Nebius: Mega Microsoft Deal, 5x Growth in Power Fueling AI Cloud Hypergrowth, But High Risk Remains

Posted on September 12, 2025June 30, 2026 by io-fund

The trend toward neoclouds is a high risk/high reward opportunity for investors. Nebius is similar to CoreWeave, dubbing itself as AI native cloud infrastructure, which means the infrastructure was built specifically for AI workloads with architectures built on bare metal servers instead of hypervisor layers, for example. As pointed out in our CoreWeave analysis, this along with a few other optimizations can result in significantly faster training.

Nebius offers an Nvidia-optimized cloud platform for teams that need to adjust compute resources, want high-performance storage and an easy-to-use AI environment yet do not want to manage the infrastructure or AI operations.

The stock surged earlier this month off the announcement of a mega deal with Microsoft worth $19.4 billion. This deal signals just how supply constrained the market is, given Microsoft is willing to partner with a neocloud to scale quickly.

Even with the big moves in neocloud stocks we’ve seen recently, they remain high risk and they are not in the “quality” bucket given their financials are messy. Nebius is even more complicated than CoreWeave given they also own an autonomous driving division, among other investments, and was formally the company Yandex. The Russian-owned Yandex was a high-profile internet search company (the Google or Baidu of Russia), which saw its stock halted after the Russian invasion of Ukraine.

Given Nebius’ roots are from Yandex, one might question if Nebius is truly AI-native as some of the Finnish data center in Europe was built in 2014. However, that is also where one of Nebius’ strength lies, which is the company offers European data centers which helps a company like Microsoft to expand quickly overseas.

To serve the AI-native market, Nebius has been building out colocation sites to increase capacity and lower latency for the incoming inference market. Companies like Cloudflare and Shopify are early customers, both of which need to power inference at the edge. The company is also expanding beyond Europe with data center expansions in New Jersey and Kansas.

With that said, Nebius is a high-risk stock given its success depends on how much capital the company can raise, and the likelihood it remains in CoreWeave’s shadow is high. Regardless of how Nebius competes with CoreWeave, it remains an AI bubble stock as the company has to hope the stock prices goes up to raise cash, which will dilute shareholders (or raise debt). It’s a vicious cycle as during any months/quarters that AI stocks are soft, Nebius stock will carry outsized execution risk.

Ultimately, we are stepping away from the stock. While it’s clear that Nebius could do well in an AI-driven market and the stock could extend rapidly, we want to be prudent about what will hold up should we see a softer AI narrative. We like other names in our portfolio better in terms of participating in the upside while protecting to the downside for when we do get the inevitable tech selloff.

Up to $19.4 Billion Deal with Microsoft, More to Come?

Last week, Nebius signed a mega deal with Microsoft worth up to $19.4 billion through 2031, for capacity at its new AI data center in New Jersey. The announcement sent Nebius shares up 50%, with the contract value being worth more than the company’s $15.5 billion market cap at the time of announcement.

Nebius outlined that it plans to bring GPUs online in several tranches, with initial capacity later in 2025 and ramping throughout 2026. At present, the deal is worth $17.4 billion, with Microsoft having the ability to sign a $2 billion extension should it determine a need for additional GPU services or capacity. Nebius expects the deal will help it greatly accelerate AI cloud growth in 2026 and beyond.

The deal does have a few major contingencies, mainly that Microsoft has the ability to cancel the contract if Nebius fails to meet agreed delivery dates and if it cannot provide alternative capacity to make up for the delay. It’s likely that Nebius will prioritize the development of the New Jersey data center above all else in the near-term to ensure it remains in good standing with Microsoft as the deal offers immense revenue upside.

Unlike miners, who already have pre-built infrastructure ready for retrofitting, this is a build-to-suit development, meaning Nebius is financing the buildout of the infrastructure, racks and related equipment and partnering with a developer who owns the land and power (in this case DataOne).

Nebius expects to fund the capex for this data center with a combination of cash flow from the deal and debt secured against the contract, while noting that additional financing options may be pursued to “enable significantly faster growth than originally planned.” The latter has already panned out, with Nebius raising $3.75 billion in a combined debt & share sale. This is important as the terms of the deal do not indicate upfront prepay, with deferred revenue only $19.3 million, suggesting Nebius will likely have to deliver the first tranche of power before payments commence.

Breaking this down, the $17.4 billion deal is worth approximately $3.48 billion in average annual revenue over the five-year period, or scaling to the several billion in revenue by 2031. Overall revenue estimates for 2028 through 2030 have been raised by $2 billion to $4 billion following the announcement, implying the ramp will be felt primarily in the later stages of the deal.

While the multi-billion dollar upside is certainly a positive, CEO and founder Arkady Volozh hinted that subsequent deals like this are possible: “In addition to our core business, we expect to secure significant long-term committed contracts with leading AI labs and big tech companies. I’m happy to announce the first of these contracts, and I believe there are more to come.”

Considering the New Jersey data center has 300MW capacity with potential expansion to 400MW, Nebius theoretically can support more deals of this size with 1GW+ in its pipeline and aggressive growth plans.

Targeting 5x Growth in Power by End of 2026

Nebius is aggressively ramping up its capacity and has outlined a plan to increase its connected power by ~5x by the end of next year, though it is still targeting 100MW of active power by the end of this year.

This quarter, Nebius stated that it is aiming to have 220MW of connected power by the end of the year, which they define as either currently active or that can be activated upon GPU installation. Management maintained its guidance for >100MW of this 220MW to be active by year-end, though this may be revised higher to account for expedited expansion for Microsoft’s deal.

This includes recent expansion at multiple data center campuses and new campuses soon to be operational:

In Finland, Nebius is currently tripling its capacity to 75MW, which can host up to 60K GPUs.
In New Jersey, Nebius has 300MW under construction, with the first ~100MW to be connected this year; some reports suggest the site can expand to 400MW. Considering this is the site powering the Microsoft deal, it’s likely Nebius will prioritize power connections on an accelerated rate here.
Nebius’ Kansas City, Missouri data center went live in Q1 2025, featuring primarily H200 GPUs, and will deploy Blackwell GPUs through year end. The first phase can expand up to 40MW and host 35K GPUs.
Nebius is launching its first UK data center in Q4 2025, expected to feature 4,000 Blackwell Ultra GPUs.
Nebius is also launching a new site in Israel, joining its global footprint with other operational data centers in Paris and Iceland.

