Nebius Group, the Amsterdam-listed AI infrastructure operator, secured a $27 billion five-year compute deal with Meta, announced Thursday. The commitment follows a $17 billion Microsoft agreement signed in December. Total hyperscaler bookings now sit at $44 billion across nine years of contracted capacity. The stock moved 38% in morning European trading before settling 22% higher by close.
The Meta contract structures as a capacity reservation with minimum annual commitments beginning Q3 2025. Nebius will deploy 100,000+ H100-equivalent GPU instances across three European data center clusters—two in Finland, one in France. Meta receives priority allocation rights and can scale to 150,000 instances by 2027 if European regulatory frameworks for AI training remain stable. The deal includes upfront infrastructure financing from Meta totaling $4.2 billion, paid in tranches as construction milestones clear. First capacity goes live in 11 months.
What changed is the pricing tier. Nebius is no longer selling interruptible spot capacity or month-to-month reserved instances. This is multi-year, take-or-pay infrastructure with contractual minimums that place Nebius alongside CoreWeave and Lambda Labs in the hyperscaler-backstopped compute layer. The Microsoft deal was proof of concept. The Meta deal is proof of category. Allocators should note that Nebius now carries $44 billion in revenue visibility against a current market capitalization of $18 billion. The forward revenue multiple compressed from 6.2x to 0.4x in four months, assuming contract terms mirror industry standard 18-22% gross margins on reserved capacity.
The infrastructure financing structure matters more than the headline number. Meta's $4.2 billion upfront payment de-risks the capex cycle entirely. Nebius will not need to tap equity or high-yield debt markets to build out the Finnish and French clusters. The company's previous funding round in October 2024 raised $1.8 billion at a $12 billion valuation; that capital now sits idle while Meta's advance payments cover steel, concrete, and Nvidia shipments. This is the financial architecture that separates infrastructure plays from software resellers.
European data sovereignty laws created the opening. Meta and Microsoft both face mounting pressure from EU regulators to localize training workloads for European user data. Nebius operates under Dutch jurisdiction with French and Finnish co-location, giving hyperscalers a compliant compute layer without building greenfield. The company's GPU fleet runs on 100% renewable energy contracts, satisfying EU taxonomy requirements that U.S. providers struggle to meet at scale. Nebius effectively became the regulatory-compliant GPU buffer for hyperscalers who need European capacity but cannot afford the 18-24 month permitting delays.
Operators should watch three follow-on events. First, Nebius will likely announce a third hyperscaler contract within 90 days—Amazon's European AI workload growth is tracking 40% quarter-over-quarter, and AWS has no comparable sovereign compute partner. Second, Nvidia's next-generation Blackwell chip allocation for 2026 will determine whether Nebius can maintain margin structure as Meta scales from 100,000 to 150,000 instances; chip supply remains the binding constraint. Third, Finland's energy grid capacity will face stress tests by Q4 2025 as the two northern clusters ramp to full load; any grid curtailment triggers contractual penalty clauses that flow back to Nebius.
The company's path from Yandex spinoff to $44 billion contract backlog took 19 months. Management did not chase consumer AI or foundation model hype. They built GPU clusters in jurisdictions where hyperscalers faced regulatory friction, then waited for the capacity crisis to arrive. It arrived.