The AI race is turning contractual
Firms are increasingly securing compute capacity through long-term deals rather than on-demand purchases — Meta deepened its CoreWeave partnership and an investor memo claims OpenAI held a compute edge in 2025 — making capacity reservations as strategic as model R&D. The consequence is that balance-sheet strength and long-duration commitments matter as much as technical edge. (digitimes.com) (qz.com)
Meta just agreed to spend another $21 billion with CoreWeave for artificial intelligence cloud capacity through December 2032, turning rented chips into something that looks more like a seven-year utility contract than a normal cloud bill. (coreweave.com) That new agreement sits on top of Meta’s earlier CoreWeave commitment, bringing total contracted spend to about $35.2 billion and pushing CoreWeave’s revenue backlog to $66.8 billion. (investors.coreweave.com) CoreWeave is not a general cloud company like Amazon Web Services. CoreWeave built its business around renting out graphics processing units, which are the specialized chips used to train and run large artificial intelligence models. (investors.coreweave.com) Meta’s new contract is for inference workloads, which means serving answers after a model is built, not just teaching the model in the first place. CoreWeave said the deal will be deployed across multiple sites and will include early deployments of Nvidia’s Vera Rubin platform. (investors.coreweave.com) At the same time, OpenAI told investors it had 1.9 gigawatts of computing capacity in 2025, versus 1.4 gigawatts for Anthropic, and said it expects to reach 30 gigawatts by 2030. A gigawatt is a power-grid-sized way to describe how much data center muscle a company can lock up. (cnbc.com) OpenAI has been building that position with long-duration infrastructure deals of its own. In September 2025, OpenAI, Oracle, and SoftBank said Stargate had nearly 7 gigawatts of planned capacity and more than $400 billion of investment lined up over three years. (openai.com) This is why the competition is shifting away from who can find spare chips on short notice. The scarce thing in 2026 is not just the graphics processing unit itself, but the reserved slot in a powered data center with cooling, networking, and a contract that guarantees access years ahead. (openai.com) CoreWeave’s own numbers show how financial this race has become. The company said it reached $5 billion in annual revenue in 2025, closed an $8.5 billion financing facility in March 2026, and is using giant multiyear contracts to fund even more buildout. (investors.coreweave.com 1) (investors.coreweave.com 2) So the new edge is not just better models or better researchers. It is the ability to sign a contract large enough, long enough, and early enough that the computers will still be there when the next model is ready. (coreweave.com) (cnbc.com)
