CoreWeave Signs $21 Billion Deal With Meta for AI Cloud Capacity

CoreWeave secured a six-year, $21 billion agreement with Meta Platforms to supply AI cloud capacity through December 2032, a contract that pushed the neocloud company's cumulative commitments with Meta to approximately $35 billion and sent CoreWeave shares up nearly 12 percent on the news.

The expanded deal builds on an existing relationship originally valued at up to $14.2 billion. Under the new terms, CoreWeave will deploy dedicated infrastructure across multiple data center locations, including some of the first commercial deployments of NVIDIA's Vera Rubin platform, the next-generation chip and rack-level architecture engineered for large-scale model training and agentic inference workloads.

CEO Michael Intrator framed the contract as a market verdict on CoreWeave's positioning. "This is another example that leading companies are choosing CoreWeave's AI cloud to run their most demanding workloads," Intrator said.

The deal crystallizes a broader shift in how hyperscalers are managing compute strategy: rather than waiting years for full data-center retrofits, companies like Meta are effectively renting early access to platform-class hardware from specialized third-party cloud providers. For Meta, the arrangement secures the raw compute needed to accelerate both internal model training and the inference pipelines powering its consumer-facing AI products through the end of the decade.

To fund the capital-intensive buildout required by the agreement, CoreWeave moved quickly to the debt markets. The company issued $1.75 billion in high-yield bonds maturing in 2031, upsized from an initial $1.25 billion, and priced an additional $3 billion in convertible notes. The combined $4.75 billion in new debt underscores how expensive it is to host Vera Rubin-class hardware at scale, and it gives investors reason to scrutinize CoreWeave's balance sheet even as its topline contracts expand rapidly.

The financing push and the deal itself raised flags for regulators tracking the concentration of AI infrastructure. CoreWeave is targeting five gigawatts of total capacity by 2030, a scale that will place sustained pressure on local power grids. Its growing role as a central node in the AI economy has attracted attention from the Department of Energy and antitrust regulators concerned about deepening ties between NVIDIA and its preferred distributors.

For competing cloud providers, the deal narrows an already tight window. NVIDIA's Vera Rubin platform is in limited early availability, and long-term committed deployments like this one reduce the supply accessible to rivals. Firms that cannot secure similar contractual access to Rubin-class racks risk falling measurably behind in the capacity race as model complexity and inference demand continue to climb.

The CoreWeave-Meta arrangement confirms that the most consequential competition in AI is no longer happening purely at the model layer. It is happening in data centers, on power grids, and in debt markets, with billion-dollar commitments locking in infrastructure dependencies well into the next decade.