Meta locked $27 billion with Nebius across five years for AI compute and data center services, following a $17 billion Microsoft commitment in January. The two deals represent $44 billion in forward infrastructure spend in less than ninety days.
Nebius, the AI infrastructure spinout from Yandex, now holds multi-year purchase obligations from two of the four hyperscalers capable of spending at this velocity. Meta's commitment specifies compute and data center services, not raw hardware resale. The structure suggests co-location, managed capacity, and potentially shared cooling or power infrastructure—arrangements that carry lock-in beyond the contract term. Nebius stock moved sharply on the announcement, though the company remains thinly traded and concentrated in European hours.
The $27 billion figure is notable not for its size but for its horizon. Five-year compute commitments were rare until 2023. Now they are table stakes. Meta spent approximately $38 billion on capital expenditures in 2024, heavily weighted toward AI infrastructure. A $27 billion Nebius deal implies $5.4 billion annually if spread evenly, roughly 14% of last year's capex budget. But these deals are not spread evenly. The first year typically carries lighter outlays while data centers ramp. The back half accelerates. If the deal follows standard hyperscaler phasing, years three through five could exceed $7 billion annually.
This is the second time in three months Meta has formalized a multi-billion-dollar infrastructure partnership outside its owned-and-operated footprint. The repetition signals a shift. Meta historically built its own data centers and negotiated spot capacity when needed. Long-term commitments reduce flexibility but guarantee supply in a market where AI-grade compute is no longer fungible. Nebius benefits from revenue visibility and the ability to pre-finance builds against contracted cash flows. Meta benefits from capacity certainty during a period when lead times for high-bandwidth networking and liquid cooling exceed eighteen months.
The Nebius-Microsoft deal in January was $17 billion over six years. The Meta deal is 59% larger and 17% shorter. That pricing and term structure suggests Meta is paying for priority or exclusivity within certain geographies. Nebius operates data centers in Finland and Kazakhstan, both of which offer low power costs and cold-climate cooling advantages. Meta's commitment may include pre-reserved capacity in facilities not yet online, a common structure when hyperscalers want to lock competitors out of specific regions.
Allocators should watch whether Alphabet or Amazon announce similar deals with non-traditional infrastructure providers in the next ninety days. If they do, the hyperscaler build-versus-buy calculus has permanently shifted. If they do not, Meta and Microsoft are securing a structural cost or capacity advantage. Nebius will likely raise debt against these contracts within six months. The company's ability to securitize forward revenue streams will determine whether it can scale to meet both commitments simultaneously. Any delay or capacity miss on the Microsoft contract would re-price the Meta deal immediately.
The infrastructure consolidation thesis assumed hyperscalers would merge overlapping facilities and retire redundant capacity. The purchasing thesis is different. It assumes hyperscalers will lock supply through long-term commitments, raising the cost of entry for competitors who arrive late. Meta just committed $44 billion to that future in fewer than ninety days.