Core42 build-outs.  
  • IREN: Cleaner balance sheet with low net leverage and industry-low power costs, though at a smaller absolute scale. 

Conclusion: 

Galaxy Digital’s Q2’25 is a case study of complexity vs. visibility. Bulls would say that GLXY provides unique optionality, allowing investors to gain exposure to both crypto financial services (trading, custody, asset management) and one of the largest AI campuses globally (Helios at 3.5GW). Compared to peers, Galaxy also has superior balance sheet flexibility (vs. APLD and WULF) and is already showing GAAP profitability.  

Bears would argue that earnings quality is lower due to market sensitivity, recurring fee revenue is declining, and Coreweave remains the sole data center tenant. Execution risk associated with ERCOT approvals and multi-phase financing is material. Ultimately, Galaxy is a barbell play consisting of crypto upside coupled with AI optionality. While APLD, WULF, and IREN are purer HPC pivots, investors must decide whether Galaxy’s dual exposure should be considered a healthy form of diversification, or distraction.  

In the near-term, upward price action could be driven by Solana’s price, tokenization of assets, or stablecoins. As we go along into 2026, additional positive price action could occur if Galaxy secures a larger hyperscaler or offers visibility on how they will fund the 3.3GW that are left to fund beyond Phase 1. However, there are an equal number of reasons Galaxy and other miners could see volatility as they all greatly depend on high beta being in favor in the broad market. For now, the bulk of a decision on entries and exits will be made with technicals.  

Note, this is a momentum stock and if we were to enter the positions, we plan to adhere strictly to risk management.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading:

TeraWulf Bitcoin Miner: Google Takes 14% Stake via Fluidstack partnership

Posted on by io-fund

TeraWulf is a front runner in the Bitcoin mining space following significant news announcements in August that Google is taking an 8% stake in the company through a partnership with European-based neocloud company Fluidstack.  

The announcement in mid-August was initially for 200 MW over a 10-year period, for $3.7 billion in contracted revenues with a potential to reach $8.7 billion through lease extensions. Of this, Google backstopped $1.8 billion by taking an 8% stake in TeraWulf.  

Days later, it was announced that Fluidstack was increasing its lease option for $6.7B in contracted revenue with a potential of up to $16B in lease obligations. Google also stepped to backstop $3.2 billion for an equity stake worth 14%. 

As you can imagine, the stock surged off the initial deal. The strong price action was supported by the second announcement to where the 130% surge in August has held up. 

The crux of the issue for Bitcoin miners is not real estate and power, rather the quality and longevity of these stocks is dependent on diversifying beyond cash-strapped neoclouds to improve their credit worthiness. To some extent, neoclouds signing lease deals with Bitcoin miners is a house of cards as both are burning cash. For creditors, this means cash flows could fall short of covering the construction costs by both the lessee building the site and the lessor funding the site. 

Therefore, when a Miner secures a hyperscaler, the prospects of raising capital and the terms for raising the capital improve substantially. Miners need access to cheap capital, and having a hyperscaler back their lease deals help to achieve this as it provides committed cash for a lender. 

Bitcoin miners that can offer investors visibility in terms of funding the construction of these massive (and rapid) multi-billion dollar projects will inevitably do better than stocks where capital is uncertain (or is coming from a company that also has to raise capital). 

Below, we explore how Google’s investment in TeraWulf lays the foundation for more durable growth and paves the way for expanded capacity that could (potentially) drive the stock higher. TeraWulf is the second report in a Series of Bitcoin Miners that we have planned for our premium members over the coming weeks. We’ve also covered a thematic overview here. 

Note, this is a momentum stock and we plan to adhere to risk management. 

TeraWulf has Roughly $700M in Annual Lease Agreements  

In the earnings call, management provided more color on the Fluidstack/Google deal, stating the original $3.7B in contracted revenue would provide $350 million a year in lease revenue with net operating margins at 85%. There was an additional 30-day exclusivity deal on data center site CB-5, which is what led to the second announcement a few days later.  

The second announcement essentially doubled the original agreement with $6.7B in contracted revenue. Therefore, one can reasonably expect that TeraWulf will see $700M in annual lease agreements from these two announcements, if we assume similar lease terms as the original announcement.  

There are still a few unknowns, such as when the site will be delivered in order to recognize the revenue on the income statement, if the lease extensions of up to $16 billion will be exercised, and how TeraWulf will fund the capex required for the data center sites.  

Here are the terms of the deal: 

  • 10-year for 200-plus MWs were the original deal terms with management stating it was “200 megawatts of critical IT load, about 250 megawatts of gross site capacity.” 
  • The additional 30-day extension, which was quickly exercised, added another 1602 MW of critical IT load. 
  • According to management commentary on the initial deal, the deployment will begin in H1 2026 with 40MW and be fully deployed by year-end. 

As stated, Bitcoin miners are seeing rapidly improving financials as the deal described above results in an 85% operating margin. In 2022, TeraWulf’s operating margin was awful at (270%) and at (43%) last year. Therefore, the fundamental profile of these deals are providing a sharp rebound that we hope to participate in. 

TeraWulf has over 1GW of Capacity Total 

TeraWulf has two locations: Lake Mariner with up to 750 MW and a new site Cayuga which offers up to 400MW. According to a previous earnings call, the company has plans to add 250 MW more. Prior to the Fluidstack/Google deal, the primary customer was Core 42, which is an AI data center company headquartered in Abu Dhabi.  

The contract discussed above is between Fluidstack and Google at the Lake Mariner site, which is located near Buffalo, New York. There is a new site located in Lansing, New York that offers up to 400 MW on a fully equipped site, “with high-capacity transmission, industrial water intake and redundant fiber.” The Cayuga site will offer 130MW initially by 2027. The amount paid for Cayuga was $95 million in stock with $3 million in cash. 

TeraWulf discusses their data centers as acronyms: 

  • CB-1 will begin to generate revenue this quarter ending in October 
  • CB-2 will begin to generate revenue in the December quarter 
  • CB-3, CB-4 and CB-5 will offer a combined 366 MWs and are stated to be on a tight timeline. Later in the call, management discussed their goal is to typically deliver within a year: “So as you think about it, once we sign a deal, we're typically delivering the facilities within a year. So we're already thinking ahead to meet that time line.” 

How TeraWulf Separates Itself from other Bitcoin Miners 

According to Terawulf, there are a few key reasons Fluidstack and Google chose them over other miners. It’s generally understood that hyperscalers go through intensive site reviews, and thus, there was a question on the call as to what helps Terawulf stand apart. 

  • The first reason is the attractiveness of the site, which management stated ticks all of the boxes: “our installed energy infrastructure, redundant grid connections, the land, the water, fiber latency, 89% 0 carbon power source, availability power.” 
  • Secondly, the company has been working with Core42 on liquid-cooled data centers, and can help Google save time by having already navigated this complexity. 
  • Third, the team has 15 years of experience working on complex infrastructure projects, and is able to confidently execute on these multi-billion dollar projects: “We've got decades of experience executing complex energy infrastructure projects. And I don't think that's something you can replicate overnight.” Building on the “decades of experience executing complex energy infrastructure projects,” TeraWulf announced an acquisition of Beowulf, an energy-powered digital infrastructure company. The deal is worth $52.4 million with $3 million in cash and 5 million in shares. As part of the deal, 94 Beowulf employees will be integrated into the workforce and this will terminate large payments TeraWulf was paying to Beowulf. 

A new fourth way that Terawulf separates itself may be the most distinguishing factor, which is that TeraWulf will have better credit terms than other Bitcoin miners as Google is their largest shareholder and is back-stopping the lease by guaranteeing 50% of the lease value. On the call, this piece was stated to be a game changer as access to capital and creditworthiness is arguably the most important piece to help separate Bitcoin miners. Here is what was stated: “You look at Google as a financial partner here working with us, a multitrillion-dollar market cap, one of the largest companies in the U.S. and with a measly $30 billion, I think, of debt or so […] And so I think we will look to finance the site in totality under this new credit regime and new support regime. So I think that's the obvious and clear takeaway.” 

Management also emphasized that having access to land and power is the easy part of the execution, whereas financing is the harder piece: “Having a signed deal is more than simply having land and access to power. You have to finance the transaction. It's critical to the magnitude of spend. And that's why it was so important to have Google involved pretty much from the start.” 

Investors obviously do not have the ability to do extensive site reviews to distinguish Bitcoin miners. Therefore, hyperscalers will help signal to the market which ones are most attractive, and will help signal to the market which ones will likely be able to execute on any lease deals. Google back-stopping TeraWulf through a deal with Fluidstack is a vote of confidence that we are paying close attention to. 

Potential for More Deals: 

On the most recent earnings call, management discussed the possibility of expanding to add more sites: 

“Finally, regarding our growth pipeline. We are constantly evaluating additional sites to add to the TeraWulf portfolio and maintain an extremely rigorous approach to this process. In 2025, we have evaluated over 75 potential expansion sites, and from that, we have a handful of progressing through negotiations.” 

Perhaps the best statement in the earnings call was the goal of TeraWulf to repeat the Fluidstack/Google deal as much as possible, with a high-level overview that 2026 could be better than 2025 in that regard: 

“On pricing, we're very happy with the economics from Core42 and the Fluidstack Google deal, and think shareholders will be rewarded if we just keep replicating them. I think there's a good argument that the market might even be tighter in 2026 than in 2025 given ongoing power constraints and rising hyperscaler CapEx.” 

Revenue 

Revenue trends tell the story of a miner on the verge of a V-shaped rebound. In Q2’25, revenue came in at $47.6 million, up 34% year-over-year and 39% sequentially, as bitcoin prices rose and hashrate climbed to 12.8 EH/s.  

The prior quarter was softer at $34.4 million, hurt by Polar Vortex and power cost spikes. Importantly, Q2 marks the last quarter of “BTC-only’ revenue. Beginning in Q3’25, WULF will start recognizing contracted HPC lease revenue from Core42’s 72.5 MW deal. 

 Management confirmed that “WULF Den” was energized in July, CB-1 followed in August, and CB-2 is on track for Q4. This timing makes Q3 the Company’s first blended quarter of mining and HPC revenues, which should begin to shift investor perception away from speculative BTC exposure toward contracted, more predictable AI infrastructure income.

As seen above, analyst top-line expectations are signaling confidence in the Company’s transition from pure-play bitcoin mining into contracted HPC infrastructure. Forward revenue growth is projected to accelerate sharply into 2026, with consensus now calling for revenue to nearly double YoY in Q3’25 (103.0% YoY), sustaining that elevated growth level through Q2’26 (87.24% YoY). These estimates imply a steep step-up in contribution from the Core42 HPC lease that began in Q3’25, alongside continued hashrate growth on the mining side. 

This quarter should be viewed as a financial bridge. Historical results remain tethered to bitcoin economics while forward consensus increasingly reflects the durability of contracted HPC leases. The mix shift from volatile BTC mining to long-dated AI hosting contracts is the key re-rating catalyst here. 

