Meta has agreed to a $21 billion deal with CoreWeave for dedicated AI cloud capacity through December 2032. This is one of the largest infrastructure contracts in the industry and highlights how quickly Meta is scaling its AI operations.
This new deal adds to a previous $14.2 billion agreement that runs through December 2031. Combined, the contracts are worth more than $35 billion, although the exact total is unclear.
The $21 billion figure includes Meta’s decision to add more capacity under the earlier deal, which means there is some overlap. Even so, the size of the commitment is notable.
After the announcement, CoreWeave shares rose by about 8%, according to Bloomberg.
What Meta is actually buying
The contract is mainly about inference, which is when trained AI models deliver real-time results to users. This is different from training, where models learn from data. Inference is often where costs and delays are most obvious when operating at scale.
Meta runs Llama-based models in its apps like Facebook, Instagram, WhatsApp, and Messenger, which reach billions of people. Handling this workload needs a fast, widely distributed infrastructure built for AI, not just regular cloud services.
CoreWeave will create dedicated capacity in several locations and will be one of the first to use the NVIDIA Vera Rubin platform, Nvidia’s next-generation AI chip following the Blackwell series. Vera Rubin should help AI models handle multi-step tasks on their own.
Getting early access to this hardware matters. Companies that use Vera Rubin first will have a performance advantage before it is available to everyone.
Why Meta is going outside its own infrastructure
Meta is one of the few companies that can build and run its own large-scale AI infrastructure. This year, it plans to spend between $115 billion and $135 billion on capital expenses, almost twice what it spent in 2024 and well above Wall Street’s expectations. Most of this money will go to its data centres.
The CoreWeave deal fits Meta’s strategy. Building new data centres takes years, but demand for inference capacity is growing rapidly due to Meta’s AI products and competition from Google and Apple. CoreWeave can set up GPU clusters faster than Meta can build new sites.
The broader picture
CoreWeave is not the only outside provider Meta works with. Meta also signed a five-year AI infrastructure deal with Nebius worth up to $27 billion. This shows Meta is choosing several vendors instead of just one. Using different suppliers reduces risk and helps keep prices competitive.
For CoreWeave, working with Meta now secures a big part of its expected revenue through the end of the decade. The company recently raised $8.5 billion in debt, partly backed by its Meta contracts.
This deal forms one of the main infrastructure partnerships in today’s AI growth. Whether Meta’s $115–135 billion spending this year is the right choice will depend on how its competition with Google, OpenAI, and Apple plays out over the next three years.