Overall, Nebius is in the process of securing more than 1GW of power by year-end 2026, or nearly 5x growth from 2025’s current guidance for connected power. This includes two new greenfield sites in the US where Nebius is in advanced discussions; management prefers greenfield construction as they can achieve TCOs around 20% below market averages by controlling building design, power flow, and server design and installation.

This growth in power supports Nebius’ medium-term expectations for scaling its fleet, from ~20K in 2024 to 60K in 2025 to 240K over the next few years. This paves the way for Nebius to continue smoothly along its hypergrowth phase, progress towards its medium-term revenue target of ‘several billions’ and support more hyperscaler deals.

To put this in perspective, primary neocloud rival CoreWeave disclosed at the end of 2024 that it had more than 250,000 GPUs installed across a 360MW active power footprint. Now, CoreWeave has 470MW of active power, and 2.2GW of total power contracted with its acquisition of Core Scientific. At its current 470MW active power size, CoreWeave is nearing a $5 billion annual revenue run rate.

For a deeper look at Core Scientific and CoreWeave, read more here: CoreWeave: AI Infrastructure Built for the Next Decade; Upside Down Business Model and Core Scientific: Expects 250MW of Billable Capacity to CoreWeave by Year-End.CoreWeave: AI Infrastructure Built for the Next Decade; Upside Down Business Model and Core Scientific: Expects 250MW of Billable Capacity to CoreWeave by Year-End.

$3.75 Billion Capital Raise to Support New Growth

The day following the Microsoft deal announcement, Nebius priced a $2.75 billion convertible debt raise to support future growth, upsized from $2 billion and combined with a $1 billion share offering at $92.50. It also builds upon a recent $1 billion convertible debt raise from June, allowing Nebius to pursue rapid capacity expansion, high-quality data center site procurement and more GPU purchases.

Considering that Nebius is not as highly leveraged with debt as CoreWeave, it enjoyed significantly better terms for this raise than the other neocloud.

Nebius priced $1.375 billion in 1.0% notes due 2030, and $1.375 billion in 2.75% notes due 2032, with effective conversion prices of $159.56/share.
For comparison, CoreWeave recently closed a 9.0% $1.75 billion note in July and a 9.25% $2 billion note in May.

These attractive low coupon notes save Nebius nearly $300 million annually in debt interest payments versus CoreWeave’s recent raises, freeing up more cash for expansion or capex. However, ~$50 million in interest is nearly half of Nebius’ current quarterly revenue, adding some strain on margins though this will ease as revenue ramps.

What raises the risk for Nebius is a constant need for more capacity – it is already outpaced by CoreWeave, and simply securing >1GW of power is not nearly enough to fuel consistent growth considering some Bitcoin miners have more in their pipeline.

Capacity at scale is not cheap – for a similar 300MW data center like the New Jersey site, total development costs would likely be in the range of $8 billion to $10 billion, assuming greenfield construction costs of $9 million to $13 million per MW, and hardware costs of $18 million to $20 million per MW. Thus, adding an additional 1GW sometime in the future to take total power to 2GW (still below CoreWeave’s 2.2GW contracted), could cost an additional $20 billion.

Vertical Integration, Custom Servers, Competitive Pricing Provide an Edge

Considering CoreWeave’s broadening presence securing capacity deals with multiple Bitcoin miners, it’s important to touch upon what separates Nebius from its neocloud rival, why it may be positioned to secure future large-scale deals and extend its hypergrowth phase.

According to Nebius, the company has a few key advantages that provide it an edge when it comes to AI infrastructure and GPU rental prices:

In-house custom designed servers offer a lower TCO, up to 20% lower versus other cloud providers
Power usage effectiveness (PUE) near industry lead, leading to increased rack density per MW
Data center ownership cuts out expensive colocation fees, allowing for competitive GPU rental pricing while maintaining margins
Vertically integrated with a full-stack, proprietary AI cloud purpose-built for AI workloads

Bringing server design in-house and bypassing traditional server OEMs such as Dell and Super Micro gives Nebius control over component purchasing and rack design, helping maximize rack efficiency and minimize costs. For example, Nebius says that average power consumption for its in-house servers can be up to 20% lower versus comparable off-the-shelf servers, at 8.2 kw versus 10 kw for a fully configured 8-GPU HGX H100 server. Not only does this lower initial hardware costs to stand up data centers, but also reduces long-term power costs from lower power consumption.

Nebius also claims to offer industry-leading power-usage effectiveness of its servers, essentially on par with Azure and slightly ahead of Oracle Cloud. To understand why this is important, let’s break down PUE first: it is a ratio that compares total facility power to core IT load. For example, a 500 MW facility that can hold 400 MW of IT load (servers, cooling infra, etc.), would have a PUE of 1.25, while a 500 MW facility that can hold 450 would have a PUE of 1.11.

Lower PUE is important as it means data centers can deliver more compute per watt, with less power lost to overhead. Considering electricity is the largest operating expense for data centers, providers that can offer the lowest PUEs are more attractive as this translates into significant savings.

For example, assuming 400MW of core IT load operating in two data centers with PUE of 1.25 and 1.11, the first data center will have total power consumption of 500MW and the second 444MW. Assuming continuous operation, the lower PUE data center will consume 650,000 MWh less electricity annual, which translates to more than $90 million in annual savings at $0.14/kWh.

Source: NebiusNebius

Additionally, Nebius is prioritizing ownership of its data centers, which eliminates colocation fees, often up to 30% of total costs. Combining this with in-house designed servers and low PUE, Nebius can enjoy 20% to 25% cheaper operating costs per GPU, allowing them to offer competitive rental pricing. SemiAnalysis says that Nebius can offer “the lowest absolute price and the best terms for short to medium-term rents.”

