Meta backs CoreWeave with $21B
Meta has committed roughly $21 billion to CoreWeave to secure AI cloud infrastructure capacity through 2032, a major long‑horizon bet on third‑party GPU hosting. The capital commitment signals hyperscalers are buying future capacity rather than depending solely on public cloud spot markets, which could reshape pricing and supplier concentration in GPU access (x.com).
Meta just agreed to spend about $21 billion with CoreWeave for artificial intelligence cloud capacity through December 2032, which means one of the world’s biggest tech companies is locking in years of chip access now instead of hoping the market has spare capacity later. (coreweave.com) CoreWeave is not Amazon Web Services or Microsoft Azure. It is a specialist cloud company that rents out giant clusters of NVIDIA graphics processing units, the chips used to train and run artificial intelligence systems. (cnbc.com) This new deal is not Meta’s first check to CoreWeave. CNBC reported that Meta already had a prior arrangement worth $14.2 billion, so the relationship has grown into roughly $35 billion of contracted business. (cnbc.com) The detail that stands out is what Meta wants the machines for. CoreWeave said Meta will use this capacity to scale inference workloads, which is the step where a trained model answers your question, writes your caption, or ranks your feed in real time. (coreweave.com) Inference is different from training. Training is the expensive boot camp where a model learns from huge piles of data, while inference is the day job where millions or billions of users keep asking it to do things every second. (coreweave.com) Meta is making this bet while already spending at a historic pace on its own data centers. Meta said capital expenditures were $72.22 billion in 2025, and Reuters reported its 2026 capital spending guidance is $115 billion to $135 billion. (investor.atmeta.com) (usnews.com) So this is not a company choosing rented computers instead of owned computers. It is a company doing both at once, because building enough power, buildings, cooling, and chip supply for artificial intelligence has become too slow and too uncertain to leave to one pipeline. (cnbc.com) (coreweave.com) CoreWeave’s side of the story is just as important. The company told investors in February that it and NVIDIA plan to build more than 5 gigawatts of “artificial intelligence factories” by 2030, and NVIDIA put another $2 billion into CoreWeave stock in January. (coreweave.com) That makes the supply chain unusually tight. NVIDIA supplies the graphics processing units, CoreWeave turns them into rentable cloud capacity, and Meta signs multiyear contracts to reserve that capacity before rivals do. (coreweave.com 1) (coreweave.com 2) It also helps explain why CoreWeave has become so valuable so quickly. CoreWeave generated $1.92 billion of revenue in 2024, and Microsoft accounted for 62% of that total, so landing Meta as another giant customer reduces the story that this business lives or dies on one buyer. (cnbc.com) (bloomberg.com) The hardware in this contract also points to how far ahead companies are trying to book supply. CoreWeave said the deployment will include some of the first installations of NVIDIA’s Vera Rubin platform, which means Meta is reserving future chip generations, not just today’s leftovers. (coreweave.com) For everyone else trying to buy graphics processing units, this is the uncomfortable part. When Meta prepays for years of dedicated capacity through 2032, fewer top-tier chips are left floating around for startups, smaller clouds, and anyone hoping spot prices will come down soon. (coreweave.com) (cnbc.com)
