Meta says it will spend an additional $21B on CoreWeave’s AI infrastructure

Meta Platforms Inc. said today it’s going to throw an additional $21 billion at the cloud artificial intelligence infrastructure provider CoreWeave Inc. as it strives to keep pace with rival AI developers.

The deal expands the social media giant’s partnership with CoreWeave. In September, it revealed it was going to spend $14.2 billion on the cloud provider’s AI compute infrastructure. That original deal was set to run until 2031, but the new deal spans 2027 to 2032, Meta said, giving it access to CoreWeave’s much-needed resources for years to come.

CoreWeave operates a network of data centers that are stacked with hundreds of thousands of Nvidia Corp.’s graphics processing units, which are the most popular chips for running AI models. Meta needs access to those resources to meet the growing demand for its AI services. Although it’s known to operate dozens of its own data center facilities globally, demand is so high that it also needs to rent extra capacity from providers such as CoreWeave, which also serves Microsoft Corp., OpenAI Group PBC and other AI firms.

In January, Meta said it will spend between $115 billion and $135 billion on AI infrastructure this year, almost double the amount it spent in 2025.

Meta has a lot of AI projects. It uses AI algorithms to enhance its core advertising business and provides tools for creators across applications such as Facebook, Instagram and Threads. It’s also trying to compete with OpenAI, Google LLC and Anthropic PBC as a frontier model developer, but it has struggled to match the success of those rivals.

Following the disappointing reception of its last Llama models, Meta pivoted, spending lavishly to set up a new Superintelligence Labs group headed by Alexandr Wang to develop more advanced models. On Wednesday came the culmination of that effort when the company released a new model called Muse Spark.

With so many AI projects going on, Meta needs tons of compute resources. It has partnered with CoreWeave since 2023, and today’s deal suggests it sees more value in its infrastructure resources beyond just extra capacity. CoreWeave Chief Executive Mike Intrator told CNBC that his company’s AI infrastructure enables Meta to take full advantage of all of the talent it has hired in the Superintelligence Labs unit. “They hired from across the space, people who have used infrastructure from all different folks, and they came back to us,” he pointed out.

Meta said the latest investment in CoreWeave expands the company’s “portfolio-based approach” to securing AI infrastructure, ensuring it has access to different kinds of resources and suppliers. The deal also helps CoreWeave to reduce its reliance on Microsoft, which remains its biggest customer. In 2024, the Redmond-based company accounted for 62% of CoreWeave’s total revenue, but that has since been reduced to just 35%, Intrator said.

CoreWeave also said today it will raise an additional $3 billion in debt to fuel its own data center buildout. It has taken on massive loans to try to establish itself as a major player in the AI chip industry, and held more than $21 billion in debt on its balance sheet at the end of fiscal 2025. Then, in March, it borrowed $8.5 billion more to expand its data centers to fulfill the demands of new contracts it has signed. Those deals have helped its stock rise 24% in the year to date.

Intrator told CNBC that he expects to strengthen his company’s relationship with Meta even further in the future, as its AI infrastructure buildout is far from finished. “They’re going to continue to do it themselves, but they’re also going to continue to do it with us,” he said.

Image: SiliconANGLE/Dreamina AI