Meta expands its GPU cloud infrastructure partnership with CoreWeave to $21 billion, the largest hyperscaler AI commitment to a single provider.

Meta expanded its agreement with CoreWeave to $21 billion for AI compute capacity, running through 2032, according to a Thursday, April 9 report from Bloomberg. The deal builds on a prior multibillion-dollar contract between the companies. CoreWeave builds its infrastructure around chips from Nvidia and provides Meta access to large-scale GPU clusters used to train and run artificial intelligence models. These systems support both training and inference workloads.

Meta has increased spending on AI infrastructure as it scales its models and products, outlining plans to spend more than $100 billion on AI infrastructure this year. The CoreWeave contract secures supply at a time when demand for GPUs exceeds available capacity. Demand for AI compute has driven the growth of specialized cloud providers often called neoclouds, which focus on AI workloads rather than general-purpose computing and offer direct access to high-performance chips and optimized infrastructure. CoreWeave is one of the largest providers in this group, according to PYMNTS.

CoreWeave has expanded its position through partnerships and acquisitions, acquiring Monolith to extend its cloud into industrial design workloads. The company has attracted investment tied to infrastructure buildout, with Nvidia committing $2 billion to CoreWeave to support data center expansion. CoreWeave's business model depends on long-term contracts with large customers, with Meta now one of its largest clients alongside Microsoft. These agreements provide predictable revenue and support continued investment in data centers and hardware. CoreWeave plans to spend tens of billions of dollars to expand its infrastructure, relying on debt markets to fund that growth.

AI workloads are shifting toward inference, where models run inside products after training, requiring sustained access to compute rather than one-time training capacity. Companies are securing infrastructure to support ongoing usage as AI systems move into production. Meta is also expanding its AI model lineup, having introduced Muse Spark, a new large language model aimed at consumer and multimodal use cases, as reported by PYMNTS.