Nebius currently offers H100 GPUs for as little as $2.00 per GPU hour with at least a three-month commitment ($2.95 per GPU hour without), a ~7% discount to the industry average at $2.15 per GPU hour, according to Silicon Data. This also compares to $2.49-$3.29 per GPU hour at Lambda and ~$6.16 per GPU hour for CoreWeave (based on $49.24 for an 8-GPU instance).

For the HGX H200, Nebius is offering rentals at $3.50 per GPU hour, a ~44% discount to CoreWeave at $6.31 per GPU, based on a $50.44 8-GPU instance price. For Blackwell GPUs, Nebius is renting HGX B200 chips at $5.50 per GPU hour, a slight premium to Lambda’s $4.99 for the B200 and a discount to CoreWeave’s $8.60 (based on a $68.80 price per 8-GPU instance).

However, it cannot be ignored that some of this pricing dynamic may stem from Nebius’ current scale, a fraction of CoreWeave’s size, and that it may cater more towards AI startups rather than larger enterprise customers who may require cluster sizes well beyond what Nebius can offer.

Nebius may also be in a tight position to demand premium pricing as bootstrapped AI startups may simply choose whatever cloud provider can offer GPU access at an affordable price, and offering competitive pricing or discounted rates can drive customer acquisition. Additionally, CoreWeave’s larger GPU fleet likely means it can demand premium pricing and long-term contracts with anchor tenants such as OpenAI, whereas Nebius is not yet at the scale to do the same beyond its Microsoft deal.

In the longer term, Nebius believes that its vertical integration with a full stack of AI services will help broaden its customer base, increase platform stickiness and capture higher margin revenue and services. Nebius offers a proprietary cloud platform with managed MLops services, low downtime and high cost efficiency, combined with its inferencing platform AI Studio. With AI Studio, Nebius says it can offer up to 3x token savings with low latency, and up to 4.5x faster time to token versus other competitors in Europe.

Nebius also reported its first MLPerf Benchmark results, training Meta’s Llama 3.1-405B model in 124.5 minutes with 1,024 Hopper GPUs, nearly double the speed of its 512 GPU result of 244.6 minutes. Nebius says that the result validated its platform’s ability to deliver bare-metal comparable performance while in the cloud.

Blackwell Ultras Drive 14% Increase in Annualized Run Rate Guidance, Not Including Microsoft Deal

In Q2, Nebius boosted its annualized run rate guidance by 14%, supported by its data center expansion, more power coming online in 2H and the rollout of more Blackwell and soon Blackwell Ultra GPUs. To note, the increased ARR guidance preceded the Microsoft deal, although it may be unlikely that ARR guidance will be raised much this year considering that capacity roll-out is more geared towards 2026.

Q2 ARR rose ~73% QoQ and 438% YoY to $439 million, marking a significant two-quarter expansion from $90 million in Q4. Nebius is targeting ARR to more than double over the next two quarters, raising its guidance to $900 million to $1.1 billion, up from its prior view for $750 million to $1 billion.

Nebius explained that the “increase in our ARR guidance reflects the strong demand we are seeing and the expected delivery of additional GPU capacity later this year, particularly the Blackwell Ultras. Because much of this capacity will come online by the end of the year, the impact will show up more in ARR than within the year’s revenue.” Much of the ARR growth is likely to be weighted towards Q4, considering the majority of GPU installations will take place that quarter, such as the ~4,000 Blackwell Ultras in the UK.

Nebius remains confident in reaching this ARR target, noting that a majority of the guidance is already under contract, while additionally capacity sells quickly upon coming online, adding an extra level of confidence.

Solidly in Multi-Year Hypergrowth Phase

Nebius is one of the more unique names in the AI universe as one of the few, if not only, AI-focused stock to have multiple years of triple-digit revenue growth ahead. Though the ARR guidance only extends two quarters ahead, it is a leading indicator of the upcoming hypergrowth phase Nebius is expected to see.

Nebius had already laid out expectations for a meaningful acceleration next year prior to the Microsoft deal, based on timing of capacity ramp in Q4 2025 and throughout 2026. The recent deal with Microsoft is extending this acceleration, with revenue revised more than $3 billion higher by 2029, or 24% to 65% above prior expectations.

Prior to the Microsoft deal, Nebius was expected to surpass $5 billion in annual revenue in 2029, doubling from $2.5 billion in 2027. After the deal, revenue estimates have been raised 24% to $3.1 billion in 2027, and 65% higher to $8.44 billion by 2029.

In terms of YoY growth, Nebius is now expected to see three consecutive years of triple digit growth from FY25 through FY27, with the possibility that additional hyperscaler deals and accelerated capacity expansion translate into a fourth consecutive year of triple-digit YoY revenue growth.

Revenue growth estimates for FY26 have moved 15 points higher to 166% YoY, while FY27 has now crossed into triple-digit territory at nearly 106% YoY. FY28 growth estimates have jumped more than 30 points to 85%, and, as noted previously, could potentially reach triple-digits if well supported by demand and capacity growth. Microsoft’s deal helps de-risk the growth story to some degree by materially upgrading backlog-like visibility to several billion in revenue through the duration of the deal.

For comparison, CoreWeave is expected to generate 128.4% YoY growth in 2026, though at a substantially larger scale with revenue estimated at $12.01 billion, before rising to $25.8 billion by 2029, or more than 3x the $8.44 billion estimate for Nebius.

Subsidiaries: Avride Progressing with Robotaxis with Key Partners Uber, Hyundai

Outside of its core AI infrastructure services, Nebius has two main subsidiaries: autonomous driving startup Avride and AI ed-tech firm TripleTen, as well as economic stakes in Toloka (now deconsolidated from results) and ClickHouse.

Avride

Avride is continuing to scale its autonomous delivery robot operations, working with Uber Eats in Jersey City, Dallas and Austin, and with Grubhub at the Ohio State University. Avride also is working with Japan’s Mitsui Fudson to deploy autonomous robots for warehouse-to-store logistics.

Avride is also advancing preparations for its autonomous ride-hailing service with Uber later this year in Dallas, based on a fleet of Hyundai Ioniq AVs with Avride handling software and systems integration. Avride said on September 4 it was beginning to ramp up testing to prepare for the Dallas rollout.

Avride is still pre-revenue and burning through cash, and due to the capital intensive nature of scaling autonomous delivery robots and AVs, Nebius is active discussions with potential strategic partners. It will be very challenging, if not impossible, to scale both AI infrastructure, which seems to be priority, with AVs simultaneously with only a few billion in cash. Consider that Waymo has raised more than $11 billion of outside capital, not including what Google has burned through.

TripleTen

TripleTen is a consumer-facing AI ed-tech platform primarily focused on offering coding or tech-skill based bootcamps. TripleTen delivered revenue of $28.8 million in 2024, up 251% YoY, while students enrolled rose 149% to 14,000.

Clickhouse

Nebius has a minority economic stake in open-source database platform ClickHouse, which recently raised $350 million in a Series C round in May. This valued the startup at $6.35 billion, more than tripling its $2 billion valuation from 2021.

Nebius has been straightforward in utilizing its 28% stake in Clickhouse as a means of raising capital, as it could unlock almost $1.8 billion (or more depending on exit valuation) without incurring dilution to shareholders or adding debt to the balance sheet.

Toloka

Toloka is a data provider for LLM and genAI developers, focusing on high-quality data solutions for training, fine-tuning, alignment, and more, counting Amazon, Anthropic, Microsoft and Shopify among its customers.

As of Q2, Nebius is deconsolidating Toloka from earnings results, as its voting share dropped below 50% following a $72 million investment from Bezos Expeditions and Shopify CTO Mikhail Parakhin. Toloka’s revenue rose 138% YoY to $26.4 million, and according to 2025’s guidance, was expected to at least double again to $50 to $70 million.

Financials

Revenue Rises 625% YoY in Q2

Before discussing the financials, it’s important to note that growth figures exclude Toloka’s impact following the deconsolidation in Q2.

Nebius reported a slight beat in Q2 with revenue of $105.1 million, up 625% YoY and 106% QoQ, versus estimates for $101.2 million. AI cloud infrastructure revenue rose more than 9x YoY in the quarter, fueled by strong demand for Hopper GPUs and almost peak GPU utilization.

Looking ahead, Q2 is marking an inflection point for revenue, with growth accelerating sequentially on a dollar basis. Q3 is estimated to see revenue up more than $52 million QoQ to $157.1 million, before rising another $106 million QoQ to $263.8 million in Q4.

For FY25, Nebius has maintained its guidance for $450 million to $630 million in revenue, excluding $50 million to $70 million in revenue attributed to Toloka. What’s interesting here is that management had explained that Nebius could grow faster, but it was “oversold on all of our supply of previous generation Hoppers, and we decided to wait for the new generation of GPUs to come. And finally, the new Blackwells are coming to the market in masses, and in parallel, we are dramatically increasing our data center capacity.” This is fueling strong growth through 2026 and beyond as these Blackwell GPUs come online.

Operating Margins Deeply Negative, Signs of Scale Emerging

While gross margin is expanding rapidly, up from 26% in Q4 to over 70% in Q2, operating margins remain deeply negative as due to the high costs of aggressively expanding data center capacity and GPUs on hand.

Excluding Toloka’s contribution, gross margin was 71.3% in Q2, up nearly 20 points from 51.5% in Q1 and improving nearly 25 points from 46.9% in the year ago quarter. While this signals strong execution to drive this degree of expansion, Nebius is now slightly below CoreWeave’s Q2 gross margin of 74.2%, and it remains to be seen how much further upside there is to gross margin as new capacity comes online.

On the infrastructure side, Nebius is pricing GPUs at a “healthy margin” on a per hour basis, while expectations for premium pricing for Blackwell GPUs may provide a tailwind through 2026. Nebius expects its Hopper GPUs to break even in two to three years on a gross profit level including both hardware and operational expenses, or even quicker when factoring in its higher-margin software and services from its full stack platform. Nebius has not commented on break-even periods for Blackwell GPUs, as it is still actively rolling out the new generation.

Moving down the line, operating margin remains deeply negative at more than (100%) of revenue as expenses continue to outpace revenue at current scale. Q2 operating margin was (105.8%), a notable improvement from (236.4%) in Q1 and (773.8%) in the year-ago quarter. Nebius is exhibiting initial signs of operating leverage, with total operating expenses up just 71% YoY versus the 625% YoY growth in revenue; this was driven mainly by 560% YoY growth in depreciation from increased server capacity, as SG&A decreased (10%) YoY.

GAAP net margin was 556%, as Nebius benefited from a one-time $597.4 million gain from revaluation of equity investments (Toloka’s deconsolidation). Adjusted net margin offers a clearer view of Nebius’ trajectory, coming in at (87.1%) in Q2, up from (164.4%) in Q1 and (424.8%) in the year ago quarter.

Quick Shift to Positive Adj EBITDA

Despite still investing heavily in capacity and its GPU fleet and kicking off its hypergrowth phase, Nebius is showing strong cost control as its core AI infrastructure business shifted to positive adjusted EBITDA in Q2. However, corporate adjusted EBITDA remains negative as Nebius continues to invest in Avride and TripleTen.

Q2 adjusted EBITDA was ($21 million), or a (20%) margin, improving from a (113.2%) margin in Q1. Nebius had laid out expectations in Q1 for adjusted EBITDA to shift to positive territory by 2H 2025, with Q3 the expected inflection point. Q2’s result does help cement this shift with higher probability now, but it will need to be proven.

For 2026, Nebius expects corporate adjusted EBITDA to be positive for the year, building on the AI cloud momentum as capacity and revenue ramp. Nebius has also offered some long-term targets for when its AI cloud business reaches scale (its medium term ‘several billion’ revenue size). At this scale, Nebius is targeting adjusted EBITDA margins between 20% to 30%, assuming a depreciation schedule of four years. Attaching this to FY29’s projected revenue of $8.44 billion, this would imply around $1.7 billion to $2.5 billion in adjusted EBITDA.

EPS Expected to Remain Negative as Nebius Scales

Nebius is expected to report negative EPS through all of fiscal 2025 and 2026, with no clear indication yet of when the company could or will shift to profitability. Current estimates show Nebius losing ($1.47) in 2025 and minimal improvement to ($1.34) in 2026.

There are no analyst estimates beyond 2026, but given the capital intensity of greenfield data center development, continuous scaling of capacity, not to mention potential investments into Avride, Nebius may face a long road to profitability.

Source: YChartsYCharts

Capex at $2B for 2025 Pressuring Cash Flows

Aggressive capacity expansion plans and a 33% increase in FY25 capex expectations from $1.5 billion to $2 billion, or nearly 4x expected revenue for the year, mean Nebius is quickly burning through cash.

Operating cash flow was ($167.7) million in Q2, for a (159.6%) margin, improving from a (357.7%) margin in Q1. However, this represented just a $30 million sequential improvement from ($197.8) million, suggesting that the path to positive cash flows may be prolonged.
Free cash flow was ($678.3) million in Q2 for a (645.4%) margin, versus ($741.8 million) for a (1341.4%) margin in Q1.
Capex was $510.6 million in Q2, or nearly 5x of revenue, down slightly from $544 million in Q1. 1H capex surpassed $1.05 billion.
Cash and equivalents totaled $1.68 billion, up only slightly from $1.45 billion in Q1 as Nebius deployed much of June’s $1 billion raise to capex and operations. Including the recent $2.75 billion raise and $1 billion share sale, cash is likely around or above $5 billion. At current burn rates, Nebius would have nearly 10 quarters of cash, but it is likely that capex will accelerate (think of CoreWeave as a leading indicator) to support growth in power and capacity.
Debt was $0.98 billion as of Q2, not including the recently priced $2.75 billion in notes.

Regarding capex, management stated that they “want to be opportunistic when it comes to really ramping up our infrastructure capacity as we see demand, and so we want to be able to sort of chase secure that demand well. And so we’ve considered some additional investments beyond the initial data center expansion plan.”

However, AI infrastructure is costly, and Nebius has already increased its guidance quite substantially for its size. The $2 billion guide also came prior to the Microsoft deal, which could add upwards force to capex and further pressure cash and cash flows, forcing Nebius into a vicious cycle of raise to build.

For example, CoreWeave is guiding to spend $20 billion to $23 billion in capex this year, also more than 4x its revenue and with potentially up to $15 billion hitting in Q4. For Nebius to keep pace with CoreWeave and scale at accelerated rates, capex needs are only going to move much higher, one that its balance sheet may not be able to sustain at the moment.

Earnings Q&A

Greenfield Site Selection

Given that Nebius is looking to acquire multiple greenfield sites for new capacity, analysts questioned why that was the route management is taking versus colocation or build-to-suit. While greenfield development provides full control over construction and design from the ground up, it can be a more expensive and longer time to power versus colocation using existing infrastructure.

Chief Product and Innovation Officer Andrey Korolenko answered why Nebius favors greenfields over build-to-suit:

“We typically favor greenfields because we can control every aspect of the data center from the design to construction to the hardware installations and deployment and phasing. We can actually tailor the phasing according to our demand. And for us, it's cheaper to build than build-to- suit, and we are not locked into the long-term leases. Also, by controlling the design of the building, starting from the — how power is piped into building and design and installation of our own racks and servers, we can achieve a lower total cost of ownership, probably around 20% less than the market average.”

Quickly Selling Through Capacity

Management also shed light on utilization trends, which support accelerated revenue growth moving forward as more capacity comes online. Higher utilization rates are important as it means GPUs are not sitting idle for long periods of time, generating revenue and shortening payback periods for high-capex server investments.

Chief Revenue Officer Marc Boroditsky explained that as Nebius “brought on more capacity, we sold through it. And by the end of the quarter, we were at peak utilization. There's a nice trend that we're actually starting to witness. As we bring on larger clusters, we are able to bring on new large customers who want to purchase greater and greater capacity. This allows us to expand and diversify our customer base and has been a clear signal there is growing opportunity in the market. This also suggests strong demand to support ramping up our capacity. If we had more capacity in the second quarter, we probably would have sold more as well.”

This ties in to comments about how growth could have been higher if Nebius was not GPU constrained, and signals a healthy demand environment for the moment as inference begins to take off. However, this is not a trend that can be assumed to last forever, given the competitive nature of the neocloud and hypercloud arena and the fact that Nebius’ growth is directly tied to its ability to deliver increasingly more capacity without delay.

Tariff Impacts:

Interestingly, Nebius discussed potential tariff impacts on Q2’s call, noting that it is still a bit early to form any definitive conclusions regarding any detrimental impacts. Chief Communications Officer Tom Blackwell said that “for now, it's a bit early to say anything definitive” about how tariffs could affect growth, though he stated that it is “possible we could potentially see some short-term fluctuations.”

Conclusion

Neoclouds present a high risk/high reward opportunity for investors, as mega-deals like Microsoft’s recent contract with Nebius highlight just how supply constrained the market is. On the other hand these firms have little access to organic cash and cash flows, and instead must turn to debt to fund aggressive capacity expansion.

The Microsoft deal and aggressive capacity expansion are fueling a prolonged period of hypergrowth for Nebius, with the company currently expected to see three, potentially four, consecutive years of triple-digit revenue growth. However, capex is outpacing revenue by a factor of 5x, and the recent debt and share sale may only provide a few quarters of runway before more cash is needed to fund this growth.

As stated in the intro, ultimately, we are stepping away from the stock. While it’s clear that Nebius could do well in an AI-driven market and the stock could extend rapidly, we want to be prudent about what will hold up should we see a softer AI narrative. We like other names in our portfolio better in terms of participating in the upside while protecting to the downside for when we do get that inevitable tech selloff. Primarily, the negative free cash flow, capital intensive operations and weak margins are among reasons we believe there are stronger names in our portfolio.

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.

Recommended Reading:

Supermicro Fails to impress; Volatility is here to staySupermicro Fails to impress; Volatility is here to stay
Bloom Energy: First Direct Hyperscaler with Oracle; More are Likely to FollowBloom Energy: First Direct Hyperscaler with Oracle; More are Likely to Follow
Positions Report – July 2025Positions Report – July 2025
AMD Reports in Line while AI Story to Improve from HereAMD Reports in Line while AI Story to Improve from Here

Oracle Stock Outlook: Revenue Could Double by FY2029, yet Targets Seem Lofty

Posted on April 4, 2025June 30, 2026 by io-fund

Late last year, Oracle outlined an ambitious plan to nearly double its revenue by fiscal 2029, forecasting sales of $104 billion that year. Oracle is aiming to capture growing enterprise AI spending in the cloud, separating itself from its hyperscale competitors with its ability to offer lower-cost AI compute via lower latency, multi-cloud AI flexibility, and AI vector search capabilities.

Additionally, Oracle is a key player in Stargate alongside OpenAI, SoftBank, and other partners, with its selection proving that, despite heightened competition from much larger cloud providers, Oracle Cloud has the capability to meet the large data, scale and performance demands for future AI workloads. Assuming high-single digit to low-double digit share of database and enterprise software spending, Stargate could represent tens of billions in revenue to Oracle with the project seeing a maximum investment of up to $500 billion over the next four years. The first phase is already underway with Oracle and OpenAI working to deploy 64,000 Nvidia GB200 GPUs at a large-scale facility in Texas.

However, Oracle is already on track to miss Cloud revenue targets for this fiscal year ending in May. Management also provided an optimistic forecast that would require Cloud revenue to nearly triple by FY29, and grow at almost a 30% CAGR, setting a high bar that includes execution risk with strong competition in AWS, Azure, and Google Cloud.

Unique AI Offerings Support Oracle’s AI Growth

Oracle’s ability to drive lower latency and high performance is one of the main reasons enterprises use Oracle for AI, as it allows enterprise customers to run demanding AI workloads faster and at a lower cost.

RDMA (Remote Direct Memory Access) is helping to drive Oracle’s AI story by enabling direct memory access between servers without utilizing CPUs, resulting in low-latency, high-bandwidth performance. Bypassing the CPU greatly accelerates data transfer rates, a necessity for large AI workloads requiring massive compute.

RDMA is integral to Oracle Cloud Infrastructure as the backbone of Oracle’s Gen2 Cloud and increasingly large Superclusters for AI training and inference, allowing ultrafast, near real-time performance. Oracle says that it can offer less than 10 microseconds of latency between nodes, improving efficiency.

With less overhead and fewer CPU cycles, RDMA helps Oracle offer its AI clusters at a lower cost: Oracle says it “consistently charges less than Amazon Web Services (AWS) for the equivalent compute capacity.”

Oracle offers the widest range of bare metal GPU instances among major cloud providers, and scalability at any size up to 65,536 Hopper GPU clusters and 131,072 B200 GPU clusters, which are expected to come online in 2025. Oracle also offers very flexible VM instances, letting customers pay for only the capacity they need as they need it for any size workload, rather than offering fixed instance sizes.

Oracle provides a wide range of AI clusters for small, medium and large-scale AI training. Source: OracleOracle

Oracle’s AI vector capabilities also stand out given Oracle’s database roots, offering native AI vector search capabilities with seamless integration to leading AI models from OpenAI, xAI, Meta, Cohere and more. AI vector search lets enterprises search both structured and unstructured data in a variety of manners, enabling intelligent, relevant and accurate AI responses utilizing their data. Oracle noted in Q3 that its Oracle Database 23ai can convert data into any vector format to be understood by an AI model of choice, facilitating AI training and inference on private data in Oracle Database.

Oracle Bets Big on Future Revenue Growth; Yet Lacks History of Meeting Targets

Oracle’s executives exuded confidence in meeting ambitious long-term revenue targets, underscored by AI and multi-cloud momentum — yet analysts aren’t so sure, with consensus estimates that are considerably lower than guidance offered during the recent earnings call.

In fiscal Q3 2025 (Oracle’s most recent quarter), management reaffirmed that confidence in reaching the $66 billion target for FY 2026 was now “stronger than ever,” while they also guided for 20% YoY growth in revenue in FY 2027, faster than previously expected. This would imply revenue of approximately $79.2 billion.

This builds upon management’s comments from Q2 that they believe they “now have a clear light of sight to our future revenue growth,” with total cloud infrastructure (IaaS) revenue in FY25 expected to grow faster than the 50% reported in FY24 and accelerate further in FY26.

Putting this in perspective, Oracle is betting on a major revenue growth acceleration driven primarily by growth in AI and cloud.

Reaching management’s optimistic $104 billion target in FY29 would require revenue growth at a 14.45% CAGR from FY24’s $52.96 billion. This is more than double the 6.03% CAGR that Oracle reported from FY19’s $39.51 billion in revenue through FY24, relying on strong, consistent AI and cloud growth to achieve these targets.

Here’s an annual breakdown of the preliminary growth guided through FY27 and what level of growth would be needed to achieve FY29’s forecast:

Oracle is targeting double-digit revenue growth from FY26 to FY29 to reach its $104 billion revenue target. Source: I/O Fund

Revenue growth would need to hit both the 15% growth and 20% growth targets in the next two fiscal years, and remain in the mid-to-high teens for FY28 and FY29 – such as 16.5% growth in FY28 and 13.3% growth in FY29.

Compare this to the prior comparable period from FY19 to FY24: Oracle reported only one year with revenue growth in the double-digit range, and just 6% in FY24 with less than 8% growth expected in FY25.

It would represent a rather unprecedented acceleration for Oracle and a monumental feat to drive a 12 point revenue acceleration in two years and maintain double-digit growth thereafter at a larger scale.

Analysts Remian Dubious of Oracle’s Growth Potential

Despite management’s confidence in achieving its growth targets in FY26 and FY27, analysts remain dubious over Oracle’s ability to hit these numbers successfully, considering the company has missed revenue estimates in both the last two quarters and in six out of the last seven.

Analysts are currently projecting Oracle to fall short of targets in FY26, FY27 and FY29, implying that they do not share the same level of confidence in future growth as management.

Analysts are projecting Oracle’s revenue to fall short of management’s forecasts for FY26, FY27 and the long-term FY29 target. Source: I/O Fund

For FY26, analysts estimate Oracle will report 14.2% growth to $65.2 billion in revenue, falling short of both the stated 15% growth and $66 billion revenue target.
For FY27, analysts estimate Oracle will report 18.2% growth to $77.0 billion in revenue, which would be nearly 2 points and $2 billion shy of what management is forecasting.
By FY29, analysts estimate Oracle will report 13.3% growth to $100.5 billion in revenue, or about 3.4% shy of management’s long-term forecast.

There has been little change in FY26’s revenue estimate over the past six months, being revised just 0.4% higher. However, FY27’s revenue estimate has been revised nearly 4.3% higher to that $77.0 billion, yet it still remains more than $2 billion below where Oracle’s 20% guidance implies.

Analyst estimates for Oracle’s FY26 and FY27 revenue have been quite consistent over the past six months. Source: YChartsYCharts

This consistent shortfall for revenue growth and doubts over Oracle’s ability to reach its stated targets likely stems from Oracle’s inability to reach its near-term targets, including its $25 billion cloud revenue target for FY25.

The I/O Fund specializes in covering lesser-known AI stocks on our research site with trade alerts and weekly webinars. Learn more here.The I/O Fund specializes in covering lesser-known AI stocks on our research site with trade alerts and weekly webinars. Learn more here.

Oracle Falls Short of $25B Cloud Revenue Target for FY25

In fiscal Q2 2025, Oracle CEO Safra Catz said that Oracle was expecting its Cloud revenue to reach $25 billion for the full fiscal year. As of Q3, Oracle is firmly on track to miss that target.

Oracle reported $6.2 billion in Cloud revenue in Q3, up 25% YoY, bringing its YTD Cloud revenue up to approximately $17.7 billion. For Q4, Oracle guided for 26% to 28% YoY growth for Cloud revenue, implying revenue between $6.68 billion to $6.78 billion, or $6.73 billion at midpoint.

This would place FY25 Cloud revenue at $24.38 billion to $24.48 billion, or $520 million to $620 million short of its goal. Not only is Oracle at risk of missing this target, but it also is running close to falling short of its 50% growth goal for Cloud Infrastructure revenue.

However, Oracle is planning on having Nvidia GB200-based servers generally available in April 2025, and if there is high demand for the new upgraded system, it could serve as a tailwind for IaaS growth in fiscal Q4 as these were not available in Q3.

Oracle Aims for Cloud Infrastructure Growth Above 50%

Oracle’s Cloud Infrastructure (IaaS) business has been a strong driver for Cloud growth, outpacing Cloud revenue growth by at least 22 points for nine consecutive quarters, rapidly expanding its share of Cloud revenue to 44% from just 29% two years ago.

Cloud IaaS revenue also surpassed a $10 billion run rate in Q3, while growing at >50% the last two quarters. Cloud IaaS is an important element in Oracle’s $25 billion target, in that at least 40% of said revenue will be coming from Cloud IaaS.

Oracle’s Cloud IaaS revenue has grown consistently over the past 2.5 years to surpass a $10 billion run rate in Q3. Source: I/O Fund

IaaS’ performance has been driven by the strength of Oracle’s AI offerings. Growth reaccelerated 6 points sequentially in Q2 to 52% YoY, driven by AI, though it dipped slightly to 51% YoY in Q3.

In Q3, Oracle noted that demand was at “record levels”, that GPU consumption for AI training had increased 244% in the last 12 months, and that the company was seeing “enormous demand” for AI inference. Chairman Larry Ellison added that “AI training and multi-cloud database are experiencing hyper growth.” This follows Q2’s 336% increase in GPU consumption for AI training and “record level AI demand.”

Despite these triple-digit points and strong growth commentary, Cloud IaaS revenue’s growth has been quite small sequentially. Q3 saw the highest sequential growth at ~$300 million QoQ, or ~12.5%.

To meet management’s “faster than 50%” growth target for FY25, Q4 would need to match that QoQ revenue increase of at least $300 million; this would place IaaS revenue at $10.3 billion, or up 51.5% YoY from $6.8 billion in FY24. Should the segment return to recent historical trends of rising $200 million sequentially, Oracle would be running extremely close to reporting growth below 50% for the full year.

While demand remains strong and “continues to outstrip supply”, as management puts it, component delays are likely a factor in seeing limited QoQ growth in FY25 despite these extraordinary AI growth figures, as these delays are not expected to ease until Q1 FY26 (likely due to Nvidia’s Blackwell system timing).

Oracle has stated that component delays have “slowed cloud capacity expansion this year,” and expects these to resolve in FY26, easing a primary bottleneck to IaaS growth. The I/O Fund discussed how Nvidia suppliers were foreshadowing delays in the most recent earnings reports. With IaaS expected to accelerate further from >50% in FY25, the segment could grow to close to $16 billion, or at least $5.5 billion higher than FY25’s projected ~$10.3 billion.

Should IaaS account for ~50% of Cloud revenue for FY26, that would imply Cloud revenue of $32 billion, or an 8 point acceleration to 31% YoY growth from FY25’s Q4-guide-implied $24.4 billion. This is likely what Oracle would need to achieve its $66 billion target, assuming this growth means Cloud grows its revenue share by mid-single digits YoY.

63% Growth in Remaining Performance Obligations

While challenges remain for FY25 with Oracle at risk of missing its $25 billion Cloud revenue target, 2026 is more optimistic, with IaaS promising to add multiple billions in revenue and RPO suggesting the segment could see a strong acceleration.

RPO rose 63% YoY to $130 billion in Q3, compared to 50% YoY growth to $97.3 billion in Q3, a nearly $33 billion QoQ increase. This was driven by record deal activity with $48 billion in signed contracts. Chairman Larry Ellison chalked this up to Oracle’s ability to provide a more cost-effective AI solution to customers: “it really is a technology advantage we have over them. If you run faster and you pay by the hour, you cost less. So that technology advantage translates to an economic advantage, which allows us to win a lot of these huge deals.”

Additionally, Cloud RPO accelerated 10 points to 90% YoY, representing 80% of total RPO, up from 75% last quarter. This places Cloud RPO at around $104 billion, up from nearly $73 billion last quarter. Management said that while the growth in RPO is evidence that the AI training business continues to grow, AI inferencing and database demand also factored into the RPO growth.

RPO continuing to outpace Cloud and IaaS growth bodes well for Oracle’s forecasted revenue acceleration in FY26, and signals strong underlying demand trends with Cloud RPO nearly doubling. 31% of total RPO is expected to be recognized over the next twelve months, or ~$40.3 billion, up 7% from $37.6 billion at the start of FY25.

Oracle noted that this RPO figure did not include any contributions from Stargate yet, which promises a large potential opportunity given the initial commitment of $100 billion up to $500 billion total for the project. Management expects the “first large Stargate contract fairly soon,” though it’s hard to put an exact figure on what management defines as large.

Oracle Continues to Quickly Add Capacity

Oracle is quickly adding data center capacity due to this growth in RPO, with management commenting that they expect to double data center capacity this year to meet the high demand they see from this RPO and additional opportunities from Stargate. Capex is also on track to exit the year at a meaningfully higher rate than expected as Oracle accelerates build-outs.

Oracle noted in Q3 that they “expect fiscal year 2025 CapEx will be a little more than double what it was last year at around $16 billion.” A year ago in Q3 2024, management laid out preliminary capex of $10 billion, which at the time was expected to go towards 100 new data centers and expansions at 66 existing facilities.

This is quickly flowing to the data center, with CEO Safra Catz explaining that Oracle expects its “available power capacity will double this calendar year and triple by the end of next fiscal year” (by mid-2027). Catz added that Oracle’s “live data center count and power capacity is the leading indicator of the conversion of RPO to revenue,” suggesting that Oracle is working to significantly increase capex to quickly translate this surge in RPO to reported revenue, to meet growth acceleration targets.

To note, this rapid acceleration in capex is negatively impacting cash flows. Oracle generated $11.8 billion in free cash flow in fiscal 2024, but reported negative free cash flow of ($2.66 billion) in Q4 and just $70 million in Q3. On a TTM basis, free cash flow was just $5.82 billion as of Q3, down (53%) YoY, as accelerated capex is far outweighing limited growth in operating cash flow.

Oracle Cloud at Smaller Scale than Peers

Oracle seemed to have taken a slight dig at Microsoft, Amazon and Google in Q3, its hyperscale partners and competitors, stating that Oracle Cloud IaaS revenue grew at “a much higher growth rate than any of our hyperscaler competitors.”

However, this growth is coming off a much smaller scale, considering that Oracle is using IaaS revenue as the comparison – to recap, that’s at a $10.6 billion annualized rate, or ~$2.65 billion in Q3.

Google Cloud, the smallest of the hyperscalers’ clouds, reported 30% YoY growth in revenue to $12.0 billion in Q4. For the full year, Google Cloud revenue grew nearly 31% YoY to $43.2 billion.

Microsoft’s Intelligent Cloud revenue increased 19% to $25.5 billion in its fiscal Q2, with Azure revenue growing 31% YoY. Microsoft’s AI run rate reached $13 billion, up 175% YoY, or 30% higher than Oracle’s entire IaaS segment and half of its entire Cloud segment.

AWS reported 19% growth to $28.8 billion in revenue in Q4, and 19% growth to $107.6 billion for 2024. Put this way, AWS is generating more revenue in one quarter than Oracle’s Cloud is in one year. AWS also generated more than 1.6x Oracle’s Cloud revenue in operating income.

Oracle is by no means operating at the same level as its customers, with Microsoft’s AI run rate dwarfing Oracle’s IaaS business while growing well into the triple-digits. Cloud to Cloud, Oracle is lagging Google’s growth, the smallest of the three hyperscalers, by 5 points in the most recent quarter while at half the scale.

Reaching FY29’s Target Likely Requires Prolonged Cloud Acceleration

Outside of Cloud, Oracle’s other businesses – licensing, hardware, services – have seen minimal growth over the past few years, generating revenue of $31.5 billion in FY22 and $33.2 billion in FY24. Assuming growth at a 3% CAGR through FY29, ex-Cloud revenue would project to approximately $37 billion, meaning Oracle’s Cloud segment would need to reach $67 billion by FY29 to reach that $104 billion forecast.

This would equate to a nearly 29% CAGR from ~$24.4 billion in FY25 to reach $67 billion, or on an annual basis, 31% growth in FY26 and FY27, followed by 26% growth in both FY28 and FY29.

Here’s what this would look like:

Oracle’s Cloud segment would need to grow at a 29% CAGR from FY25 to hit $67 billion, a likely threshold for it to reach the $104 billion target. Source: I/O Fund

Conclusion

Projecting 30% growth for multiple years does not account for execution risk as AI is still quite early. While there are large competitors such as AWS, Azure and Google Cloud, there are also neocloud competitors, such as CoreWeave, that are shaking up the cloud IaaS market with GPU-as-a-service. These neoclouds are entirely optimized for AI, able to capture a higher rate FLOPs utilization (MFUs) — a metric that is quite important when considering time to market for AI models. My firm recently covered CoreWeave’s IPO for our premium members here.

Stargate will certainly provide a tailwind to Cloud revenue through FY29 should the maximum of $500 billion in investments materialize. Prior estimates from IDC place database and enterprise software spending at a high-single digit share of overall infrastructure spending. This could ultimately represent tens of billions in long-term revenue to Oracle that will complement existing growth in Cloud.

However, analysts do not share the same confidence in Oracle’s ability to meet its long-term targets, with consistent revenue misses in six of the past seven quarters; additionally, it is already falling behind its $25 billion Cloud revenue target for FY25 after just one quarter.

The I/O Fund is ultimately passing on Oracle and we are instead accumulating small and mid-cap positions that are better poised to benefit from the ongoing AI spending war. Premium and Advanced members receive real-time trade alerts and technical setups in our weekly webinars. Learn more here.

Disclaimer: 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.

Recommended Reading: