Meta expands cloud pact with CoreWeave
Meta has signed a fresh long‑term deal to scale its AI cloud capacity with CoreWeave, locking in a multi‑billion‑dollar commitment to run inference workloads. The expanded agreement is reported as a $21 billion deal designed to give Meta more outsourced GPU and hosting capacity as it chases model improvements. This is an operational move that converts model ambitions into guaranteed datacentre capacity and financial commitments. (reuters.com, investors.coreweave.com)
Meta just turned an artificial intelligence plan into a utility bill. On April 9, CoreWeave said Meta signed an expanded agreement for about $21 billion of cloud capacity running through December 2032. (investors.coreweave.com) This is not Meta buying a new model or a new app. It is Meta reserving the computing warehouses, graphics chips, power, and networking needed to keep artificial intelligence systems answering real users every day. (reuters.com, investors.coreweave.com) CoreWeave said the new capacity is for inference workloads, which is the part after training when a model takes a prompt and produces an answer. Training is like teaching a student for months; inference is the exam happening millions of times a day. (investors.coreweave.com) That distinction explains the size of the deal. A company can train one big model once, but serving Facebook, Instagram, WhatsApp, and Meta’s other products means paying for constant responses every hour those apps are open. (reuters.com) CoreWeave is the specialist landlord in this story. It rents out cloud systems built around high-end graphics processing units, and its April 9 statement said some of Meta’s new capacity will include early deployments of Nvidia’s Vera Rubin platform. (investors.coreweave.com) Meta is not doing this because it lacks data centers of its own. Meta said earlier this year that 2026 capital spending would be $115 billion to $135 billion, up from $72.22 billion in 2025, with the increase tied to artificial intelligence infrastructure and data centers. (datacenterdynamics.com, finance.yahoo.com) So the CoreWeave contract sits on top of Meta’s own buildout, not instead of it. Meta is paying to build some of the highway itself while also reserving extra lanes from a company that already has chips, racks, and power lined up. (investors.coreweave.com, datacenterdynamics.com) The timing also tells you where the pressure is. Reuters reported that Meta is trying to catch up after an underwhelming model release last year, so this deal is less about a splashy launch and more about making sure the next models have enough room to run. (reuters.com) For CoreWeave, this is the kind of contract that can reshape a company. CoreWeave said in March 2026 that it reached $5 billion in annual revenue in 2025, and this single Meta expansion is worth roughly four times that figure spread over the life of the agreement. (sec.gov, investors.coreweave.com) It also deepens a relationship that was already large. CNBC reported the new $21 billion commitment comes on top of a prior $14.2 billion arrangement, which means Meta is steadily moving more of its artificial intelligence load onto reserved outside capacity instead of relying only on whatever it can build in-house first. (cnbc.com) The plain-English version is simple: the artificial intelligence race is now a construction race. The winners are not just the companies with the best models, but the companies that can lock up enough chips, electricity, buildings, and long-term contracts to keep those models running at internet scale. (reuters.com, investors.coreweave.com)