CoreWeave expands OpenAI partnership to $22.4B as AI data center race heats up

OpenAI is spending big to secure the computing horsepower behind its next generation of models. The company has expanded its agreement with AI cloud provider CoreWeave by as much as $6.5 billion, pushing the total value of its deals to $22.4 billion, the company announced on Thursday. It’s the third time this year the two have deepened their partnership, highlighting the breakneck demand for data center capacity in the middle of the AI boom.

Back in March, OpenAI struck an initial pact with CoreWeave worth up to $11.9 billion. That was followed by a $4 billion extension in May. Thursday’s announcement adds yet another layer, with CoreWeave confirming the capacity will be used to train “most advanced next-generation models.” Reuters first reported the deal.

The deals show how central CoreWeave has become to OpenAI’s infrastructure strategy. Even as it works with Oracle on a $300 billion project and teams with SoftBank to build new data centers, OpenAI is securing multiple suppliers to fuel its Stargate initiative — a massive buildout targeting 10 gigawatts of compute and as much as $500 billion in investment over the next three years.

“We are proud to expand our relationship with OpenAI, a company consistently at the forefront of advancing artificial intelligence,” said Michael Intrator, Co-Founder, Chairman and Chief Executive Officer of CoreWeave. “This milestone affirms the trust that world-leading innovators have in CoreWeave’s ability to power the most demanding inference and training workloads at an unmatched pace. ”

CoreWeave’s stock initially pared losses on the news, trading about 4% lower in premarket hours. Nvidia, which owns more than 5% of CoreWeave, has a direct interest too. The chipmaker signed a $6.3 billion supply agreement with CoreWeave this month and said it will buy up any cloud capacity left unsold. Nvidia earlier this week also announced plans to invest as much as $100 billion in OpenAI, a move that ties the two most high-profile names in AI even closer together — and raises antitrust questions along the way.

For OpenAI, the stakes are enormous. Training frontier models requires infrastructure on a scale that rivals entire national grids. With billions in commitments to CoreWeave, Oracle, SoftBank, and Nvidia, the company is betting big that its next wave of products will justify one of the most expensive computing buildouts in history.

Founded in 2017 as a crypto mining outfit, CoreWeave pivoted into AI infrastructure and now rents out access to Nvidia GPUs for training and running large AI models. Microsoft is by far its biggest customer, with others including Meta, IBM, and Cohere.