Margins 

Gross margin improved significantly in Q2 as power costs normalized. Cost of revenue fell from 71.4% of revenue in Q1 to 46.4% in Q2, bringing gross margin back to a healthier 54%. On a year-over-year basis, gross margin was down from Q2’24, reflecting the impact of the April 2024 halving and rising network difficulty. While volatile, sequential improvement in gross margins should be viewed as a clear positive. WULF’s gross margin of 54% in Q2 now looks closer to peer average (IREN / RIOT mid-50’s), showing that the underlying economics are competitive. 

Operating performance remains pressured by on-going build costs. While gross margin recovered, operating losses continue, reflecting elevated SG&A and depreciation tied to the Lake Mariner expansion. Operating leverage should improve once HPC lease revenues flows, but in Q2, WULF still posted an operating loss. Operating margin headwinds are largely transitional and tied to scaling corporate overhead in anticipation of HPC ramp 

At the net income level, losses narrowed to ($18.4) million in Q2 from ($64.1) million in Q1, still worse than the ($10.9) million in Q2’24. The YoY comparison underscores the ongoing headwinds from the halving and the timing of new infrastructure coming online.  

Earnings 

EPS dynamics remain noisy due to financing and non-cash accounting charges. On a GAAP basis, EPS losses have been volatile: Q2 GAAP net loss equated to ($0.05) per share, improving from ($0.18) in Q1 but worse than ($0.03) in the year-ago period.  

This mirrors the broader group where GAAP EPS is distorted by financing and non-cash charges: APLD shows the same dynamic while IREN is the rare exception already printing GAAP profits due to its GPU cloud mix. For WULF, the cleaner signal is adjusted EBITDA, which turned positive this quarter. 

The underlying picture is more stable on an adjusted basis. Adjusted EBTIDA swung positive in Q2 at $14.5 million compared to ($4.7) showing a meaningful improvement in operating performance when one-time and non-cash charges are stripped out. This mirrors the broader pattern across the miner group, where adjusted results are the cleaner signal of progress while GAAP optics remain distorted by convertible debt, accounting, depreciation, and stock compensation. 

Analyst estimates paint a very constructive picture for WULF’s EPS trajectory – moving from consistent GAAP losses through FY25 into profitability beginning in FY26. From there, the ramp is steep: consensus sees EPS flipping positive in Q4’26 and growing more than 200% YoY by Q4’27. This progression mirrors the broader miner-to-AI pivot: short term pain, medium term inflection, and long-term structural profitability. Analysts are effectively modeling WULF as a credible AI infrastructure provider by 2026, with the EPS trajectory confirming that transition.  

Cash Flow and Balance Sheet 

Operating cash flow swung from an inflated $56.5 million gain in Q1’25 to a $54.8 million outflow in Q2. The sharp sequential decline was largely mechanical. Q1 benefited from a one-time $90 million deferred rent prepayment while Q2 represents a more normalized bun rate tied to the company’s build cycle. On a year-over-year basis, the trend also deteriorated, with Q2’25 OCF falling well below the $16.4 million generated in Q2’24, underscoring the heavier working capital load associated with the HPC construction. 

Free cash flow outflows widened to $174.8 million in Q2, compared to $37.2 million in Q1 and $30.2 million in Q2 of the prior year. The step down was driven by accelerated capex into Lake Mariner data halls, $120 million in Q2 versus $94 million in Q1, on top of weaker operating inflows. Put differently, WULF’s free cash burn in Q2 was nearly six times higher than the same period last year, highlighting the front-loaded nature of its HPC investment cycle. 

Liquidity contracted alongside these outflows. Cash balances fell to $90.0 million at quarter-end, down from $219.6 million in Q1 and $274.5 million in Q4’24, representing a sequential draw of $130 million and a modest $15 million decline year-over-year. Debt held steady at roughly $489 million, leaving the balance sheet pressure solely on cash. 

Working capital shifts add nuance to the liquidity story. Accounts receivable rose modestly to $1.2 million from $0.5 million in Q1. With DSO still under 10 days, collections remain swift, leaving receivables immaterial in the broader picture. Accounts payable held flat at $38.8 million sequentially but remained elevated from $24.4 million in Q4’24, suggesting WULF is leaning slightly on vendor float as a temporary source of liquidity while funding its HPC build. 

Project Costs and Financing 

Cash / Debt Position: At the end of Q2’25, cash stood at $90.0 million and debt at $488.72 million, yielding a Cash / Debt ratio of .18x. This lags peers such as IREN (0.59x) and RIOT (0.38x), while being broadly in line with APLD (0.18x). While liquidity has compressed to $90 million against $489 million of debt, the balance sheet is not deteriorating structurally. Debt has remained flat, and the drawdown is tied directly to front-loaded HPC spend. With Core42 lease revenue beginning in Q3, WULF is positioned to pivot from cash burn to recurring contracted inflows, a trajectory that analysts have started to model into FY26 and beyond. 

Conclusion: 

With Google now a strategic backer through its Fluidstack partnership, TeraWulf is positioned to access cheaper capital and expand its Lake Mariner facility toward its 750 MW potential. The Cayuga site pushes the company past the 1GW capacity mark, and counting.  

While execution risks remain around rapid build-outs and financing, the company stands out as one of the few Bitcoin miners offering investors an opportunity to participate in high-margin AI data center deals. 

We hold a small allocation in TeraWulf and plan to actively risk manage the position.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in WULF at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

Applied Digital: Bitcoin Miner Hinting at Rare, Hyperscaler Deal

Posted on by io-fund

The energy crisis brought about by AI is a strong thematic hat will surface many opportunities for stock investors, yet it’s those companies that supply power quickly that we are building our portfolio with. 

As a Bitcoin miner, Applied Digital has repositioned itself as an AI data center energy stock, with campuses primarily in the Dakotas. With long-term lease agreements — including multi-hundred-megawatt deals with AI cloud providers like CoreWeave — Applied Digital is a front runner that offers lower power usage effectiveness (PUE), an important metric that measures how much electricity is consumed by a data center. This PUE is achieved through free cooling as the Dakota region is colder than other regions and has surplus power the State exports. The company also hinted they are in advanced, direct negotiations with a hyperscaler. If this is confirmed, it will be the first among the miners to sign a direct deal. 

With that said, many Bitcoin miners offer financials that are deep in the red yet are expected to sharply rebound from the high-margin revenue that AI data center deals will provide. Layer-in that Bitcoin miners have a highly volatile trading history, and what you get is a high risk/high reward approach to participating in stocks that promises to solve AI’s largest bottleneck – which is that AI's power consumption is outpacing what the grid can supply. 

For investors, the challenge in participating in this trend is two-fold – how to identify opportunities in the vast energy sector and how to manage volatility given the margins and cash in this sector is lumpy (to put it kindly). We recently covered Bitcoin miners in our analysis “Bitcoin Miners Addressing AI’s Near-term Time to Power Bottleneck with up to $50 Billion in Commitments"Bitcoin Miners Addressing AI’s Near-term Time to Power Bottleneck with up to $50 Billion in Commitments" 

As stated in the analysis: “Instead of having to worry about the prolonged process of site selection, permitting, planning, and more before final power delivery, these neoclouds instead have near immediate access to the powered shell. While retrofitting for liquid cooling, networking and connectivity may still be necessary and pose some challenges, [Bitcoin] miners offer a fast time to operation and relatively cheaper capex costs for a hyperscale-size data center outfitted with tens to hundreds of thousands of GPUs.  

Bitcoin miners can only meet a fraction of this growth, likely around several gigawatts in total. Yet their innate ability to deliver this power over the next 12 to 24 months, supporting up to hundreds of thousands of high-end GPUs in larger-scale facilities, is why miners are prime targets to meet hyperscalers and neoclouds’ immediate power needs.” 

Below we look at Applied Digital in more detail, the first in a Series of Bitcoin Miners that we have planned for our premium members over the coming weeks. 

Note, this is a momentum stock and we plan to adhere strictly to risk management. 

Applied Digital Located in the Dakotas with PUE of 1.18 

Applied Digital made a strong argument in a recent white paper that offering a more strategic location in the Dakotas can result in savings of up to $60 to $90 million per year by offering free cooling.   

We’ve covered in the past the importance of liquid cooling, as AI/ML require massive amounts of data processing, and as future generations of CPUs and GPUs are released, these systems will exceed air cooling capacity. Liquid cooling solves throttling, which occurs when CPUs and GPUs overheat and are throttled back to avoid damage to the chip.  In the case of high-performance computing, liquid cooling reduces total cost of ownership as air cooling requires air conditioning and server fans to run constantly. 

Cooling data center servers is responsible for 40% of the data center energy consumption. According to Dell, enclosed DLC solutions can save up to 23% of energy compared to traditional air-cooled racks. McKinsey places this number at 27% savings when there is 75% liquid cooled and 25% air cooled servers. 

What Applied Digital is proposing is that mechanical liquid cooling is not needed as much in cooler regions such as the Dakotas, stating their mining operation “uses ambient air to cool liquid directly – no mechanical refrigeration needed for significant portions of the year.” 

This represents a PUE improvement from the industry average of 1.58 to Applied Digital’s stated 1.18, translating into estimated annual savings of $60–90 million. In the example below, the calculation highlights $84 million in savings, driven by a significantly lower cost per kilowatt-hour. 

Source: Applied Digital white paper 

When comparing the Dakotas to other regions, such as North Virginia and Texas, the Dakotas have surplus energy with 33% exported compared to these other power-constrained regions. The region also offers 220 days of free cooling, which drives the power savings described above.  

Management discussed these benefits in the recent earnings call, stating: “This design seeks to achieve a projected PUE of 1.18 and near 0 water consumption intended to ensure exceptional efficiency and sustainability. We like this location for its abundant low-cost synergy, some of which is generated from stranded power with over 200 days of free natural cooling. We have calculated that 100-megawatt data center customer could save up to $2.7 billion over a 30-year period as compared to the current industry data centers in other regions.” 

CoreWeave is Primary Customer; Potential Incoming Hyperscaler(s) 

As it stands, CoreWeave is the primary customer for Applied Digital’s capacity. Last quarter, CRWV signed a 15-year lease agreement for 250 MW and an extension for 150 MW for the Polaris Forge 1 data center site located in Ellendale, North Dakota. 

The roll-out is expected to look like this: 

  • First 100-megawatt facility scheduled to be operational in Q4 of 2025 
  • Second 150-megawatt facility coming online in mid-2026 
  • Third 150-megawatt facility planned for 2027  

There are two paths to revenue for this deal. The first is “fit-out” revenue where Applied Digital is paid to prepare the site. This revenue is expected to kick-in for the current quarter and next quarter, leading to a “significant sequential” increase in revenue. From there, the lease deal is worth $11 billion over a 15-year time frame, or about $733M per year averaged out.  

