Meta signed a five-year contract with Nebius worth up to $27 billion for AI compute infrastructure, the largest known procurement commitment in the company's history and the second such deal in three months. The agreement surfaces weeks after Meta committed $17 billion to Microsoft for cloud compute, indicating the Facebook parent is locking in capacity across multiple vendors rather than building exclusively in-house. Nebius stock moved sharply on the announcement. Zuckerberg's net worth fell $18 billion the same day as Meta shares dropped 10% following disclosure of a 91% collapse in free cash flow and a $145 billion aggregate AI spending forecast.
The Nebius contract structures payments across sixty months with no disclosed minimums, suggesting Meta retains flexibility to scale usage based on model training cycles and inference demand. Nebius, a relatively young infrastructure provider spun out of Yandex's international operations, now holds two of the three largest known hyperscale AI procurement contracts in the market. The Microsoft deal closed in December. The Meta commitment implies Nebius will deploy data center capacity at a pace consistent with $5.4 billion annual run-rate, a figure that exceeds the company's prior twelve-month revenue by a factor of six. The stock response reflects both the revenue certainty and the execution risk.
The twin deals—Microsoft in Q4, Meta in Q1—signal a structural shift in how frontier AI labs and large platform companies approach compute procurement. Meta spent $38 billion on capex in the prior fiscal year, nearly all of it on AI infrastructure. The $145 billion multi-year guidance disclosed in the earnings call implies Meta will spend more on AI compute over the next three years than it has on total capex in its entire history as a public company. Free cash flow fell to $2.4 billion last quarter from $28 billion a year earlier, driven almost entirely by data center construction and GPU procurement. The market punished the stock, but the Nebius and Microsoft contracts suggest Meta views captive capacity as insufficient. Renting at scale hedges against hardware obsolescence and allows the company to shift workloads across vendors as models evolve.
Operators should monitor two follow-on events in the next ninety days. First, whether Alphabet or Amazon announce comparable multi-vendor procurement deals, which would confirm the shift from build-to-own toward lease-at-scale as the dominant infrastructure strategy among hyperscalers. Second, whether Nebius discloses the geographic distribution of capacity under the Meta contract. If the deal concentrates capacity in non-U.S. jurisdictions, it implies Meta is preparing for regulatory or geopolitical scenarios that restrict cross-border data flows. Both outcomes carry implications for the sovereign cloud build-out and for secondary-market pricing of GPU clusters.
Nebius now holds commitments worth $44 billion from two of the four largest AI spenders in the market, with contract terms extending into 2030.