Better AI Data Center Stock: Nebius vs. CoreWeave

CoreWeave's contracted revenue backlog rose 246% year over year to $104 billion, while Nebius' backlog more than quadrupled to over $40 billion, underscoring strong demand for AI cloud infrastructure. Analysts project 2025-2028 revenue CAGRs of 101% for CoreWeave and 253% for Nebius, while CoreWeave trades at roughly 7x current-year adjusted EBITDA versus Nebius at 15x next-year EBITDA. The article favors CoreWeave due to its larger 51-data-center footprint, simpler business model, and lower valuation, though leverage, equity issuance, high rates, and a potential AI-spending slowdown remain key risks.

Analysis

The investable question is not backlog size but backlog quality and the financing spread required to turn contracted GPU capacity into cash flow. CRWV's scale should produce better utilization, procurement economics, and fixed-cost absorption, while NBIS faces a steeper construction and customer-acquisition curve; that makes CRWV the cleaner near-term operating-leverage vehicle. Conversely, the market is likely underpricing the risk that contracts contain ramp, cancellation, or performance contingencies while depreciation and interest expense are immediate and largely fixed.

Over the next 1-3 months, funding conditions matter more than AI demand headlines: a 100 bp rise in long-end yields or widening in data-center financing spreads can compress equity multiples sharply because both models require repeated external capital. CRWV is more exposed to customer concentration and renewal pricing pressure from MSFT, META and other sophisticated buyers; NBIS is more exposed to execution risk on new sites, power availability, and potentially dilutive equity issuance. NVDA benefits initially from capacity buildouts, but accelerating neocloud leverage raises the probability of a later GPU-order air pocket if utilization or financing weakens.

The contrarian view is that dedicated GPU clouds are not necessarily structurally cheaper than hyperscalers once customers demand network integration, storage, security, and multi-region reliability. AMZN and MSFT can accept lower AI-cloud margins to defend strategic accounts, which would cap neocloud pricing once supply catches up over 6-18 months. The bullish thesis is falsified by sequential backlog conversion slowing, utilization failing to rise with deployed capacity, or EBITDA guidance missing because power, network, and depreciation costs outpace revenue ramps.

Key Decisions for Investors

• Prefer a 3-6 month long CRWV / short NBIS pair rather than outright exposure: CRWV has the stronger scale-to-utilization path, while NBIS carries greater buildout and dilution sensitivity. Size modestly; exit if NBIS reports materially faster revenue conversion or CRWV discloses rising customer concentration, adverse contract amendments, or weaker utilization.
• Do not chase either name solely on contracted-revenue disclosures. Establish an alert for quarterly capex, net debt, interest expense, committed-power capacity, and backlog-to-revenue conversion; a widening gap between backlog growth and recognized revenue is a short-risk signal rather than confirmation.
• Maintain NVDA as the higher-quality AI-infrastructure exposure, but hedge a portion of GPU-demand risk with a small CRWV put position dated 6-9 months out if CRWV rallies ahead of funding or earnings updates. The hedge pays if leveraged capacity expansion is delayed, while NVDA retains broader customer diversification.
• Watch AMZN and MSFT AI-cloud pricing and capex commentary over the next two earnings cycles. Evidence of aggressive GPU capacity additions or price concessions would support adding to the CRWV/NBIS short leg, as hyperscaler bundling can pressure neocloud gross margins before it visibly affects reported revenue.