Meta committed up to $27 billion over five years to Nebius for AI infrastructure capacity, the largest single compute contract disclosed in 2026 and the second hyperscaler anchor for a company that traded as a Russian tech conglomerate eighteen months ago. The Facebook parent will draw down GPU clusters and training infrastructure as its Llama family scales beyond current 405B parameter models, with first capacity scheduled for Q3 2026.
Nebius closed a $17 billion Microsoft contract in December 2025, putting combined hyperscaler commitments at $44 billion before adding smaller enterprise accounts. The company operates 12 data center campuses across three continents, with 8.2 gigawatts of contracted power capacity and 47,000 H100-equivalent GPUs already racked. Meta's deal includes priority access to Nebius-designed liquid cooling infrastructure that cuts training costs by 22% compared to air-cooled alternatives, according to disclosed benchmarks. Microsoft's contract included similar cooling provisions, suggesting Nebius positioned thermal efficiency as the wedge into both accounts.
The hyperscaler consolidation matters because it separates infrastructure providers into two tiers. Nebius, CoreWeave, and Lambda Labs now hold multi-year commitments that lock in utilization rates above 80% and create predictable revenue models Wall Street can underwrite. Smaller GPU brokers without anchor contracts face spot-market exposure and single-digit operating margins as H100 lease rates compress 38% year-over-year. Nebius traded at $14.20 before the Meta announcement and touched $22.80 intraday, valuing the company at $8.6 billion against $44 billion in signed backlog, an unusual discount that reflects skepticism about execution risk on this scale.
The contract also reshapes Nvidia's customer concentration. Meta and Microsoft together represent 31% of Nvidia's datacenter revenue in fiscal 2026, and both are now committed to infrastructure they do not own. If Nebius hits contractual minimums, it becomes Nvidia's third-largest indirect customer by 2028, behind only the hyperscalers' direct purchases. That shift gives Nebius pricing leverage in future Blackwell and Rubin orders, particularly if Meta or Microsoft negotiate capacity guarantees into their contracts. The company has not disclosed whether it holds GPU allocation rights, but the five-year term suggests some form of supply commitment.
Allocators should track Nebius's Q2 2026 earnings in mid-August for utilization ramp details and whether Meta's contract includes minimum take-or-pay clauses. The company must add 2.4 gigawatts of live capacity by year-end to meet stated deployment schedules, requiring roughly $6 billion in capex if historical cost structures hold. Debt markets will price that financing in May and June. CoreWeave is expected to announce a comparable hyperscaler deal before summer, which would set the benchmark for whether three-player oligopoly pricing sustains or whether competition reopens.
Nebius now holds more signed hyperscaler paper than its market capitalization, a gap that closes through execution or recalibrates through equity dilution. The next six months determine which.