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: