Meta committed $27 billion over five years to Nebius Group, the Amsterdam-listed AI infrastructure operator spun out of Yandex in July. The contract covers compute capacity, networking infrastructure, and sovereign-grade data center access across North America and Europe. Nebius shares moved 31% intraday on NASDAQ before settling 22% higher at close. The company now trades at $41.20, a 340% gain since the Microsoft announcement.
The deal follows a $17 billion Microsoft contract announced December 2024, giving Nebius $44 billion in forward commitments from two of the three largest AI compute buyers. Meta's structure mirrors Microsoft's: phased capacity delivery tied to specific model training milestones, with penalty clauses if Nebius misses Q2 2026 and Q4 2027 buildout deadlines. Payment terms are quarterly in advance, weighted 60% toward years three through five. Meta will embed technical staff at Nebius facilities in Finland and Nevada starting April 2025.
This is the second time in six months Meta has moved AI infrastructure spend outside the AWS-Google-Azure triad. The company spent $38.7 billion on capex in 2024, 68% of which went to NVIDIA hardware deployed on third-party infrastructure. By 2027, that ratio is expected to invert: $52 billion total capex, $34 billion external, with Nebius holding the largest single contract. Meta's disclosure filing notes "geographically distributed resilience" as the primary justification, a phrase that typically signals regulatory hedging in the EU and data sovereignty concerns in post-CLOUD Act enterprise sales.
Nebius was carved out of Yandex after sanctions pressure forced the Russian search giant to divest non-domestic assets. The company retained Yandex's custom silicon roadmap—an in-house inference chip called Graphcore-adjacent architecture—and its Scandinavian data center footprint. It went public via SPAC in October at $9.40 per share. The Microsoft deal gave it credibility. The Meta deal gives it inevitability. Operators should note that Nebius is now the fourth-largest non-hyperscaler infrastructure provider by forward contract value, behind only CoreWeave, Lambda Labs, and Crusoe Energy.
The timing matters. Meta's Llama 4 training is scheduled to begin Q3 2025, requiring 1.2 exaflops of sustained compute, roughly 4x the Llama 3.1 workload. AWS and Azure cannot deliver that capacity in Meta's preferred Nordics-plus-Nevada geographic split without cannibalizing OpenAI and Anthropic allocations. Nebius can, because it built for this exact constraint. The company's Q1 2025 earnings call, scheduled May 8, will clarify whether it can meet the Q2 2026 milestone without sacrificing margin. If it does, Oracle and Salesforce are the next likely customers. If it does not, Meta's contract includes a $2.1 billion clawback provision.
Nebius is now a $14.6 billion market cap company with $44 billion in revenue visibility and $1.8 billion in net debt. The financing closed last week with Barclays and Deutsche Bank as lead arrangers.