CoreWeave expands its Meta AI partnership to $21 billion and raises $4.25 billion in new convertible and unsecured debt.

CoreWeave has expanded its agreement to supply Meta with AI computing capacity, lifting the total value of the deal to $21 billion, as reported by Bloomberg. This builds directly on the $14.2 billion pact the companies struck last September, which originally ran through 2031 with an option for extension. The updated terms extend AI cloud services through December 2032, with the additional capacity coming from multiple data centers equipped in part with Nvidia's next-generation Rubin AI chip systems. The move gives Meta more assured access to specialized GPU clusters as it scales training and inference workloads for its expanding lineup of large language models. CoreWeave now holds $35 billion in contracts with Meta, making the tech firm one of CoreWeave's largest customers.

CoreWeave has dramatically ramped up borrowing in recent years to finance deals in which it rents access to high-end artificial intelligence processors. The company's debt load has grown to around $30 billion, triple what it was a year earlier. To fund its capital-intensive expansion, CoreWeave plans to offer $3 billion in convertible senior notes due 2032 with a 1.5% to 2% coupon that investors can choose to convert into stock at a premium, alongside $1.25 billion in senior notes due 2031 at just above 10%. In February, the company had been seeking about $8.5 billion from banks including Morgan Stanley and Mitsubishi UFJ Financial Group Inc. to help finance its buildout of cloud computing capacity for Meta.

CoreWeave, described as a "cash-incinerating provider of GPU-accelerated cloud computing," has positioned itself as a neocloud alternative to traditional hyperscalers, serving clients including Microsoft and OpenAI, while competing with smaller rivals like Nebius and Nscale. The company has benefited from Nvidia's backing, including an additional $2 billion investment to speed construction of new AI factories. Meanwhile, Meta has emerged as one of the top spenders on AI infrastructure, with CEO Mark Zuckerberg planning to spend hundreds of billions of dollars over the next few years on the energy, computing power and talent needed to build, train and run AI models. In its latest earnings call, Meta raised its 2026 capex projections to $115-$135 billion, nearly doubling its 2025 capex spend. The deal underscores a broader truth in the current cycle: hyperscalers are willing to lock in enormous, multi-year contracts to guarantee scarce high-performance computing resources, with contracts continuing to materialize to feed exponential growth in demand.