CoreWeave expands OpenAI deal by up to $6.5bn, total contracts reach ~$22.4bn

CoreWeave, the GPU-cloud provider that went public earlier in the year, agreed to supply OpenAI with up to a further $6.5 billion of data-centre capacity, its third expansion of the relationship in 2025. Combined with an initial contract worth up to $11.9 billion signed in March and a $4 billion add-on agreed in May, the company said its total contracted business with OpenAI had reached roughly $22.4 billion.

The agreement was framed around the capacity needed to train and run OpenAI’s next generation of models, with CoreWeave’s chief executive, Michael Intrator, citing the scale of demand from “world-leading innovators” for training and inference infrastructure. Unlike the hyperscalers, CoreWeave does not build its own chips; its business is renting out data centres built around large numbers of Nvidia GPUs, financed substantially through debt secured against those long-term contracts.

The repeated expansions illustrate the pattern that defined AI infrastructure spending through 2025: compute demand from frontier labs outstripping what any single supplier could commit to in one agreement, producing a series of successive top-ups rather than one large contract. For CoreWeave, whose revenue is heavily concentrated in a small number of large customers including OpenAI and Microsoft, the growing total also underlined a dependency that analysts flagged as a risk to the business model even as it drove the company’s valuation.