CoreWeave Inc., an emerging “neocloud” provider of AI computing power, has signed a deal to supply Meta Platforms Inc. with up to $14.2 billion in computing resources, highlighting the escalating costs of developing and running advanced AI models, Bloomberg reports.
The agreement will provide Meta access to Nvidia Corp.’s latest GB300 systems, according to CoreWeave Chief Executive Officer Michael Intrator. “They loved our infrastructure in earlier contracts and came back for more,” Intrator told Bloomberg, emphasizing the social media giant’s reliance on cutting-edge technology to support its AI initiatives. Meta declined to comment.
CoreWeave, which competes with companies such as Nebius Group and Nscale Global Holdings Ltd., has seen its stock more than triple since its March initial public offering, reflecting soaring demand for AI infrastructure. The deal also helps diversify CoreWeave’s customer base away from Microsoft Corp., which accounted for 71% of revenue in the quarter ending June. “When we came out in the IPO, we got dinged because of our customer concentration. This is clearly a step in the right direction,” Intrator said.
The neocloud model, which rents access to high-end AI chips, is capital-intensive, and CoreWeave plans to continue tapping debt markets to expand its infrastructure. “We will tap the debt markets periodically as we continue to expand,” Intrator told Bloomberg, noting that other major cloud providers, including Meta and Oracle, have recently used debt financing for large-scale data center projects.
The deal follows a multibillion-dollar commitment from OpenAI last week and underscores the growing investments by major tech companies in AI hardware and talent. Meta has projected capital expenditures of up to $72 billion this year, much of it directed toward AI and the data centers that train and run these models.
