Meta locks in compute

Meta just agreed to spend another $21 billion on rented artificial intelligence computing from CoreWeave, and the contract runs through December 2032 instead of for a single product cycle or a one-year burst of demand. (Reuters: ) (CoreWeave: ) That money sits on top of an earlier $14.2 billion arrangement, which means the two companies now have roughly $35 billion tied up in one infrastructure relationship. (CNBC: ) CoreWeave is not a consumer app company or a chip designer. It is a cloud provider that packs data centers with Nvidia graphics processing units and rents that capacity to companies that need to train and run large artificial intelligence models. (CNBC: ) Meta is building its own data centers at the same time, but this deal shows that owning the factory is not enough when demand is moving faster than construction. CoreWeave chief executive Michael Intrator told CNBC that companies able to buy their own compute still want outside capacity because “there’s just too much risk not to.” (CNBC: ) The new contract is aimed at inference workloads, which is the part where a trained model answers real user requests instead of learning from data in the lab. That is the expensive, always-on layer you need when millions of people start asking chatbots and recommendation systems for results at the same time. (CoreWeave: ) CoreWeave also said the capacity will include some of the first deployments of Nvidia’s Vera Rubin platform across multiple sites. That turns the deal from a simple rental agreement into a place-in-line for the next generation of scarce hardware. (CoreWeave: ) Meta has room to make bets this large because it told investors in January that 2026 capital spending would land between $115 billion and $135 billion, nearly double what CNBC said it spent on capital expenditures in 2025. (CNBC: ) (Meta Investor Relations: ) The urgency comes from competition as much as from engineering. Reuters reported that Meta is trying to catch up after an underwhelming model release last year, while CNBC said the market for frontier models is still led by OpenAI, Anthropic, and Google. (Reuters: ) (CNBC: ) For CoreWeave, a contract that stretches to 2032 helps justify raising fresh debt and building more facilities before the revenue fully arrives. CNBC reported that the company said on Thursday it would also raise $3 billion in new debt. (CNBC: ) For Meta, the purchase is a hedge against the one thing money alone cannot instantly fix in artificial intelligence: time. You can hire researchers in a quarter, but you cannot conjure a fully powered cluster of top-end Nvidia machines overnight, so Meta is paying now to make sure the machines are there when its next models are ready. (CNBC: ) (CoreWeave: )