CoreWeave vs Nebius—Which "AI Cloud" Power is Stronger?

CoreWeave vs Nebius—Which "AI Cloud" Power is Stronger?

Author: Xu Chao

Source: Hard AI

Morgan Stanley released an in-depth report, systematically breaking down for the first time the strategic similarities and differences between the two listed Neocloud companies—CoreWeave and Nebius.

According to Morgan Stanley analysts, CoreWeave has established industry leadership through its scale advantage, solid execution record, and highly visible financing pathway; Nebius, on the other hand, builds differentiated optionality value through greater spot market flexibility, a diversified strategy focused on enterprise customers, and a more flexible capital structure.

The gap in financial data is obvious. In 2025, CoreWeave will achieve revenues of $5.13 billion (168% year-on-year growth), with EBITDA reaching $3.1 billion and a profit margin of 60%. Nebius’ revenue is only $530 million (up 479% year-on-year), and EBITDA shows a loss of $65 million (margin -12%). Yet, the growth itself is impressive—Morgan Stanley forecasts CoreWeave’s revenue to reach about $12.5 billion in 2026, while Nebius may leap to about $3.3 billion, with both companies on a super-fast growth track.

The most crucial point for investors is: More than 98% of CoreWeave’s business is tied to multi-year “use or pay” long-term contracts, providing strong revenue visibility; yet Nebius’s larger spot market exposure may offer greater short-term flexibility in the current environment of continuing strong GPU pricing. Nebius’s approximately $550 million ARR outside the Microsoft and Meta contracts mainly comes from on-demand/spot markets and shorter-term contracts, giving more room for repricing in today’s high-demand, supply-constrained climate. Strong demand, supply bottlenecks, memory price inflation, and surging token consumption driven by autonomous agents (like OpenClaw, etc.) all support continued positive GPU pricing trends.

Spot market exposure: Nebius dominates, may benefit from GPU price hikes short-term

One of investors’ biggest current concerns: Who stands to gain most from GPU price hikes? The answer points to Nebius.

About $550 million of Nebius’s ARR (excluding Microsoft and Meta contracts) comes mainly from on-demand/spot markets and customers with shorter-term contracts (mostly estimated under 1 year). These contracts have upward repricing potential given strong demand and tight supply. In contrast, over 98% of CoreWeave’s business is locked in multi-year "pay-as-you-go" contracts, securing revenue visibility but sacrificing spot market flexibility.

Morgan Stanley notes that memory price increases drive capex inflation, deployment challenges, and the continued increase in computing consumption driven by autonomous agents like OpenClaw, possibly keeping GPU pricing strong in the short term. If GPU prices continue climbing, Nebius is likely to keep outperforming.

Data Center Strategy: Neck-and-neck, but structurally different

Both CoreWeave and Nebius use a "hybrid" data center strategy, combining co-location leasing, build-to-suit, and self-build ownership, but their current infrastructure structures differ significantly.

CoreWeave currently operates 43 data centers, mainly via co-location leasing, and is actively introducing self-build and joint-venture models, such as the 300MW self-built Lancaster, Pennsylvania project (initial 100MW, scalable to 300MW) and the 250MW joint venture in Kenilworth, New Jersey (CoreWeave holds 15%, expected to launch in early 2027).

Nebius has a 75MW self-built data center running in Mäntsälä, Finland, and is planning a 300MW project in Birmingham, Alabama (to start Q2 2026) and a super-sized 1200MW campus in Independence, Missouri (Phase 1: 250MW slated for October 2027 completion, full completion expected by end of 2029). Nebius regards self-building as the best long-term economic solution, as it provides better cost and operational control.

Both see co-location leasing as key for fast scaling, and plan to increase self-build ratios in the future for improved economic efficiency and operational control.

Power scale: CoreWeave leads by a wide margin, Nebius quickly catching up

Power is the core resource in the AI cloud compute competition.

By end-2025, CoreWeave will have signed power capacity of 3.1GW, with active power over 850MW; Nebius has signed 2GW with only 170MW active—CoreWeave’s active power is 5 times Nebius’.

Looking at mid-to-long-term targets, CoreWeave plans to reach over 8GW signed power by 2030, while Nebius aims for 3GW by end-2026 and 5GW by 2030.

Morgan Stanley expects CoreWeave to add about 1GW of active power annually, and Nebius about 600MW. The gap largely reflects differing historical execution—CoreWeave has already proven large-scale delivery capability, while Nebius’s story remains at a relatively early stage.

Capital structure & financing: CoreWeave’s path is clear but debt-heavy; Nebius is flexible but uncertain

To support massive capex, both companies need to continue large-scale financing, but their modes differ significantly.

CoreWeave uses delayed draw term loan (DDTL) as its core financing tool, having developed a mature asset-level debt financing system.

By end-2025, CoreWeave’s total debt principal will reach $21.6 billion (45% of which is DDTL). DDTL interest rates have steadily declined: DDTL 1.0 (July 2023, $2.3 billion) at 15%; DDTL 4.0 (March 2026, $8.5 billion) at a fixed ~5.9%. Morgan Stanley predicts CoreWeave's debt scale will grow from $21 billion in 2025 to $46 billion in 2026, and further to $92 billion by 2028.

Nebius started with a cleaner balance sheet, relying mainly on convertible bonds, equity issuance (ATM plan, 25 million shares), and customer prepayment financing. The recent $4.338 billion convertible bond issue (with $2.588 billion due 2031 and $1.75 billion due 2033) has further boosted liquidity.

For 2026 capex plans, CoreWeave's guidance is $30-35 billion, Nebius’s is $16-20 billion; both have received $2 billion in strategic investment from NVIDIA.

Customer concentration & market strategies: scale first vs enterprise diversification

CoreWeave’s strategic core: “Get to maximum scale at the fastest speed.”

Currently, about two-thirds of its revenue comes from Microsoft, with projections that two-thirds of its ARR by end-2026 will come from Microsoft, OpenAI, and Meta combined. This highly concentrated customer structure delivers strong revenue visibility but brings concentration risk.

Nebius’s strategy is “sell AI cloud to a broader set of enterprise customers and AI startups.” Nebius’s current backlog is also highly concentrated, but its anchor contracts with Microsoft and Meta are very high quality.

Specifically, $15 billion in Meta’s latest contract is “backstop commitment for residual capacity,” providing Nebius “firepower” to sell GPU compute to SME clients and AI startups with lower credit ratings (CoreWeave and NVIDIA have a similar $6 billion backstop agreement).

Nebius aims to achieve $4-4.5 billion in Microsoft and Meta contract ARR by end-2026, moving progressively toward a broader customer ARR goal of $7-9 billion.

Software stack: CoreWeave more mature, Nebius pursues differentiation

Both companies aim to position themselves as “full-stack AI infrastructure platforms,” not just GPU lessors.

CoreWeave’s software stack is more mature, especially in infrastructure orchestration, observability, workload management, and storage. Its Object Storage business ARR has exceeded $100 million (as of Q3 2025), and via NVIDIA partnership is pushing software directly to global cloud service providers and enterprise customers.

Nebius is more focused on building tightly integrated secure AI cloud environments, emphasizing virtualization, tenant isolation, workload orchestration, and inference toolchains. The target is a vertically integrated AI cloud platform for enterprise customers.

Overall, CoreWeave leads in software maturity and platform breadth, while Nebius shows long-term potential in differentiated products for the application layer and inference scenarios.

Profitability: Distinct starting points, converging endpoints?

CoreWeave already demonstrates strong EBITDA profitability (2025 EBITDA margin at 60.8%). However, due to a timing mismatch between rapid capital deployment and revenue recognition, adjusted EBIT margin is facing temporary compression (2026 projected at 8%, rebounding to 15.9% in 2027, and 22.4% in 2028).

Nebius is currently incurring heavy losses (2025 adjusted EBIT margin at -96.8%), mainly due to: inheriting a large employee base and operational cost structure from Yandex, loss-making assets (like AVride), and depreciation/interest pressures during significant expansion.

Nevertheless, Morgan Stanley predicts Nebius’s EBITDA margin will quickly improve from -12.2% in 2025 to 40.8% in 2026, 62.1% in 2027, and 70.6% in 2028. If Nebius can realize the TCO advantages of self-built centers and shift to higher-margin enterprise customers, its long-term profitability potential warrants attention.

Both face valuation challenges

Morgan Stanley gives both a “neutral” rating, with target prices below current stock prices, reflecting that current valuations have already fully priced in optimistic expectations.

CoreWeave’s key risk is its $46 billion (projected 2026) debt scale and highly concentrated customer structure; Nebius’s core risks are its aggressive short-term revenue targets, unproven profitability, and significant uncertainties in transitioning to a broader enterprise customer base.

Both companies share potential risks including: normalization of AI demand, internalization of AI compute by large clients, and shortened effective GPU life span. For investors seeking scale and certainty, CoreWeave’s execution and financial visibility is more appealing; for those chasing growth flexibility and diversified optionality, Nebius’s spot market exposure and enterprise cloud strategy offer differentiated logic.

This article is from WeChat Public Account “Hard AI”. For more cutting-edge AI information, click here.

Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it consider the individual user's specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their own specific circumstances. Invest accordingly at your own risk.