Here is what was stated: “We expect revenue to increase significantly sequentially, beginning in the quarter ending for August 2025 due to the technical fit out of our first Polaris Forge 1 building. Note, our customer pays the cost of this fit-out with a small margin to the company. This fit-out revenue will largely be recognized in both the current fiscal quarter and as well as the quarter ending November 30, 2025. Now this is before the actual lease revenue for the facility begins to be recognized.” 

CoreWeave has a few strategic advantages over hyperscalers when it comes to pursuing energy deals with hyperscalers. The first is the speed in which a smaller company like CoreWeave can negotiate, the second is the company’s willingness to acquire power that is located away from major metro hubs, and third, CoreWeave’s core business is supply fast and AI-optimized access to GPUs, whereas Big Tech has many priorities across their core segments. 

However, with that said, the market will favor any miner that can sign a hyperscaler given CoreWeave’s financials are upside down and require creative ways to raise cash to finance their ambitions. When it comes to a steady, stable flow of cash for a Bitcoin miner (who they, themselves have high debt leverage), there is no better customer than a hyperscaler. You can think of CoreWeave signing lease deals with Bitcoin miners as somewhat of a house of cards – it works well when stocks are up, it does not work well if we go through a longer period of AI softness.  

This is where Applied Digital stands out – the company has hinted they are in negotiations with two hyperscalers with one of the hyperscalers being in “advanced negotiations.” Management made sure to point out how rare it is for a Bitcoin miner to have a hyperscaler as a potential customer. 

The crux of the issue for Bitcoin miners is the quality and longevity of these stocks as  they will have to diversify beyond cash-strapped neoclouds to raise their credit ratings and offer investors visibility in terms of funding the construction of these massive multi-billion dollar projects. Therefore, I’m quoting in full the discussion on the high likelihood that Applied Digital not only secures a hyperscaler deal – but do so soon: 

“Besides CoreWeave, we have completed the diligence and onboarding process with 2 other investment- grade North American hyperscalers […] We also expect to benefit from this competitive advantage as new entrants to the market confront time, money and effort it takes to overcome these industry syncretic barriers to entry for other players. We also, for us, we feel we are now in a position to do business with these companies in the future with a much shorter negotiating and contracting completion process. In fact, we are currently in various stages of negotiation with several investment-grade hyperscalers for large capacity campuses other than our Polaris Forge 1 campus with 1 of those negotiations being in an advanced stage.” 

How Quickly can Applied Digital Add Capacity (and How Much):

Applied Digital has a few paths for announcing more deals. The first is that Polaris 1 offers 1GW of capacity and yet CoreWeave has contracted 400 MW of the available 1 GW.  

Secondly, Applied Digital broke ground on a new $3 billion campus this month called Polaris 2 in Harwood, North Dakota. This new campus is expected to “scale beyond scheduled operations in 2026 and full capacity in early 2027.” 

In terms of size for Polaris 2 and stated delivery for full capacity by 2027, the following was shared in the call indicating it could be at least 1 GW and will be completed in 12-14 months time: “Building on the momentum from these leases and the surging demand for AI infrastructure, we're actively marketing our multi- gigawatt pipeline to a diverse group of customers. […]  As a result, we believe we've reduced our projected build times from 24 months to 12 to 14 months, allowing us to deliver on large-scale commitments faster and more efficiently than before.” 

Note, the exact size and timeline was not officially provided yet the comment above helps to frame initial expectations.  

The Value of 2 GW  

If we assume Applied Digital will lease 2G by early 2027, and if we figure that each 400 MW is worth $73 million based on the CoreWeave deal, then a rough estimate of what Applied Digital's AI factories will be worth is $18.25M per 100MW. This means the total 2GW is roughly worth $365 million. This does not take into account the “fit-out” revenue, which we will get a better idea of in the upcoming quarter.  

Given the 2027 estimates are for $501 million in revenue, taken at face value, one could argue Polaris 1 and Polaris 2 are priced in. This does not take into account additional sites, and if a hyperscaler pays more than what CoreWeave is paying. According to commentary on the call, APLD may be breaking ground on a third site: “We do expect to break ground and work has already started for that on 1 additional campus and potentially 2 before the end of this year.” 

Project Costs and Financing 

When we look at capex costs of $11-$13 million per 100 MW, the cash flow for Applied Digital should pay for the project costs. Financing partnerships and capital raises are central to funding APLD and its HPC buildout. For example, Macquarie has committed up to $5 billion, with $900 million already allocated to Ellendale and $4.1 billion available for future development. In addition, Applied raised roughly ~$875 million in FY25 though equity and debt financing, with another $268.9 million post-quarter.  

Cash / Debt Position: At the end of Q4 FY25, cash stood at $120.9 million and debt at $688.2 million, yielding a Cash / Debt ratio of .18x. Including the post quarter raise, the ratio improves to 0.57x, providing significantly more liquidity runway. Still, this lags peers such as IREN (0.59x) and RIOT (0.38x), while being broadly in line with WULF (0.18x). 

As you can see from the tables below, Applied’s Cash / Debt ratio has been extremely volatile. FY24 cash levels hovered around $30-40 million with minimal debt, spiked to $314 million in Q2 FY25 due to financing inflows, and declined again as capex surged in Q3 and Q4. This underscores the company’s reliance on external financing, followed by rapid deployment into HPC infrastructure.  

Looking deeper into the Company’s investor base, another strong signal comes from Nvidia’s involvement. In September 2024, NVDA participated in APLD’s $160 million private placement financing, landing NVDA a ~3% stake in the Company valued around $63.7 million at the time. Importantly, Applied was already recognized as a Preferred NVIDIA Cloud Partner prior to this deal, meaning the investment represents more than just capital, it reinforces strategic alignment. For Applied, having Nvidia on the cap table provides credibility, potential preferential access to GPU’s, and validates the Company’s pivot into high-performance compute hosting. Relative to peers, few digital infrastructure firms can point to this level of backing, which helps APLD as a trusted partner for hyperscalers and AI-native cloud firms. 

While historical financials illustrate the growing pains of a transition year, investors should focus on the Company’s financing position and go-forward revenue estimates, which accelerate meaningfully over the next four quarters. The $7 billion contracted HPC backlog implies a structural step-up in revenue visibility, while the post-quarter raise improves the cash/debt ratio from .18x to .57x, extending liquidity runway. These forward-looking metrics are more indicative of Applied’s trajectory than backward-looking margins which are distorted by non-cash charges and build-out spend. Ultimately, Applied’s ability to balance its liquidity runway with execution on HPC energization will determine whether the $7 billion in contracted backlog translates into the sharp revenue acceleration that analysts expect. 

Revenue 

Revenue optics are noisy due to Cloud Services fluctuations, but the underlying Hosting growth is much healthier than the FY headline suggests. The Q4 exit run-rate and $7 billion plus in HPC lease backlog indicate a more favorable forward profile. 

Q4 Revenue came in at $38.0 million, representing 41% growth YoY and 4% growth QoQ. YoY growth reflects Ellendale operating at full capacity (vs. partial outages in Q4’24) and the transition away from related-party contracts, leaving a more third-party-driven base. QoQ growth of $1.5 million shows the strength of Hosting revenues amid Cloud Services decline, which was reclassified as “held for sale”.   

FY2025 Revenue of $144.2 million represents 6% YoY growth or $7.6 million compared to $136.6 million in FY24. The growth this year looks weak compared to FY24, mainly due to the termination of legacy contracts in Q1‘25 and inconsistent Cloud Services revenue across the fiscal year. 

While FY25 revenue printed slightly below consensus ($144.2 million vs. $148.0 million), the shortfall was isolated to Cloud Services, which has since been discontinued. Hosting was broadly in line, underscoring the durability of the core business. As you can see in the chart below, analysts expect top-line growth to accelerate sharply moving forward, with estimates calling for $250 million in FY26 and $502 million in FY2027. These strong forward estimates help explain why shares have traded less on historical misses and more on execution / timing of HPC energization. 

Shifting our focus toward the segment breakdown, Hosting is the durable core business with stable recurring revenues between $35 – $38 million per quarter. As mentioned above, Q4 marked the first time both Jamestown (106 MW) and Ellendale (180 MW) facilities ran at stable full capacity. 

Cloud Services drove growth early in FY25 during Q1 and Q2, shrank in Q3, and has now been re-classified as held for sale. These fluctuations are largely driven by GPU contract structure changes and management’s decision to wind down the unit. 

HPC / AI Leases are not currently contributing to FY25 revenue but these are critical for the Company’s trajectory and should be viewed as the growth driver. Management highlighted that signed leases with Coreweave represent $7 billion in contracted revenue to be recognized over 15 years. For added context, this contract alone is roughly 49x FY25 revenue of $144.2 million. Suffice to say, the Company is undergoing transition among its key segments but the strong backlog implies a change in trajectory. 

Margins: Underlying gross margin strength offset by OpEx, SG&A caution signal 

Gross margin trend is positive but material swings in Opex and financing costs mask any improvement. Q4 spike in SG&A is a red flag on cost control that should be monitored going forward. The key takeaway here is that unit economics are improving at the gross level but cost discipline and financing overhands remain key risks. 

Q4 gross profit of $7.8 million, down $7.6 million or 49% compared to Q3, but up $3.7 million or 90% compared to Q4’24. Gross margins of 20.5% vs. 15.2% in Q4’24 reflect 530 basis points of improvement, driven largely by the increase in revenue, along with better facility uptime and efficiency, plus mix shift away from lower-margin Cloud Services. 

Q4 operating income of ($20.7) million, $1.7 million lower than Q3’25 and $18.0 million lower than Q4’24. Operating margin of (54.5%) is down substantially from (9.9%) in Q4’24 and heavily pressured by a 115% increase YoY in SG&A spend as the company expanded headcount, increased stock comp, and ramped corporate overhead associated with the HPC buildout.  

Q4 Net income of ($26.6) million reflects sequential improvement of $30.1 million against Q3’25 and annual improvement of $8.7 million compared to Q4’24. Net Income Margin of (70.0%) is up from (131.3%) reported in Q4’24, driven by higher revenue and better gross margins, slightly offset by increase in opex mentioned above. 

FY25 gross profit $42.7 million is up $12.8 million compared to FY24. This represents a gross margin of 29.6%, 770 bps of incremental progress compared to 21.9% seen in FY24. The full year improvement is largely driven by increased scale and less power-related disruptions. 

FY25 operating loss of ($16.8) million reflects improvement of $16.1 million compared to ($32.9) million operating loss in FY24. This represents an Operating margin of (11.7%) compared to (24.0%) in FY24. Despite the heavy expenditures noted in Q4, margins improved on a full-year basis, indicating operating leverage. 

FY25 Net Loss of ($161.0) million continues to expand versus ($74.0) million loss reported in FY24. This represents Net Margins of (112%), down compared to (54.2%) in FY24. Despite the top line growth and improving unit economics, net margins for the full year worsened due to $119 million in non-cash losses tied to debt conversions, warrants, and derivative liabilities. 

