Meta agreed to spend up to $27 billion with Nebius over five years for AI data center capacity, the largest single infrastructure commitment disclosed by a hyperscaler this cycle. The deal follows a $17 billion arrangement Meta signed with Microsoft months earlier, bringing the company's disclosed long-term compute obligations above $44 billion in under six months.
Nebius, the Amsterdam-listed successor to Yandex's cloud infrastructure unit, operates primarily in Finland and Kazakhstan with expansion underway in the United States. The company trades under ticker NBIS and operates outside traditional U.S.-China tech spheres, which made it palatable to Meta's compliance desk. Shares spiked 37 percent in European trading before settling at a 29 percent gain by New York open, adding roughly $4.2 billion in market capitalization on volume twelve times the three-month average. Meta has not disclosed whether the $27 billion figure includes power costs or represents pure compute lease commitments, a distinction that matters to margin modeling.
The size and sequencing matter more than the headlines suggest. Meta spent $38 billion on capital expenditures in 2024, with roughly $32 billion directed to AI infrastructure. A $27 billion five-year deal implies annual outlays near $5.4 billion, or seventeen percent of last year's capex run rate, now locked into a single vendor relationship. That Meta signed two deals of this scale within months indicates internal forecasts show training runs and inference workloads outpacing in-house buildout by a widening margin. The Microsoft deal was structured as a hybrid lease-and-equity arrangement; Nebius disclosed only lease terms, suggesting Meta is prioritizing speed over balance-sheet optimization.
Hyperscalers typically build proprietary infrastructure to avoid vendor lock-in and margin dilution. When they lease at this scale, it signals either urgent capacity shortages or strategic hedging against geopolitical supply-chain risk. Nebius operates data centers in jurisdictions with stable power grids and cooling infrastructure but limited exposure to U.S.-China semiconductor export controls, making it a natural hedge for workloads that cannot wait for domestic GPU clusters. The Finland facilities run on hydro and nuclear baseload, which matters to Meta's 2030 carbon neutrality commitments but also insulates compute costs from natural gas volatility that has plagued Texas and Midwest data center operators.
Allocators should watch Nebius's contract disclosure in the next 10-K filing, expected late April, for revenue recognition terms and whether Meta secured capacity options beyond the initial $27 billion. If the deal includes equity warrants or preferred pricing on future builds, it signals Meta is locking in marginal cost advantages before Nvidia's next-generation Rubin architecture ships in late 2026. Watch also for whether Amazon or Google announce similar offshore compute commitments in the next 90 days; if neither moves, it suggests proprietary chip roadmaps (Trainium, TPU) are meeting internal demand and Meta is the capacity outlier. Nebius's investor deck will clarify whether it plans to repatriate cash to shareholders or plow proceeds into U.S. expansion, a decision that determines whether this is a liquidity event or a growth inflection.
Meta now has more disclosed long-term compute liabilities than any peer except Microsoft, which owns Azure and monetizes third-party workloads at gross margins above 60 percent. Meta monetizes compute exclusively through ad targeting and engagement, a lower-return use case that makes $44 billion in external commitments look less like optionality and more like necessity.