I've Changed My Mind on Nebius Stock. The AI Supercycle Has Room for More Than Just Hyperscalers.

Nebius signed four AI-cloud contracts exceeding $1 billion each in Q2, with new-customer contract value rising more than ninefold. Meta has agreed to lease $12 billion of dedicated capacity from 2027, with a potential additional $15 billion, while Nvidia invested $2 billion and could support deployment of more than 5GW of Nvidia systems by 2030. Nebius is expanding U.S. and U.K. capacity and adopting a partner-financed data-center model to scale its AI cloud platform with lower infrastructure ownership requirements.

Analysis

NBIS is evolving from a capacity-constrained cloud vendor into a leveraged infrastructure aggregator: partner-financed buildouts can accelerate revenue recognition and reduce upfront equity needs, but they also shift risk into minimum-payment commitments, utilization guarantees, and counterparty concentration. The key underwriting question is not headline contract value but whether contracted pricing clears GPU depreciation, power, networking, and financing costs at normalized utilization. A modest delay in customer ramp or a fall in rental rates could disproportionately impair equity value because the operating model will carry substantial fixed-cost exposure.

The near-term equity catalyst is evidence that signed demand converts into commissioned, billable capacity without material gross-margin dilution; this is likely a 1-3 month estimate-revision story rather than a 2027 revenue story. NVDA benefits from incremental system deployments but the financial impact is immaterial relative to its existing data-center base; NBIS is the higher-beta expression. CRWV, APLD and IREN should re-rate on any proof that third-party AI-cloud capacity remains scarce, although NBIS's sovereign/enterprise positioning may command a premium only if it can demonstrate durable, non-hyperscaler demand.

Consensus appears to be capitalizing maximum contract values while discounting execution friction too lightly. The relevant downside is not simply AI demand weakening: it is hyperscalers bringing capacity back in-house, GPU generations shortening useful lives, grid interconnection delays, or a large customer exercising flexibility before infrastructure utilization reaches scale. PLTR integration could improve enterprise distribution over 6-18 months, but it should not be treated as a direct infrastructure revenue commitment until customer deployments and economics are disclosed.

Key Decisions for Investors

• Maintain NBIS as a watch-to-buy rather than chase on promotional momentum; initiate only after disclosure of contracted versus commissioned megawatts, expected gross margin, and financing obligations. Target a 3:1 upside/downside structure, with thesis invalidated by delayed site energization or a material reduction in customer prepayments.
• Express the capacity-shortage theme via a pair: long NBIS / short CRWV only after comparing EV per contracted and operational MW plus customer concentration. NBIS should outperform if its asset-light model funds growth without dilution; exit if CRWV's utilization or EBITDA guidance accelerates faster than NBIS commissioning.
• Use 6-12 month NBIS call spreads rather than outright common for event-driven upside around capacity commissioning and backlog conversion; avoid selling naked puts until obligations, lease terms, and customer termination rights are transparent.
• Keep NVDA as the lower-risk basket exposure and avoid adding solely on this development; incremental orders matter only if they contribute to a broader upward revision in networking, systems, or data-center revenue guidance.
• Monitor META capex guidance, AI-cloud rental pricing, power-interconnection milestones, and NBIS quarterly cash burn. Any combination of lower external-compute spend, falling GPU rental rates, or financing-cost escalation would invalidate the high-multiple infrastructure thesis within 1-3 quarters.