EPS 

Adjusted figures provide a more accurate view of operating performance, showing narrowing core losses, while GAAP EPS was heavily distorted by financing and derivative accounting. Investors should focus on Adjusted EPS, which shows sequential and YoY improvement, while GAAP EPS remains noisy until the financing structure stabilizes. 

Q4 GAAP EPS of ($0.12) improved from ($0.28) in Q4’24, reflecting stronger revenue and gross profit at both Jamestown and Ellendale as uptime normalized. On an adjusted basis, EPS narrowed to ($.03), close to breakeven, as higher gross margin partially offset heavier SG&A tied to HPC build-out. 

FY25 GAAP EPS of ($0.80) worsened from ($0.65) in FY24 due to $119 million in non-cash charges (debt conversions, warrant issuances, and derivative liabilities). Adjusted EP of ($.06) versus ($0.11) in FY24 highlights meaningful progress in narrowing core losses despite heavy investment and Cloud services volatility. 

Cash Flow and Balance Sheet 

FY25 was defined by record capital deployment into HPC infrastructure. While Q4 cash burn moderated sequentially, full year OCF and FCF deterioration underscore the scale of investment and reliance on external financing. Liquidity was extended through large capital raises but leverage ballooned. Sustained improvement in OCF and eventual FCF stabilization will hinge on timely energization and monetization of HPC leases.  

  • Q4 Operating Cash Flow of ($15.7) million, up from ($37.2) million in Q3’24, as working capital normalized. Operating Cash Flow margin improved to (44.3%) versus (101.9%) in Q3. 
  • Q4 Free Cash Flow of ($79.2) million, improved from ($261.5) million reported in Q3, with FCF margin narrowing to (208.4%) from (716.4%). Sequentially, lower capex spend drove the improvement. 
  • FY25 Operating Cash Flow of ($115.4) million down significantly from $13.8 million in FY24, with OCF margin falling to (80.0%), down from 10.1% reported in FY24. 
  • FY25 Free Cash Flow of ($797.0) million, down significantly from ($128.0) million in FY24. FCF Margin of (553.0%), down from (93.7%) in FY24, reflecting the heavy investment cycle for Polaris Forge and HPC facilities. 
  • Cash of $120.9 million, up from $31.7 million in Q4’24 and up $46.7 million sequentially, was boosted by roughly $875 million in equity and debt financing. As noted by Management, this “does not include the additional $268.9 million in proceeds from our ATM and Series G preferred stock offering that occurred post quarter.” 
  • Debt of Q4 was $688.2 million, up from $653.3 million in Q3’25 and a significant step up compared to $79.5 million as of Q4’24 (nearly 9x YoY increase). 

Conclusion: 

Applied Digital asserts they are unique due to being located in the Dakotas where free cooling is available for more than half the year. The company also points toward this region’s surplus of power as a strategic advantage compared to power constrained regions like Texas and North Virginia.  

Most importantly, should Applied Digital announce a deal with a hyperscaler, then it will be the first among the miners. Should the deal materialize, it would indicate their operations are attractive on a competitive basis given hyperscalers have a slow, thorough process for site review.  

We hold a small allocation in Applied Digital and plan to actively risk manage the position.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in APLD at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

Updated Nvidia Stock Price Target – AI “Bubble” Narrative Ignores Re-Acceleration in Big Tech Capex 

Posted on by io-fund

Nvidia stock faced a rare price target cut last week following a weaker-than-expected Q2 and heightened competition from Broadcom. The company has faced numerous headwinds from China and also production delays in their current generation of GPUs. 

As someone who has tracked Nvidia’s cyclical downturns closely since 2018,the culmination of negative narratives in-between GPU generations is eerily familiar. Needless to say, in the past, Nvidia’s cyclical bottoms were buying opportunities where getting the stock lower helped to increase stock returns in the span of a few months. 

Buying Nvidia stock at the October 2022 market bottom delivered over 10X returns within three years

Investors who bought Nvidia stock at the October 2022 bottom have seen more than 10X returns in just three years, underscoring Nvidia’s explosive AI-driven growth. Source: YChartsYCharts

Nvidia’s softer price action this past month is partly due to Broadcom hinting they will see strong AI growth next year. There were some important numbers in the report that we covered for our premium members to help gauge just how much growth Broadcom could see.  

Although I’ve been quite vocal around Broadcom’s importance as a contender on AI accelerators, the chances are low that custom silicon can compete with the powerful and versatile NVL72 systems that Nvidia is shipping now. There will certainly be a time when custom silicon helps to drive down costs for Big Tech, especially around inference. However, it is too early for this to be a serious threat to Nvidia given one of their most important GPU generations to date is (finally) shipping in volume. 

There are other clues that Nvidia’s reign will continue. Prior to the earnings report, I discussed why Big Tech is providing an important green light, stating the capex numbers we saw as recently as two weeks ago are “beautifully packaged” to help guide AI investors on where Nvidia’s stock can go next.

In the analysis below, my firm crunched the hard data on Q2 capex numbers and what is coming down the pipe for Q3. If you are an AI investor like we are, this is an analysis you will not want to miss as the numbers below help to dictate how much room is left in many AI stocks. 

As a final highlight, we are rolling out a brand-new buy plan for Nvidia stock – which is highly dependent on capex growth. When it comes to Nvidia buy plans, many of you are aware the I/O Fund has a strong track record — from our $3.15 entry in 2018, to buying at $10 in October 2022 on the very day the stock bottomed before the AI surge, with real-time trade alerts sent to Members.  

More recently, we stood apart by urging caution just before the DeepSeek news, outlining a “wait to buy” strategy in the analysis “Where I Plan to Buy Nvidia Stock Next” as we anticipated Nvidia could dip below $100. That call paid off, allowing us to secure shares at $95 and $88, again with real-time alerts to Members. 

Below, we present a new buy plan for you below plus a few reasons why I am (once again) saying the Street has Nvidia all wrong. 

Looking back, Q1 Capex was Weak with 2% QoQ Decline 

Before deep diving into Q2’25 growth, let’s contextualize Q1 and why the AI trade was soft earlier this year beyond tariff impacts. Total Capex of $77.3 billion, was up 62.4% YoY from $47.6 billion in Q1’24 but down (2.0%) QoQ when compared to $78.9 billion in Q4’24.  

As seen below, this was the first quarterly decline in capex in 8 quarters, likely triggering some doubts regarding the durability of spend for short-sighted investors.  

Big Tech reported a rare 2 percent capex decline in Q1, contributing to Nvidia stock weakness earlier this year

In Q1, Big Tech reported a rare 2% decline in capex, pressuring Nvidia’s stock earlier in the year as AI infrastructure spending slowed.

It is important to note that this weakness was broad-based, with three of four hyperscalers declining, signaling a timing issue rather than a fundamental demand problem. During earnings calls, management teams largely cited timing issues and power constraints as the main contributors to the QoQ decline — specifically grid hookups, permitting delays, and PPA negotiations. Needless to say, investors were especially focused on these capex numbers coming into Q2.   

AI Stocks, like Nvidia, set to benefit from Q2 cap-ex acceleration 

In the table below, Big Tech Capex is surging QoQ in Q2 with Alphabet, Amazon, and Meta all spending 24% and up to 30% more QoQ on AI infrastructure. This far exceeded previous estimates which led to further upward revisions of capex guidance for the full year.  

Aggregating each company’s reported capex—Microsoft and Meta including finance leases, Google technical infrastructure, and Amazon cash capex—for an apples-to-apples comparison while respecting disclosures

In Q2 2025, Big Tech companies — Microsoft, Meta, Alphabet, and Amazon — accelerated capex spending by 24% QoQ to $95B, raising full-year guidance to $359B. This surge in cloud and AI infrastructure investment is a strong tailwind for AI stocks, including Nvidia.

Q2 hyperscaler capex hit a combined $95.0 billion (+23% QoQ, +63% YoY) with multiple raises and Q3 run-rate signals. Street’s FY25 roll-ups moved from ~$300B to ~$359B (+20% upward revision) and still look light if energization stays on track. These dollars are concentrating in servers / accelerators, DC shells, interconnects, cooling, and networking – setting up a sustained spend momentum into 2026. 

The graph below shows a sharp uptick specifically in Q2 2025 compared to the previous quarter, helping to illustrate why AI spend is not slowing down (quite the opposite). This had an important readthrough for Nvidia and other hardware players like Broadcom. 

Rising big tech capital expenditures are supporting a higher move in Nvidia’s stock price

Big Tech capex growth is fueling AI infrastructure demand and supports a higher move in Nvidia’s stock.

Below, we take a look at each Big Tech company and what they are individually communicating about the year ahead for AI stocks.

Microsoft Guides Capex 27% Higher than Street Estimates 

Q2 capex totaled $24.2 billion (+13% QoQ, +27% YoY), reversing Q1’s brief dip and setting a quarterly record. MSFT confirmed this was driven by AI data center build-outs, with a heavy mix of short-lived IT gear (GPUs, servers, networking) versus real estate.  

CFO Amy Hood stated in the earnings call that capital expenditures were expected to exceed $30 billion in Q3, driven by “continued strong demand signals”, specifically referencing AI-related services and Azure.   

Azure revenue was reported as a standalone metric for the first time in its latest quarterly report, being stripped out of “Azure and Other Services.” The company stated Azure saw $75 billion in revenue or growth of 34%. For comp purposes, the original segment grew 39% up from 33% / 35% on CC basis last quarter.  

According to the CEO, Microsoft is ahead of other hyperscalers in speed of data center buildouts: “We continue to lead the AI infrastructure wave and took share every quarter this year. We opened new DCs across 6 continents and now have over 400 data centers across 70 regions, more than any other cloud provider. There is a lot of talk in the industry about building the first gigawatt and multi-gigawatt data centers. We stood up more than 2 gigawatts of new capacity over the past 12 months alone. And we continue to scale our own data center capacity faster than any other competitor.” 

The heightened guide was notably higher than both analysts’ expectations at the time and higher than Microsoft’s own Q2 level (~$22.6 billion), marking a sharp step-up for the quarter. FY25 total capex is now expected to reach ~$80 billion, well ahead of original Street expectations of ~$63 billion going into the year.  

Alphabet Q2 is Highest Quarterly Capex Spend Ever 

Q2 capex came in at $22.4 billion, up 30% QoQ & 69% YoY, marking its highest quarterly spend ever. This quarter made a significant step-up as Google accelerated TPU vNext production and bulk GPU orders alongside new data center builds.  

Alphabet raised FY25 capex guide by ~$10 billion, now targeting ~$85 billion versus ~$75 billion previously. Management guided continued sequential growth in Q3, though not the same YoY % step-up seen from ’23 to ’24. Investment consists of Custom TPUs, Nvidia GPUs and Data center shells and land acquisition. Cloud remains a bright spot, with revenue growth up +35% YoY in Q2.  

