Nebius vs. CoreWeave: Which Is the Better Artificial Intelligence (AI) Infrastructure Stock to Buy Right Now

CoreWeave and Nebius are both positioned for sustained AI data-center demand with large contractual backlogs, led by Bank of America estimating a combined $2.3T backlog for Microsoft/Oracle/Amazon/Google. CoreWeave’s Q2 2026 revenue grew 112% YoY to $2.6B, but losses worsened with EPS down 90% YoY, while Nebius’ revenue rose 454% YoY to $582M and adjusted net loss narrowed 64% to $33.2M—yet Nebius trades at a much more expensive sales multiple (CoreWeave is comparatively more value-oriented). As of 2026, Nebius shares are up 146% vs CoreWeave up 18%, creating a split between fundamentals (Nebius improving profitability) and visibility (CoreWeave’s $104B revenue backlog, up 246% YoY).

Analysis

The market is likely mispricing the distinction between demand visibility and economic quality. In neocloud, backlog is only valuable if it converts into cash after GPU depreciation, power costs, financing, and utilization risk; that means the real winner over the next 3-12 months is not necessarily the fastest-growing name, but the operator with the best unit economics and customer concentration profile. If hyperscaler capex stays elevated, suppliers like NVDA, power/cooling vendors, and data-center REIT-adjacent infrastructure names benefit more reliably than the lease-heavy operators themselves.

The bigger second-order risk is that backlog can become a crowded consensus signal exactly when investors start underwriting 2027-28 revenue too aggressively. If either NBIS or CRWV reports a step-up in capex without a corresponding margin inflection, multiple compression can happen fast because these names trade more like duration assets than infrastructure monopolies. Over 1-3 months, the key catalyst is whether next earnings show backlog conversion, gross margin stability, and capex discipline; over 6-18 months, the question is whether customer workloads migrate from rented neocloud capacity back to hyperscaler-owned stacks once deployment pipelines normalize.

Contrarian view: the article frames this as a valuation choice, but the more important debate is whether the category deserves high-multiple software treatment at all. If AI compute supply catches up, pricing power likely shifts back to customers, not providers, and the leaders with the most contracted revenue could still disappoint on returns on invested capital. That makes this less a simple long-vs-longer-longer story and more a potential short-basket setup against the most expensive names if revenue growth decelerates before operating leverage appears.

Key Decisions for Investors

• Prefer a relative-value short NBIS / long CRWV pair only if the next print confirms backlog-to-revenue conversion and NBIS's multiple stays meaningfully richer; otherwise avoid chasing either name outright.
• If long exposure is desired to the AI infrastructure theme, express it via NVDA, MSFT, AMZN, or GOOGL rather than the neocloud operators; the risk/reward is better because supplier economics are cleaner and balance-sheet risk is lower.
• Set an earnings watch item on CRWV and NBIS: if either name shows capex growth outpacing gross profit dollars for another quarter, treat that as a multiple-risk trigger and fade rallies.
• Use a basket hedge: long AI infrastructure suppliers (NVDA/SMCI-type exposure where applicable) against a short basket of high-multiple neocloud operators if the market starts valuing backlog as guaranteed earnings.