Executive Summary / Key Takeaways
-
The Capacity Arbitrage Thesis: Nebius is exploiting a historic supply-demand imbalance in AI compute by deploying $21 billion in capacity through 2026 while competitors remain constrained, positioning it to capture outsized economics as the market matures from scarcity to equilibrium.
-
Full-Stack Differentiation as Margin Engine: Unlike commodity GPU renters, Nebius's vertically integrated model—spanning data center construction, hardware orchestration, and a 100%-attach-rate software stack—creates structural cost advantages and customer stickiness that should drive EBITDA margins from 24% today toward a 40% target by 2026.
-
Financing the Hypergrowth: The March 2026 $4.3 billion convertible note offering at sub-2% coupons, combined with $3 billion in cash and $4.6-5.2 billion in monetizable non-core assets, provides a self-funded growth model that minimizes dilution while funding an unprecedented $16-20 billion CapEx program for 2026.
-
Execution at Scale as Critical Variable: With $20.5 billion in total estimated capital needs and guidance implying 6x ARR growth to $7-9 billion by end of 2026, the investment thesis depends on the execution of data center construction timelines and GPU deployment schedules—any slippage creates valuation compression risk from current multiples.
Setting the Scene: The AI Infrastructure Land Grab
Nebius Group N.V., founded in 1989 and headquartered in Amsterdam, spent decades building technical DNA as Yandex N.V. before executing a decisive pivot in August 2024. The rebrand represented a full strategic exit from legacy businesses to focus exclusively on full-stack AI infrastructure. This gives Nebius a rare combination: the engineering depth of a mature technology company combined with the clean slate of a pure-play AI provider, unencumbered by legacy revenue streams that distract traditional cloud incumbents.
The company operates in a high-demand environment. Global data center AI power demand is set to quadruple to 1,500 terawatt-hours by 2034, while the AI Infrastructure-as-a-Service market races toward $419 billion by 2028. Big Tech capital spending has already been revised up 31% to $405 billion for 2025, yet supply remains the primary constraint. Management's observation that "everything we deploy, we sell" reflects a structural industry condition where demand for AI compute so far outstrips supply that pricing power remains firmly in the hands of providers who can deliver capacity.
Nebius's position in the value chain is deliberately integrated. Unlike CoreWeave or Lambda Labs that lease data center space and focus on GPU orchestration, Nebius controls the entire stack: it secures land and power, builds greenfield data centers, and deploys GPUs only when demand is visible. This vertical integration creates a 20% total cost of ownership advantage versus build-to-suit alternatives, a structural edge that becomes more valuable as pricing normalizes. The strategy is to use the current scarcity window to lock in long-term capacity at advantaged economics, then capture margin as the market matures.
History with a Purpose: From Yandex to AI Pure-Play
The Yandex lineage provides Nebius with a deep bench of systems engineers who have spent decades solving hyperscale infrastructure problems at one of Europe's largest technology companies. When Arkady Volozh notes that Nebius is "one of very few AI cloud providers that actually can serve corporate clients in multiple geographies," he's referencing this inherited expertise in building resilient, multi-tenant platforms that handle extreme scale.
The August 2024 rebrand marked the point where management decided to monetize non-core assets to fund AI infrastructure growth. The Q2 2025 deconsolidation of Toloka, while retaining a majority economic interest, exemplifies this discipline. By allowing Toloka to raise capital from Bezos Expeditions independently, Nebius freed management attention and created a valuable stake that can be monetized later without diluting shareholders. This demonstrates capital allocation discipline at a time when many neocloud competitors are diluting shareholders to fund growth.
Technology, Products, and Strategic Differentiation
Nebius's full-stack architecture is its primary moat. The software stack spans three layers: hardware management, a virtualized cloud platform, and an application layer with pre-configured AI tools. This structural advantage creates a 100% software attach rate across all AI cloud customers. The significance lies in the transformation of Nebius from a commodity compute provider into a platform with switching costs. Customers aren't just renting GPUs; they're building on Nebius's APIs, using its inference platform (Token Factory), and integrating its agentic search capabilities (Tavily). This stickiness enables premium pricing that management describes as strong across all GPU families, even for previous-generation hardware.
The Aether platform launch in Q3 2025 represents a deliberate expansion upmarket. By adding enterprise-grade compliance, security certifications, and control features, Nebius is targeting large enterprises that require trust and auditability for critical AI workloads. This expands the addressable market beyond AI-native startups to include regulated industries with deeper pockets and longer contract durations. The 50% increase in average contract duration for new cloud customers in 2025 validates this strategy—enterprises are committing to multi-year deals that provide revenue visibility and reduce churn risk.
Tavily's acquisition adds nearly 700,000 developers and agentic search capabilities, creating a developer ecosystem effect. Roman Chernin's comment that Tavily "very much fits in our strategy to become the platform where all the AI developers are building their applications" reveals the long-term vision: Nebius wants to be the default infrastructure for the next generation of AI applications, much like Amazon (AMZN) Web Services became the default for web applications. Software monetization is still early, but the 100% attach rate ensures that as customers scale, Nebius captures more value per GPU-hour.
Financial Performance: Evidence of a Working Model
The Q4 2025 results provide evidence that the strategy is working. Group revenue of $228 million grew 547% year-over-year, but the core AI cloud business's 830% growth shows the pivot is gaining traction far faster than legacy businesses are declining. The ARR trajectory is also significant: from $249 million in March 2025 to $1.2 billion by December, a nearly 5x increase in nine months. This demonstrates that Nebius is winning customers who are scaling rapidly and committing to long-term usage.
Margin expansion validates the full-stack thesis. The core AI cloud business achieved positive adjusted EBITDA in Q2 2025, ahead of schedule, and expanded margins from 19% in Q3 to 24% in Q4. Management's 2026 guidance of approximately 40% group adjusted EBITDA margin implies continued operating leverage as revenue scales faster than fixed costs. The gross margin of 68.63% provides cushion to absorb the operating expenses required to build out global operations.
The cash flow story is nuanced. Q4 2025 operating cash flow of $834 million was driven by upfront payments on long-term agreements, a working capital benefit that reflects customer commitment. However, annual free cash flow of -$3.68 billion reveals the capital intensity of the model. This frames the financing challenge: Nebius must continuously raise capital to fund growth, but the prepayment model and non-core asset monetization provide non-dilutive sources of funding that many peers lack.
Capital Allocation: The $21 Billion Bet
Nebius's 2026 CapEx guidance of $16-20 billion represents one of the most aggressive infrastructure buildouts in the sector. Management's strategy is methodical: secure land and power first, build data centers second, then deploy GPUs only when demand is visible. Ophir Nadav's comment that "we deploy the GPUs in a short time period when we have great visibility about the demand" reveals a risk management approach that treats the largest capital component as variable rather than fixed.
The financing plan is equally deliberate. Approximately 60% of 2026 CapEx will be self-funded through cash on hand, operating cash flow, and prepayments from long-term agreements. The remaining 40% will come from the $4.3 billion convertible notes raised in March 2026 at a blended sub-2% coupon, plus potential asset-backed financing. This minimizes dilution while securing capital at attractive rates. The convertible notes have conversion premiums that likely protect shareholders from immediate dilution, and the lack of existing corporate debt gives Nebius flexibility.
Non-core assets provide a hidden buffer. The ClickHouse stake alone, valued at over 25% of a $15 billion company, represents roughly $3.75 billion of latent value. Avride's comparison to Waymo's (GOOGL) $40-50 billion valuation suggests additional upside. Management explicitly frames these stakes as potential sources of capital to support hyper-growth plans beyond 2026 while maintaining a disciplined balance sheet. This provides a non-dilutive funding option that many competitors lack, potentially allowing Nebius to maintain ownership while rivals dilute shareholders.
Outlook and Execution Risk: The $7-9 Billion ARR Bridge
Management's 2026 guidance implies a 6x increase in ARR to $7-9 billion, which would place Nebius among the largest AI infrastructure providers globally. The key assumption is that capacity deployment will accelerate throughout 2026, with the majority of 800MW-1GW coming online in the second half. Dado Alonso's clarification that "revenues and ARR reflect the deployment schedule of our capacity throughout the year" is crucial—it means Q1 and Q2 2026 may show modest sequential growth, with acceleration backloaded to Q3 and Q4.
The Microsoft (MSFT) and Meta (META) deals provide revenue visibility but different timing. Meta's $3 billion contract will contribute roughly 12 months of revenue for the first tranche and 11 months for the second in 2026, suggesting approximately $600 million in recognized revenue. Microsoft's $17.4-19.4 billion deal won't reach full run rate until 2027, meaning 2026 revenue will be weighted toward the second half. This matters because it creates a potential "air pocket" in early 2026 where capacity deployment delays could cause revenue to miss expectations, triggering multiple compression.
Marc Boroditsky's statement that the 2026 ARR target "is not dependent on any new mega deals" provides confidence—more than half of the $7-9 billion is already booked through existing contracts. However, the dependency on large hyperscalers introduces customer concentration risk. If Microsoft or Meta shift strategy or experience their own demand slowdown, Nebius's growth trajectory could be affected.
Risks: What Can Break the Thesis
The convertible notes, while cheap, carry delayed dilution risk. With $4.3 billion in notes outstanding and potential for additional raises to fund the full $16-20 billion CapEx program, shareholders face ownership dilution over the next 5-7 years. The sub-2% coupon is attractive, but conversion will likely occur if the stock performs well, capping upside for existing holders.
Execution risk on the CapEx program is significant. Andrey Korolenko's comment that "building data centers is a complex task with inherent risks" highlights the challenge of deploying 800MW-1GW in a single year. Supply chain constraints for data center equipment, power grid connection delays, and construction bottlenecks could push capacity online later than planned, creating a revenue shortfall.
The AI market cycle risk is present. Arkady Volozh acknowledges that "this situation of unbalanced demand supply is temporary," implying that today's pricing premiums will normalize as new capacity comes online across the industry. If Nebius is still deploying its $21 billion capacity build when demand-supply equilibrium returns, it could face margin compression and asset utilization challenges.
Tariffs pose a margin threat. U.S.-China trade tensions could affect GPU procurement costs or data center equipment imports. Given the 68.63% gross margin, even a 5-10% cost increase on hardware could pressure profitability, especially if pricing power weakens simultaneously.
Competitive Context: David vs. Goliath with a Slingshot
Nebius competes against hyperscalers with trillion-dollar market caps and decades-long head starts. Microsoft's Azure grew 39% year-over-year in Q2 FY2026, Amazon's AWS grew 24%, and Google Cloud grew 34%—all from bases significantly larger than Nebius. These giants benefit from ecosystem lock-in, enterprise relationships, and massive balance sheets.
Yet Nebius's specialization creates openings. Its full-stack control allows it to be hands-on practitioners along with customers, a level of service that hyperscalers may not match at scale. The 20% TCO advantage from greenfield development is relevant as enterprises scrutinize AI spending. And the 100% software attach rate creates switching costs that pure infrastructure providers lack.
The neocloud competitive set includes CoreWeave, which has also secured major deals but trades at different risk premiums due to financing structures. Nebius's self-funded model and strategic asset monetization provide a capital structure advantage. However, CoreWeave's first-mover status in certain GPU categories and its partnership with OpenAI create competition for advanced workloads.
NVIDIA (NVDA) sits in a unique position as both supplier and partner. While Nebius depends on NVIDIA for GPUs, its status as a launch partner for NVIDIA Dynamo and one of five reference platform partners provides preferential access. In a supply-constrained market, allocation priority is a significant advantage.
Valuation Context: Pricing in Perfection
At $105.97 per share, Nebius trades at a $26.81 billion market cap and $28.02 billion enterprise value. With TTM revenue of $529.8 million, the stock fetches 50.6x trailing sales—a multiple that reflects expectations for the 2026 guidance. The forward multiple is lower: 2026 revenue guidance of $3-3.4 billion implies 7.9-8.8x forward sales, which is comparable to Microsoft's 8.75x.
The valuation reflects growth (547% in Q4) and margin trajectory (24% core EBITDA margins expanding toward 40%). However, the operating margin and ROA reveal that Nebius is still in heavy investment mode. Investors should focus on price-to-operating-cash-flow, which at 28.4x based on Q4 annualized OCF of $834 million, appears more aligned with growth expectations if they are sustained.
The balance sheet provides both strength and risk. The current ratio of 3.08 and quick ratio of 2.96 show liquidity, while debt-to-equity of 1.06 is manageable given the asset base. However, the $4.3 billion in convertible notes creates a potential overhang if the stock trades below conversion prices, and the $16-20 billion CapEx plan will require additional financing beyond current resources.
Conclusion: The Capacity Arbitrage Bet
Nebius represents a pure-play bet on AI infrastructure scarcity and the company's ability to vertically integrate its way to a durable competitive advantage. The $22 billion capacity buildout through 2026, supported by $20.5 billion in estimated capital raises, positions the company to capture outsized economics as the AI market transitions from experimental to essential. The full-stack differentiation—evidenced by 100% software attach rates, 20% cost advantages, and enterprise-ready platforms—provides a path to 40% EBITDA margins.
The investment thesis hinges on execution velocity and financing discipline. If Nebius deploys 800MW-1GW of capacity on schedule in 2026 while converting its $4 billion pipeline, the $7-9 billion ARR target is achievable. If construction delays, supply chain issues, or demand normalization occur, the high fixed-cost base and valuation multiple create downside risk.
For investors, the key monitoring points are Q1 2026 pipeline conversion, quarterly capacity deployment metrics, and software monetization progress. The company's ability to fund 60% of its CapEx internally while maintaining a 100% software attach rate suggests a business model that can scale. However, the convertible note overhang and potential for additional financing remain material risks that could affect returns even if operational execution is successful.
Nebius is an investment in an AI infrastructure market that remains supply-constrained and growing at 45%+ annually. The question is whether Nebius can build fast enough to capture the opportunity before the window of scarcity closes.