This elevated spend is highly correlated with Google’s new chip, Ironwood, considered its largest, fastest, and most power efficient TPU yet – designed for both AI training and inference workloads. Although Management has not provided any direct figures around spend, they noted that in April 2025 that Ironwood “ships sometime later this year. We reckon it will be towards the end of the year, perhaps in the fall and maybe even in the early winter depending on how tough manufacturing ramp is.” 

This move signals Google’s ambition to reduce dependence on Nvidia, win AI cloud workloads, and monetize inference at hyperscale. Alphabet’s AI growth engine is monetizing faster than expected, validating its aggressive hardware pipeline.  

Amazon Plays Catch-Up with 29% QoQ Capex Growth 

Q2 capex totaled $31.4 billion, up 29% QoQ and 63% YoY, reflecting a strong ramp in AWS capacity buildouts. CEO Andy Jassy reiterated that AWS demand outstrips supply, with Q2 spend focused on GPU clusters and Trainium/Inferentia ASIC deployments.  

Amazon is now signaling >$118 billion for FY25 capex vs ~$100 billion prior. Management expects Q3 capex to stay near Q2 run rate as new data centers energize and custom silicon ramps. Capex is split between Nvidia GPUs (~196K Hoppers deployed in 2024, growing further in 2025) and Next-gen Trainium 3 clusters. This capex spend is necessary to accelerate growth for AWS to compete in the marketplace, as Azure is currently growing twice as fast and gaining market share. 

AWS remains the core profit driver, with >30% OP income growth for five straight quarters. Market analysts see potential for AWS re-acceleration as Morgan Stanley analysts noted that “Amazon Web Services could experience growth above 20% in 2026 as the company expands capacity to meet increasing demand”. The bank’s analysts now have “more conviction that AWS growth has the potential to accelerate to 20%+ in ’26’” which would be “ahead of our base model and key driver of AMZN’s multiple,” with a base-case valuation of $300 per share and a bull-case scenario of $350.”  

While extremely optimistic on AWS, Morgan Stanley did note that that the company was continuing to operate among “capacity constraints (data center builds, delivery of chips, racks, cables, power, etc.), which our new analysis suggests AWS is working through… which we view is a positive signal of faster AWS revenue growth ahead.”  

Meta Capex up 100% YoY 

Q2 capex reached $17.0 billion, up 24% QoQ & 100% YoY, making it Meta’s largest quarterly spend ever. Sequential growth was driven by server deployments, with data centers and networking also stepping up. Meta tightened FY25 capex guide to $66–72B billion vs $64–72 billion prior. Management emphasized “significant growth” in 2H25, led by servers for AI training/inference and core product refresh cycles. Servers remain the largest driver and largest portion of spend.

In its latest quarter, Management stated that while the “infrastructure planning process remains highly dynamic, we currently expect another year of similarly significant CapEx dollar growth in 2026 as we continue aggressively pursuing opportunities to bring additional capacity online to meet the needs of our AI efforts and business operations.”

Meta’s rapid deployment of custom MTIA accelerators and GPUs positions it as a top-3 AI infrastructure buyer. AI-driven ad optimizations have boosted pricing and are offsetting slowing impression growth, adding incremental ROI to Meta’s infrastructure spend.

FY Capex Guidance: What These Guides Imply for Q3 & Q4 

On the topic of full year guidance, all of these firms have significantly increased the fiscal -year 2025 capitalization budgets, largely due to AI, data center buildouts, and cloud infrastructure. As seen below, coming into FY25, Big Tech was expected to spend $300 billion in capex. Following Q1 & Q2 earnings announcements, these estimates have now shifted to $359 billion.

Alphabet raised 2025 capex guidance to 85 billion dollars, Amazon may exceed 100 billion with analysts expecting 117 billion, Microsoft plans 80 billion, and Meta narrowed guidance to 66–72 billion

Alphabet raised 2025 capex guidance to $85 billion, Amazon is projected to exceed $100 billion with analysts estimating $117 billion, Microsoft plans $80 billion in 2025 capex, while Meta narrowed its guidance to a range of $66–72 billion.

Of the Big Tech players, only Microsoft is indicating that growth could moderate:

  • Microsoft provided commentary around its capex spend, noting that elevated capex in the first half of its fiscal year (which includes Q3 of calendar year) is due to large finance lease site deliveries and datacenter ramp-ups. This indicates there may be some moderation in H2 growth driven by a softer Q4 figure.
  • Meta has continued to raise full year guidance which indicates spend remains elevated, with Ad revenue seasonality and infrastructure ramp-ups likely pushing strong spending into H2.
  • Amazon indicated that quarterly capex will “mirror second-quarter spending” so the expectation there is a consistent ~$31 billion spend each quarter, in line with Q2.
  • Alphabet will likely distribute spending evenly across H2 to meet its FY capex guide, with potential upside in cloud infrastructure should construction projects advance in Q2.
As shown in the table, Big Tech capital expenditures typically increase in the second half of the year

As shown in the table above, Big Tech capex spending typically skews toward the second half (H2) of the year, supporting stronger growth for AI stocks.

In FY23, H1 represented roughly 45% of total spend while H2 accounted for ~55%. Similar trends exist as we move forward to FY24, as H1 spend accounted for roughly 43% of annual spend while H2 reflected ~57% spend.

If we assume that budgeting systems remain consistent and that H1’25 spend remains within that 43-45% band, that would imply that H2 capex spend of ~$218 billion and full year spend of ~$390 billion, well above current estimates of $359 billion.

AI bulls will point towards this calendar spend cadence as continued fuel for potential upside against current guidance.

What Big Tech Capex Means for Nvidia’s Stock

Nvidia’s Q2 results reported a decline of (1%) for the Compute segment. This alone should have tanked the stock. Instead, we saw a mild reaction because the forward-looking guidance for the Q3 quarter was quite strong. The ramp from Big Tech capex helps to support Nvidia’s strong Q3 guide. That part is key – that Big Tech and Nvidia remain in lockstep, as they have been since the start of the AI boom. 

Sign up for free below to find out the following information: 

  • My prediction on when Nvidia will reach a $10 trillion market cap 
  • Why a semiconductor trough can often be the best time to buy 
  • The I/O Fund’s updated buy plan. We nailed Nvidia buys at $3.15 in 2018, $10 in 2022, and more recently at $95 and $88 after calling for patience ahead of the DeepSeek news. Now we’re rolling out a new buy plan — access for free below.

Prediction: Nvidia Will Reach $10 Trillion by 2028 

Big tech capex will reach mid-$300 billion to high-$300 billion this year with Nvidia downwind from a large portion of this spend.  

The company’s Q3 guidance – supported by strong capex spending – essentially puts the data center on the brink of a $200 billion run rate with $48.5 billion at the midpoint. If Q4 delivers 10% sequential growth, revenue would reach $53.4 billion — roughly 62% higher than analyst estimates from May 2024. That substantial disconnect has been a profitable one for investors who tracked our nuanced analysis. 

Why does this matter? Because hitting a $50 billion quarter this year sets the trajectory toward $75 billion per quarter by the end of fiscal 2027, assuming 50% CAGR across six quarters and 10–11% sequential growth. That would imply roughly 60% growth in the AI segment compared to the most recent $47 billion quarter. This is why we emphasize buying at cyclical lows, even as the market tends to sell semiconductor stocks during the trough. 

In my premium research, I laid out the case for a $200 billion data center run rate — or $50 billion a quarter — followed by $300 billion and eventually $500 billion by 2028. On a 20x forward sales multiple, that path implies a $10 trillion market cap by 2028; an upgrade to my original target of $10 trillion by 2030. 

Our $6 trillion and $10 trillion market cap scenarios remain intact. At a 20x forward sales multiple, $300 billion in data center revenue supports a $6 trillion valuation, while $500 billion supports $10 trillion.  

Analysts don’t see Nvidia hitting the milestone of a $500B data center until after 2033, but I believe it can arrive much earlier by 2028 — especially since AI software will proliferate and is expected to match hardware in size. CUDA plus the AI stack (TensorRT, Triton, NIM, cuDNN and more to come), Enterprise AI software and licensing from robotics hasn’t been fully factored into the consensus. If Nvidia grows revenue at a 30% CAGR between 2026 and 2028, that timeline accelerates by at least five years.  

We’ll adjust our view if the market shifts, but as of now, Q3 guidance puts Nvidia firmly on track toward the $50 billion data center quarter — the first major milestone in the path toward a double-digit market cap. 

Nvidia’s Buy Plan 

Nvidia’s Q3 guide helped to keep the overall trajectory on track, but near-term price action has been soft. The China narrative has been beaten to death, Q2 marked a cyclical bottom as hyperscalers pause before Blackwell ramps, and semis rarely shine at their lows. For now, the AI trade is running on fumes until Nvidia can carry the market again in Q3–Q4.  

Fortunately, we have a buy plan that allows us to capitalize on any softness seen in the stock. 

Nvidia’s Updated Buy Plan 

Since the April 7th low of this year, Nvidia has advanced nearly 100%. The first higher low on April 21st established the trend, which has since accelerated in a near-straight line with only small dips along the way. The structure unfolding aligns with a standard 5-wave pattern, which remains incomplete and suggests at least one additional high before a meaningful correction sets in.

The current decline appears to be a 4th wave consolidation. This pullback should hold above $135 and, once complete, set the stage for a final 5th wave advance toward the $210–$240 region. We are tracking two primary scenarios based on this interpretation:

  • Blue Scenario – The 4th wave extends lower into the $155–$135 target zone, providing a potential buying opportunity. The areas of interest are: $155, $152, $146, $138. As long as price action holds above $135, the 5th wave should advance into the $210–$240 range, completing the larger 5-wave structure from the April 7th low. Below $135 and this scenario gets invalidated, which opens the door to a more prolonged period of volatility.
  • Green Scenario – The 4th wave was shallow and may already be complete. A breakout above $185 would confirm that the 5th wave is underway, with upside momentum likely targeting the upper range of the $210–$240 target zone.

In both cases, the technical setup remains constructive for another leg higher before a larger corrective phase begins.

Chart showing Nvidia’s nearly 100% advance since the April 7, 2025 low, highlighting a potential 4th-wave consolidation and projecting a 5th-wave rise toward the $210–$240 price range with key support near $135.

Closing Point 

The re-acceleration of hyperscale capex into 2H25 is a powerful sign for investors. While Street estimates are catching up, they still underestimate the scale of AI infrastructure build-out. In the near-term, we are looking for Nvidia to grow 50% year-over-year or about 10% to 11% QoQ on average. Further out, the company must clear 30% YoY growth to reach a $10 trillion market cap by 2028. Given Nvidia’s rapid product road map and the AI software opportunity, this is a low bar for the clear leader in AI.  

The I/O Fund issued 22 buy alerts between March and April of this year, targeting the AI economy. We now have a handful of positions up over 80%, one position up over 100% and two entries up over 300% in 2025.  Our cumulative returns of 210% over a five-year period would place us as #2 if we were a hedge fund and #5 if we were an ETF. Learn more here

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in NVDA at the time of writing and may own stocks pictured in the charts.

Recommended Reading:

Bitcoin Miners Addressing AI’s Near-term Time to Power Bottleneck with up to $50 Billion in Commitments

Posted on by io-fund

This analysis was published on the Discovery tier on August 29, 2025. Learn more about Discovery hereLearn more about Discovery here

Time to power is the most important focal point for AI investors at the moment, considering the high level of demand for compute, the inability of Big Tech to meet said demand in the cloud, and the rapid upgrade cycle between GPU generations.  

Once exposed to the volatility and uncertainty of Bitcoin prices and halving, transitioning to offering AI hosting provides more lucrative, predictable, and higher-margin revenue streams for Bitcoin miners. The AI industry is now giving miners a vote of confidence, with up to $50 billion committed in long-term deals this year for AI/HPC data center hosting. 

This boils down to two key advantages miners can provide: the ability to retrofit or repurpose existing space for GPUs with lower incremental capex than building AI data centers from scratch, and quicker time to power.  

Below, we discuss why time to power is of utmost importance to the AI industry and for AI investors, how Bitcoin miners can bridge the gap and meet near-term power needs, and which BTC mining stocks are engaged in multi-billion dollar AI deals.  

The I/O Fund bought shares in Core Scientific early in this trend after first analyzing its AI angle exclusively for our Discovery members. We closed the position shortly after CoreWeave’s acquisition announcement for a 193% return. To subscribe to Discovery with 30% off, please click here to email us or email premium@io-fund.com and mention code DISCOVERY30.click here to email us or email premium@io-fund.com and mention code DISCOVERY30.

Why Time to Power is Mission-Critical 

Power connection is quickly becoming a primary constraint for the AI industry as GPU bottlenecks ease relative to the last two years. Essentially, long lead times for grid interconnection means hyperscalers and specialized AI neoclouds like CoreWeave simply cannot get enough capacity online to meet high demand. This was noted by Microsoft in April and Amazon this quarter, with Amazon CEO Andy Jassy not shy in saying that “the single biggest constraint is power.” 

Here's why time to power is now emerging as a critical focal point: 

Grid connection timelines in key regions are 36+ months, per data from TD Cowen and Riot Platforms, with major Data Center Alley in North Virginia at 42 months or longer. Even in 2024, Bloomberg reported that utility Dominion said large data centers in Virginia (>100MW) were facing seven year wait times for new connection hookups.  

Additionally, the most powerful GPUs from Nvidia (and now AMD) are being upgraded on an annual cadence, so any delays in getting GPU clusters online shortens the time these chips are ‘useful’ before becoming outdated by the next generation. Think of it this way: hyperscalers and GPU providers do not want to spend tens of billions of dollars on AI hardware to then have it sit idle while waiting for power, as this translates directly into lost revenue and lost profits.  

Tying into this, if a company like Microsoft can get new data centers connected to the grid and stood up faster than competitors such as Amazon, they can then meet higher demand, win new customers and gain market share. For example, AI startups simply seeking capacity to train models likely have lower cloud switching costs, meaning that these companies could easily and quickly shift workloads to whichever cloud provider can offer them the capacity they need the quickest.  

Bitcoin Miners Provide a Path to Meet More Immediate Power Needs 

While some companies like Meta are building 1GW+ clusters from scratch, for others in the neocloud arena that do not have access to tens of billions in cash, Bitcoin miners are recently attracting increased levels of interest. Miners help address immediate power needs with multiple gigawatts of grid-connected facilities that can quickly be converted from mining to AI GPUs. As TeraWulf’s executives put it, the “ability to scale quickly provides a meaningful advantage in today’s race to secure power and compute capacity.” 

Consider some of those challenges that exist on the power side when building data centers from the ground up – short power supply in key markets like Virginia, grid connection requests extending four to seven years, or permitting delaying full power delivery to new infrastructure. Miners, on the other hand, streamline this process by offering access to readily available grid-connected power, cooling infrastructure, and low latency to major hubs.  

Core Scientific, Applied Digital and Galaxy are expecting to bring their first AI-focused facilities online by Q4 2025 through 1H 2026, or four to ten months from now, substantially quicker than new construction. Transitioning existing facilities also offers rapid time to operation; for example, IREN stated that it seamlessly transitioned from mining ASICs to AI GPUs at its 50 MW Prince George facility in six to eight weeks, though this is likely at a very small scale considering the size of its GPU fleet. 

For a company like CoreWeave or Fluidstack, signing long-term contracts for capacity with BTC miners and fronting the cash for capex allows them to build hyperscale clusters for a fraction of the cost and time. As CoreWeave’s CEO Michael Intrator put it, “Right now the key bottleneck really is the powered shell. When you think about that, that’s really the building, the cooling, the distribution of electricity.”  

Instead of having to worry about the prolonged process of site selection, permitting, planning, and more before final power delivery, these neoclouds instead have near immediate access to the powered shell. While retrofitting for liquid cooling, networking and connectivity may still be necessary and pose some challenges, miners offer a fast time to operation and relatively cheaper capex costs for a hyperscale-size data center outfitted with tens to hundreds of thousands of GPUs.  

Capex Costs Highlight Miners’ Attractive Positioning 

Considering that most of the miners that have struck deals are rather cash-strapped, these GPU cloud firms behind the deals are putting up all (or the majority) of the capex required to retrofit existing facilities for high-end AI GPUs. Miners are highlighting the capex costs per MW for these facilities, and these costs, while varying quite wildly between miners, help emphasize the industry’s rather attractive positioning when factoring in quick time to power.   

Core Scientific had offered one of the cheapest project costs, costing CoreWeave approximately $1.5 million per MW, though Core Scientific’s executives had stated in Q3 2024 that “they were able to significantly buy down their rates.” 

IREN is touting the next lowest capex costs at ~$6-7 million per MW of IT load, for its Horizon 1 facility in Texas, which it says is catered to liquid-cooled NVIDIA Blackwell GPUs and provides ~6ms latency to Dallas. IREN says it expects the capex to be funded primarily by colocation customer prepayments and debt financing. 

TeraWulf expects its capex per MW of critical IT load to be in the range of $8-10 million for its recent deal with Fluidstack, while offering <8ms latency to New York City and Boston and <2ms to Toronto.  

Applied Digital and Galaxy are a bit on the higher end – both are anticipating capex of $11-$13 million per MW, with Galaxy recently securing $1.4 billion in project financing to aid development of its first 133MW phase with CoreWeave. However, Galaxy expects capex to move higher for its second 260 MW phase.  

Compared to new construction costs per MW, miners can offer some capex savings, with savings becoming more attractive the lower capex costs go, such as in the case of IREN and TeraWulf. Cushman & Wakefield estimates construction costs per MW in the range of $9-15 million across key markets, averaging $11-13 million. This aligns with estimates from CBRE for $10-14 million per MW, though costs can reach $16-20 million per MW in certain cases (or higher).  

Source: Cushman & Wakefield  Cushman & Wakefield  

While Applied Digital and Galaxy may not offer much of a discount relative to new construction, IREN and TeraWulf can offer more pronounced capex savings. Based on a $12 million per MW construction cost estimate, TeraWulf could offer up to 33% savings per MW and IREN up to 50% savings, translating into hundreds of millions to billions in savings at a 400-800 MW scale.  

Time to Power Comparison 

Theoretically speaking, miners with the quickest time to power should enjoy a two-fold advantage: quicker time to revenue recognition, and potentially better deal economics arising from the ability to meet customers’ demand faster.    

For example, Core Scientific’s ability to bring 250MW of power online for CoreWeave’s AI compute needs supports why it was the first to be acquired, as no other miner could match that scale within the same timeframe. 

There are nuances in delivery timelines and expectations that make it difficult to offer a direct quarter-by-quarter comparison, such as Galaxy only offering vague delivery comments for 1H or throughout the year. However, roughly speaking (and excluding Core Scientific), by the end of 2025, Applied Digital should lead with 100MW, yet this shifts by year-end 2026 with TeraWulf aiming to deliver more power.  

By Q4 2025: 

  • Applied Digital expects its first 100MW of power online. 
  • TeraWulf expects 60MW of power online. 
  • IREN could have 50-130MW of AI-dedicated power online, depending on how it retrofits its Prince George and Mackenzie facilities. 
  • Galaxy does not expect to have capacity online. 

By Q4 2026: 

  • TeraWulf expects to have 420MW of capacity online as it progresses with quick expansion. 
  • Applied Digital expects to have 250MW of capacity online. 
  • Galaxy expects to have 133MW of capacity online. 
  • IREN is aiming to energize its 1.4GW Sweetwater facility in early 2026, though there is no indication yet that it will be able to stand up fully operational infrastructure by year-end.  

Long-Term Power Needs Far Outstrip Miner Capacity 

Looking at the bigger picture, the contracted power that miners can provide will only meet a small portion of the projected demand growth over the next three to ten years, and are unlikely to be the sole solution to this structural power shortfall. As we had covered in our free newsletter, Nuclear Power Emerging as a Clean AI Data Center Energy Source, data center power demand is expected to grow at an accelerated 16% CAGR from 2023 to 2028 and beyond. 

For example, Boston Consulting Group forecasts 45 GW of growth in global data center power demand in just three years, from 82 GW in 2025 to 127 GW by 2028.  

In the US, data center demand was forecast to rise as much as 5x over the next decade. Deloitte estimated US data center power demand to triple from 2025 to 2030, from 41 GW to 120 GW, before rising further to 176 GW by 2035. In terms of power consumption, the DOE recently forecast data centers will consume 6.7% to 12% of total US power production by 2028, up from 4.4% in 2023. 

Bitcoin miners can only meet a fraction of this growth, likely around several gigawatts in total. Yet their innate ability to deliver this power over the next 12 to 24 months, supporting up to hundreds of thousands of high-end GPUs in larger-scale facilities, is why miners are prime targets to meet hyperscalers and neoclouds’ immediate power needs.  

Below, we discuss miners that have secured large-scale deals and/or progressing with building out GPU fleets for self-hosted AI cloud services. As it stands, CoreWeave has been the primary undertaker of major AI deals with miners, committing to over $35 billion including its acquisition of Core Scientific; though Fluidstack is making a splash with a deal worth up to $16 billion with TeraWulf.  

Core Scientific Acquired by CoreWeave Following $10B+ Deal 

CoreWeave made a statement to the industry with its acquisition of CoreScientific for $9 billion, a Bitcoin miner and key compute partner. The acquisition communicates that immediate power at scale is paramount. This stems from Core Scientific’s key advantage, as the only miner that will be able to bring 250MW online this calendar year, capable of hosting >100,000 GB200 chips.  

