Nebius vs. CoreWeave: Which AI Cloud Stock Is the Better Buy?

Nebius reported Q2 revenue of $582 million, up 454% year over year, with adjusted EBITDA of $236 million and a 41% margin; it reaffirmed 2026 revenue guidance of $3.0-$3.4 billion while raising capex to $20-$25 billion. CoreWeave delivered $2.6 billion of Q2 revenue, up 112%, and expanded backlog 246% to $104.2 billion, but posted a $626 million net loss as capex reached $9.4 billion and interest expense rose to $640 million. The analysis favors Nebius due to faster growth, improving margins, lower 5.47x price/book valuation versus CoreWeave's 7.25x, and a more capital-efficient expansion model, despite both companies facing substantial infrastructure funding and supply-chain requirements.

Analysis

The relevant distinction is not growth versus growth; it is who bears utilization, financing, and residual-value risk. NBIS’s partner-funded model can produce superior equity returns if customer prepayments are genuinely non-cancellable and matched to construction milestones, but it shifts value to counterparties if capacity pricing normalizes. CRWV’s owned-infrastructure approach creates greater operating leverage in a tight-GPU/power market, yet its debt service makes the equity increasingly a long-duration claim on uninterrupted utilization and refinancing access.

Over the next 1-3 months, CRWV faces a more demanding catalyst burden: investors need evidence that incremental contracts convert into cash collections fast enough to contain net leverage, rather than merely expand backlog. The critical disclosure is contract concentration, term, cancellation rights, and whether backlog is indexed to power/GPU delivery; a delay in energized capacity can leave interest expense accruing without corresponding revenue. NBIS has the cleaner estimate-revision setup, but its sharp relative outperformance raises the risk that any margin dip during the build phase produces a crowded-growth unwind.

The second-order winner is NVDA: both strategies require accelerated GPU deployment, and cloud-provider competition supports system demand beyond hyperscaler procurement. Conversely, AMZN and GOOG are not direct losers, but third-party GPU clouds can cap pricing for excess AI capacity and force greater product differentiation; the threat rises if enterprises adopt portable inference workloads rather than locking into proprietary cloud stacks.

Consensus is likely over-crediting headline contracted capacity as equivalent to recurring software revenue. In a 6-18 month normalization scenario, power availability improves, GPU generations turn over faster, and short-duration pricing compresses; the better equity is the operator with contractual pass-throughs, low refinancing needs, and demonstrable cash conversion—not necessarily the largest backlog.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: long NBIS / short CRWV, sized beta-neutral. The thesis is financing asymmetry and likely estimate dispersion; reassess if CRWV demonstrates two consecutive quarters of positive operating cash flow after interest and capex funding, or if NBIS’s customer prepayments fail to cover expected construction draws.
• Do not chase NBIS outright following momentum. Accumulate only on a post-results margin/capex-driven pullback, contingent on verification that prepayments are enforceable and customer concentration is manageable. Target a 6-12 month holding period; exit if management lowers its contracted-power build schedule or EBITDA-margin trajectory.
• Maintain an overweight NVDA versus the AI-cloud cohort for the next 6 months. It captures deployment demand without taking project-finance or customer-utilization risk; trim if lead times normalize materially and next-generation system pricing or gross-margin guidance weakens.
• Use CRWV quarterly results as a credit-risk watch item rather than a directional long: monitor interest expense versus EBITDA, construction-in-progress growth, and disclosed debt maturities. A widening in its bond/loan spreads or another financing required before capacity becomes revenue-producing would be a catalyst to add to the short leg.
• Treat PLTR as an indirect beneficiary only if commercial workload wins translate into disclosed consumption growth, not partnership announcements. A sustained acceleration in commercial revenue would support a tactical long; absent that evidence, the infrastructure partnership has limited standalone valuation relevance.