Prior to the acquisition, CoreWeave had contracted 590MW of total capacity from Core Scientific as of February, worth $10.2 billion in cumulative revenue to the miner over the 12-year lease terms. Core Scientific stated in May that it was on track to deliver 250MW of billable capacity to CoreWeave by the end of 2025, with expectations for the full 590MW to be delivered by 2027. 

CoreWeave says that with the acquisition, it will now own 1.3 GW of gross power across Core Scientific's national footprint (including BTC mining facilities), with an incremental 1 GW+ of potential gross power available for future expansion. The acquisition also adds $500 million of estimated fully ramped, annual run rate cost savings by eliminating CoreWeave’s $10 billion lease obligations to Core Scientific.  

This combination of hundreds of MW of deliverable power by year end and substantial long-term cost savings from owning infrastructure outright versus leasing underscore why CoreWeave was quick to progress with the acquisition. 

For a deeper look at Core Scientific, refer to our prior analyses, Core Scientific Q1: Expects 250MW of Billable Capacity to CoreWeave by Year-End and Core Scientific: Hypergrowth with 21X AI Segment Growth Potential.Core Scientific Q1: Expects 250MW of Billable Capacity to CoreWeave by Year-End and Core Scientific: Hypergrowth with 21X AI Segment Growth Potential. 

TeraWulf Signs up to $16B Deal with Fluidstack, Backed by Google 

TeraWulf is no stranger to AI/HPC hosting, having signed a $1 billion, 72.5 MW deal with Core42 in December 2024, with dedicated GPU compute infrastructure coming online throughout 2025. However, its recent deal with Fluidstack represents one of the largest deals to date between an AI cloud firm and a Bitcoin miner. 

  • TeraWulf had originally announced a $3.7 billion, 10-year AI hosting deal with AI cloud startup Fluidstack for 200+ MW capacity at its CB-3 and CB-4 buildings at its Lake Mariner data center in upstate New York.  
  • Only a few days later, TeraWulf announced that Fluidstack was expanding its deal to include ~160MW at CB-5, bringing its total contracted IT load to 360MW and contracted revenue to $6.7 billion. Including potential lease extensions, the new deal could be worth up to $16 billion.  
  • Perhaps more importantly, Google is backstopping $3.2 billion of Fluidstack’s lease obligations to support project financing, and has taken a 14% stake in the miner (up from a $1.8 billion backstop and 8% stake with the original deal).  

TeraWulf is aiming for quick delivery of power to Fluidstack, with the first 40MW phase expected to come online in the first half of 2026 and all 200+MW delivered by year-end; additionally, operations are expected to commence for the 160MW tranche at CB-5 in 2H 2026. Multiple different OEM GPU equipment is expected to be deployed, likely GPUs from Nvidia and AMD, able to serve a variety of AI training and inference workloads. 

Overall, Fluidstack’s deal represents just over 70% of the current available power at Lake Mariner of 500MW, though TeraWulf does have 250MW additional power pending regulatory approval for expansion. This would bring the facility’s total power potentially up to 750MW with its targeted upgrades, or allowing a second deal of similar size to be signed in the future.  

TeraWulf also recently signed an 80-year ground lease with purchase option at Cayuga, securing its exclusive rights to develop 400MW of data center infrastructure “on a fully equipped site, with high-capacity transmission, industrial water intake and redundant fiber.” The company expects to bring the first 130MW online in 2027, while boosting its long-term power capacity to over 1GW.  

Applied Digital a Secondary Benefactor of CoreWeave with $11B Deal 

If CoreWeave’s acquisition of Core Scientific wasn’t enough, it also committed to leasing 400MW of capacity from Nvidia partner Applied Digital at its Ellendale, North Dakota facility, worth ~$11 billion over the 15-year terms. Ellendale (now referred to as Polaris Forge 1) has the potential to scale up to over 1GW capacity over the long run, though the 400MW represents the maximum current capacity.  

Applied Digital expects the first 100 MW data center to be ready for service in Q4 2025, while the second 150 MW data center is expected to be ready in mid-2026. The third 150 MW facility is expected to be ready in 2027. Applied shared some details on the campus, saying that the campus “will feature high-density racks and direct-to-chip closed-loop liquid cooling and air cooling combo” as it entered a $150 million, 36-month convertible preferred equity financing to advance construction. 

On August 18, Applied announced that it is expecting to break ground on its $3 billion, 280MW Polaris Forge 2 data center in Harwood, North Dakota. Applied is aiming to have initial capacity online in 2026 and the facility reaching full capacity by early 2027, a quick 16-20 month time to power. The company also claimed in 2024 to offer approximately 30% lower costs than AWS, Azure and GCP for cloud compute on Nvidia’s GPUs with short, 8-10 week lead times, a key advantage in the race to deliver compute.  

Galaxy: Another CoreWeave Partner with ~$15B Deal 

CoreWeave has struck another deal with digital asset platform and now data center infrastructure provider Galaxy Digital, committing to the entire 800 MW of approved capacity (~526MW critical IT load) at Galaxy’s Helios data center. With 2.7GW of power under load study, Helios has the potential to expand up to 3.5GW, which would make it one of the largest single data center facilities in the US and the world.  

Galaxy stated that they anticipate average annual revenue of more than $1 billion over the 15-year term, based on committed contractual terms, internal capex estimates, and full capacity utilization. This estimate would place the deal value above $15B.  

Source: GalaxyGalaxy

Galaxy is expected to deliver its first 133 MW phase of power to CoreWeave in the first half of 2026, followed by the next 260 MW phase throughout 2027 and a subsequent 133 MW phase throughout 2028.  

To fund this first phase, Galaxy has secured $1.4 billion in project financing debt, providing the $350 million equity requirement for the funding. Management also stated in Q2 that its $480 million in cash proceeds from its May equity raise would go towards capex related to the Helios DC buildout. Overall, the combined phases are no small task, likely requiring close to $10 billion over the next few years, especially considering management stating Phase 2’s will likely be slightly higher than Phase 1 due to the size of the committed load.  

Galaxy made it clear that for Phase 2, discussions for project financing are still “pretty preliminary,” though management expects that as they produce results with Phase 1 and generate returns/revenue, they will “earn the right to achieve larger financings at lower cost.” Simply getting the second phase financed is the company’s primary goal, considering the 16-20 month timeline to delivery.   

IREN Building Out its Nvidia GPU Fleet, Targeting Deals 

In March, IREN hit the pause button on its bitcoin mining expansion as it pivoted to focus more on AI and HPC, given the revenue potential stemming from its >2.9 GW of grid-connected power.  

Unlike peers, IREN is primarily focusing on short-duration contracts, from on-demand use to three-year term lengths, while expanding its GPU fleet to drive growth in its self-hosted AI cloud business. Revenue for its AI Cloud Services surged 94% QoQ to $7 million in fiscal Q4, after rising 33% QoQ to $3.6 million in its fiscal Q3. The company laid out an aggressive $200-250 million annualized AI Cloud Services revenue target by December 2025, up 8-10x versus its current annualized run rate. 

IREN is working to expand its GPU fleet, which is still extremely small considering its power pipeline as GPUs are not cheap at scale. IREN purchased 2,400 Blackwell GPUs in early July for ~$130 million including fit-out costs, comprised of 1,300 B200s and 1,100 B300s.  

In late August, IREN doubled its fleet to ~8,500 GPUs with another purchase of 4,200 B200s for ~$193 million, while securing $102 million in funding for the July purchase. The company has already contracted out its first batch of 256 B200s to an undisclosed customer.  

In its fiscal Q4 report, IREN also stated that it has secured $200 million in GPU financing, aiming to increase its GPU fleet to ~10,900, or an additional ~2,400 GPUs.  

Given that a 50MW data center could be outfitted with ~25K GPUs, IREN is far from outfitting its own facilities with its current fleet, though it is somewhat capital constrained given its $564 million cash on hand.   

The company says that it is “observing demand for multi-thousand air-cooled Blackwell GPUs,” and has ~47MW of capacity still available at its Prince George, British Columbia facility, capable of supporting ~20,000 Blackwell GPUs. 

For its other AI data centers, IREN is targeting calendar Q4 2025 delivery at its Horizon 1 facility in Texas with up to 50 MW of IT load. IREN says that it has several customers undergoing due diligence and contractual negotiations with interest expanding beyond 50MW, though it has not announced any signed deals. The company is also eyeing a complete transformation for Horizon to reach the full 750MW power capacity, noting in fiscal Q4 that procurement is underway for a second 50MW. 

IREN is looking to create an interconnected 2GW data center hub at its Sweetwater 1 and 2 facilities, with the 1.4GW Sweetwater 1 facility expected to be energized by April 2026 and the 0.6GW Sweetwater 2 facility expected to be energized in late 2027. This is a slight acceleration for Sweetwater 2 from early 2028. Together, Sweetwater could support 700K liquid-cooled Blackwell GPUs, in close proximity to Stargate’s Abilene data center.  

Other Miners Considering AI Pivots 

Riot Platforms and Hut 8 are two other miners that are pivoting towards AI hosting, though the two have yet to secure long-term contracts.  

Riot is aiming to transition from Bitcoin mining to AI hosting when “economic and feasible,” and said last quarter that it is “actively progressing toward securing a lease with a high-quality tenant” at its Corsicana facility. As of January, Riot was evaluating using the remaining 600MW of power capacity at the facility for AI/HPC, with 400MW currently geared towards mining. Needham analysts are encouraged by Riot’s possible pivot, saying they “believe Riot’s Corsicana site is one of the most attractive HPC sites in our universe” with amplefiber connectivity for low latency AI inference. Needham believes Riot could be in advanced discussions with potential tenants by 2H 25 and sign a lease as early as Q1 2026. 

Hut 8 has ~3.4MW of HPC capacity operational and ~670MW of mining capacity, and similar to IREN is leasing Nvidia GPUs in the cloud. Hut reported a $2.3 million increase in revenue from leasing ~1,000 H100 GPUs to an unnamed AI developer, which launched in September 2024. Hut 8 also just broke ground on a ~300MW data center in Louisiana, and is expected to invest ~$2.5 billion, while undisclosed future tenants are expected to contribute ~$10 billion in compute equipment. However, it is reported that Hut will likely need external financing or a partner for the project.  

Earlier in 2025, Bitfarms announced it was mulling a shift to AI, and in early August, the miner announced a partnership with T5 Data Centers to advance the development of its Panther Creek facility in Pennsylvania. The two are expected to engage in pre-construction design planning and development approval processes, with Bitfarms also submitting its master site plan to Macquarie for future development.   

Cipher is also considering a pivot, saying that it created a new strategic plan at its 150MW Black Pearl Phase II to bridge both AI compute and hydro-bitcoin mining, though in the long run it expects the site to be fully leased by AI/HPC tenants. Cipher added that it is seeing continuing HPC interest at its Barber Lake site. 

AI/DC Deals are High-Margin, Highly Visible Revenue for BTC Miners 

These AI hosting deals are very attractive for miners as they provide highly visible, high-margin revenue streams, a major operational shift from the prior business model with growth and earnings tied to Bitcoin’s price volatility, network difficulty and halving.  

Applied Digital’s deal with CoreWeave has pricing set upfront with an annual escalator, with average annual revenue of ~$733 million based on the $11 billion, 15-year terms. Revenue is likely to start small and begin to ramp rapidly once full capacity is online. Applied is also targeting 88%, +/- 3% net operating income margins on the AI hosting revenue, or ~$645 million average annual net operating income for the duration of the deal.  

TeraWulf’s expanded deal with Fluidstack will generate average annual revenue of $670 million at its current scope, though the full extension to $16 billion could provide significant revenue upside. TeraWulf is eyeing an 85% net operating income margin for the Fluidstack deal, or ~$570 million on average annually. This is at the low end of Applied Digital’s range, likely accounting for higher operating costs due to location (upstate New York vs North Dakota). Combined with the Core42 deal, TeraWulf has visibility into ~$770 million in average annual revenue, with the first revenue from Core42 now being recognized.  

Galaxy is targeting 90% adjusted EBITDA margins on its AI hosting deal with CoreWeave, implying adjusted EBITDA of ~$900M+ on its average annual revenue estimate of more than $1 billion. Given energization times spanning into 2028 and a longer 15-year term structure, Galaxy’s initial revenue ramp may be more prolonged than peers. 

IREN does not have a firm hosting deal, though it has touted a 97-98% hardware profit margin for its AI Cloud Services, or revenue minus electricity costs. This is more than 20 points above its Bitcoin hardware profit margin, highlighting why AI is more attractive than mining at scale. However, margin tailwinds are likely minimal in the near-term given AI Cloud Services contributes less than 3% of revenue. 

Accelerated Revenue CAGR  

While there are nuances in deal sizes and lengths, IREN, Applied Digital and TeraWulf are seeing accelerated forward revenue growth CAGRs as AI capacity soon comes online. These miners’ growth rates are much stronger than others such as Riot and MARA who have not jumped headfirst into AI hosting.  

Applied Digital and TeraWulf are expected to see revenue increase at an 88-90% 2-year CAGR from 2025 to 2027, a significant acceleration from their respective historical 61% and 73% CAGRs from FY23 to FY25.  

IREN is expected to see revenue grow at a 42% CAGR from FY25 to FY27, though this is a bit skewed as the company has benefitted from rapid mining hashrate expansion and rising Bitcoin prices over the past two quarters. 

Compare this to Riot and MARA, with both expected to see FY25 to FY27 revenue growth at a 17-18% CAGR, decelerating sharply from the 55-60% level over the past two years. While revenue is coming off a higher base than say APLD or WULF, the difference in forward growth rates is notable for AI-engaged miners and these two who have not yet pivoted in full force.  

Thin Balance Sheets Present Capital Raise Risk 

Miners are rather cash-strapped, and while neoclouds and partners are fronting the cash for capex, capital raises and dilution are still a risk, considering IREN and TeraWulf both recently launched larger-scale convertible note offerings.  

Here’s a quick snapshot into the health of IREN, TeraWulf, Applied Digital, and Galaxy’s cash versus debt, with the chart below showing how thin and lumpy cash balances are: 

IREN reported cash and equivalents of $196 million in Q3, down from $455 million in the prior quarter. As of Q4, IREN reported $564 million in cash and equivalents, after closing an upsized $550 million convertible note offering in June, while debt is now at $963 million. This presents possible dilution risk in the high-teens, based on IREN’s $5 billion valuation.  

Applied Digital reported cash and equivalents of $114 million, down from $254 million in the prior quarter.  Applied has a deal with Macquarie for up to $5 billion in financing, including a $900 million initial investment at Ellendale and $4.1 billion on retainer to for future data center expansion. For any future builds, Macquarie would invest $2.25 million per MW and Applied would invest $0.75 million per MW. 

TeraWulf reported cash and equivalents of $90 million, down from $220 million in the prior quarter. However, shortly after the Fluidstack deal, the company announced the full exercise of its $1 billion convertible note offering, or ~27% of its current market cap. Based on prior capital allocation projections, this would likely leave TeraWulf with ~$600-700 million in unallocated cash, for project cost overruns or other expansion needs. 

Galaxy reported cash and stablecoins of $1.18 billion in Q2, including $691 million in cash and $489 million in stablecoins, approximately flat from the prior quarter. Notes and loans payable were $1.07 billion.  

Galaxy President and CIO Christopher Ferraro offered some very important perspective on funding and capacity growth, and why capital is likely to be the limiting factor for miners’ current buildouts:  

“There’s also a practical component, which is, these are very large-scale, long-term development projects that take a lot of capital. And so our ability to grow into the opportunity is wholly dependent on 2 things: one, us executing excellently; but then also two, growing and getting bigger as a company so that we can actually support the growth, meaning like it would be totally imprudent for us to now take on in parallel, for example, like another $10 billion build, because that requires a capital base and the attention and resources that we're just not built out for today.” 

This matches our statements in our Core Scientific analysis from May, Core Scientific Q1: Expects 250MW of Billable Capacity to CoreWeave by Year-End, where we explained that if CoreWeave did not front the capex for the data centers, the “business model would not work as CORZ would struggle to raise the level of capital required to acquire more sites and modify the existing infrastructure.” 

These comments that the current builds underway for CoreWeave and Fluidstack are likely to be the main focus of the miners over the next few years, especially considering the thin balance sheets, convertible note raises, and difficulty from CoreWeave to keep funding multiple different projects.  

Customer Concentration Another Risk to Consider 

One other risk to consider is customer concentration, given the fact that CoreWeave is the sole backer for a majority of the miners discussed here. TeraWulf has the benefit of Google backstopping Fluidstack’s obligations, offering early termination protection for the first six years.  

Considering CoreWeave has made the move to acquire Core Scientific and has signed deals with Applied Digital and Galaxy, it may be more limited in its ability to fund future projects. It also means that miners signing away all or a majority of their power to CoreWeave find it hard to diversify revenue streams away, with Galaxy noting that the deal with CoreWeave “is going to take up the vast majority of our attention over the next few years” and prevent other hyperscaler engagements. 

Also, if CoreWeave should pull away from a deal down the road, it could create a significant blow to revenue and earnings for miners it is currently engaged with. This is due to two factors: high-margin, high concentration of revenue CoreWeave’s deal will drive, and that there is little room for diversification as new capacity is expensive for miners to handle without major financing.

Technical Analysis 

The risk associated with Bitcoin miners is present in the potential setups outlined below. As you will see, some have the potential for wild swings in either direction, which follows very messy and overlapping uptrend patterns. For this reason, we approach these charts only from the mindset of risk management. If we do take a position in any of these names, we do not view them as buy-and-hold vehicles, and all will come with relatively tight risk controls.

Galaxy Digital (GLXY) 

  • Green – GLXY appears to be tracing a very large diagonal pattern. If we zoom into the current drop in GLXY, it appears to be a 4th wave, and needs one more swing higher to complete the larger pattern. The target for this swing is $34 – $43. As long as we hold $19, this setup remains valid. 
  • Blue – Instead of getting one more swing higher, we should see two more. Once we reach $35 – $43, we’ll see another 3-wave drop and a final swing to $56 – $67. Once again, if we break below $19, then both of these scenarios are no longer valid, and a larger top will likely be in. 

TeraWulf (WULF)

  • Blue- We have completed wave 1 in a very large diagonal, as well as wave 2. We are just now completing wave A of 3. This should be a double top that turn lower in 3 waves toward $5.25 – $3.65. The drop needs to be a 3-wave drop to confirm this count. Once this ends, we should see a 5-wave pattern turn higher. This would point toward $69 – $105 in a pretty direct path.  
  • Green – If we instead keep pushing higher over $12.50, with volume and momentum expanding with price, then we could be in a more direct path higher. Instead of a large diagonal pattern, it would be a standard 5 wave pattern. Wave 1 of 3 would push toward $35 – $69. 
  • Please note, while the pattern below allows for these counts, we still only have 3 waves up off the 2022 low. So, until we see either the green count get confirmed, or a 3-wave correction, risk remains high. Any drop below $4.20 will be the first warning. If we see a 5-wave drop break below this level, the risk will increase that something more bearish might be in play.  

Applied Digital (APLD) 

  • Green – APLD is tracing a very large 3 wave pattern, which best fits as a diagonal pattern.  If we can see a breakout on heavy volume and expanding momentum over $17.25, it will confirm this count, which should see a continuation towards $23 – $29.  
  • Blue – We fail to breakout over $17.25 and instead drop back into the $14.50 – $12.50 range. We’ll then break through the $11.35 region and head toward $9 in a larger 2nd wave. 

Riot Digital (RIOT) 

  • Blue – We are completing the B wave in a larger 2nd wave. We will fail under $14.10 – $15.35 and see a sharp, 5-wave drop back to $9 – $8. This will hold, as we set up for a larger 3rd wave breakout to new highs.  
  • Green – We completed the 2nd wave, and it was shallow. We’ll see a strong breakout over $14.10 – $15.35 on expanding volume and momentum.  the 3rd wave target is $27 – $31.  
  • Both counts must hold $7.56 on any weakness, or they will get invalidated.  

Iren Limited (IREN) 

  • Green – Note how the most recent push higher is happening on less volume and momentum. This is the 5th wave, which can continue as high as $35. Once complete, this will end a large 3rd wave within a diagonal pattern. The target for the 4th wave is around $8. We should then see a large 5th wave to new highs.  
  • Red – We only have 3 waves up off the 2022 low.  This is not ideal, and until we see a 3 wave retrace, it is a risk that should not be ignored. If this is all we get, then we should see a large 5 wave drop develop that takes us through $8 and then $2.80. We would then be on our way to new all-time lows.  
  • While both counts have IREN in a 5th wave, if we can see volume and momentum expand with price over $35, then something more bullish may be going on. If we see this then we will adjust accordingly. 

Conclusion 

Bitcoin miners are not able to solve the long-term power deficit that the industry is coming head-to-head with, but in the 12 to 36 month window, miners are uniquely positioned to meet hyperscale and neocloud power needs. AI hosting deals provide highly-visible, high-margin revenue over the next decade and beyond, a major operational shift for miners once reliant on volatile Bitcoin prices for growth. 

The catch is that miners cannot finance these AI data center builds themselves, as they have thin balance sheets and larger debt loads. Instead, they are reliant on their backers such as CoreWeave, and in TeraWulf’s case, Fluidstack and Google, to front the cash for the projects and provide the necessary compute.  

While risk does stem from the fact that CoreWeave is the sole backer for a majority of these AI hosting deals, limiting opportunities for revenue diversification and minimizing concentration risk, AI presents much stronger, higher-margin forward growth opportunities.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Please note: The I/O Fund conducts research and draws conclusions for the Fund’s positions. We then share that information with our readers. This is not a guarantee of a stock’s performance. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis.

Recommended Reading: