In brief: Meta has committed an additional $21 billion to CoreWeave for dedicated AI cloud capacity from 2027 to December 2032, bringing the total value of the two companies’ infrastructure relationship to approximately $35 billion. The new contract will deliver early deployments of Nvidia’s Vera Rubin platform at multiple sites and is designed specifically for inference rather than training workloads. Along with the announcement, CoreWeave revealed plans to raise $4.25 billion in new debt, $3 billion in convertible notes and $1.25 billion in junk bonds, to fund continued expansion. CoreWeave shares rose about 5% on the news; Meta shares gained about 3%.
From Ethereum mining to a goal relationship of 35 billion dollars
CoreWeave was founded in 2017 in New Jersey as Atlantic Crypto, a commodity trading side project that mines Ethereum using graphics processing units. When the 2018 cryptocurrency crisis made mining uneconomical and Ethereum’s eventual move to proof-of-stake threatened to make GPU mining completely obsolete, the founders, Michael Intrator, Brian Venturo, and Brannin McBee, recognized that the GPU inventory they had amassed was also exactly what machine learning researchers needed and couldn’t easily access through conventional cloud providers. The company was renamed CoreWeave in 2019 and moved into GPU cloud infrastructure. It went public on March 28, 2025 at $40 per share, valuing it at $23 billion. Its revenue in 2025 reached $5.13 billion, a year-on-year increase of 168%, and its order book is estimated at more than $66 billion. The first Meta deal, valued at $14.2 billion and announced in September 2025, was the deal that established CoreWeave as a serious counterpart to hyperscale cloud providers. The April 9, 2026, additional $21 billion expansion makes Meta the largest commercial relationship in CoreWeave’s history, with a combined commitment that will sustain the company’s revenue base through the end of the decade.
What Meta is really buying
The contract is specifically structured around inference rather than formation. The Llama de Meta model family is open weight and free to download, meaning the capital-intensive training phase is virtually complete before any cloud contract is signed; the ongoing cost is serving those models to billions of users in real time. Inference at Meta’s scale—hundreds of millions of daily active users across Facebook, Instagram, WhatsApp, and Meta AI—requires sustained, low-latency computing across distributed infrastructure in a way that Meta’s own data centers can’t always absorb at full capacity. CoreWeave will deploy that capability in multiple locations and will include some of the first commercial implementations of Nvidia’s Vera Rubin platform, which the chipmaker unveiled at GTC 2026 in March as the next generation of its AI infrastructure hardware.. The new agreement complements, rather than replaces, Meta’s internal development. Meta has forecast between $115 billion and $135 billion in capital expenditures by 2026, identifying AI infrastructure as the main driver, and the company has been explicit that it is building its own data centers and sourcing external capacity simultaneously. CoreWeave’s expansion follows a $27 billion infrastructure deal Meta signed with Nebius in March 2026.under which the Dutch neocloud operator will supply dedicated computing starting in early 2027, and will also include early implementations of Vera Rubin. The two deals together illustrate that Meta is not just acquiring cloud capacity, but is building a diversified, multi-vendor infrastructure position designed to give you flexibility and redundancy at hyperscale.
The customer diversification game
For CoreWeave, Meta’s expansion solves a problem that has plagued the company since its IPO: excessive revenue concentration. Microsoft accounted for 62% of CoreWeave’s revenue in 2024, a figure that made institutional investors uncomfortable and which the company has been working to reduce. With the new Target commitment in place, CoreWeave CEO Michael Intrator said no customer would account for more than 35% of total sales. It’s still a significant concentration, but it’s a materially different risk profile than a position where a single hyperscale customer controls the majority of your revenue. Nvidia, which made a $2 billion strategic investment in Nebius in March 2026 and has deepened its business relationships with all major AI cloud providers.lies at the center of CoreWeave’s business model: the entire CoreWeave infrastructure is based on Nvidia GPUs, and Vera Rubin’s implementations in the Meta contract will extend that dependence to the next generation of hardware. CoreWeave also recently expanded its deal with OpenAI by up to $6.5 billion, further expanding its customer base beyond Microsoft. The company’s shares hit an all-time high of $187 in mid-2025 before retreating to around $65 in late 2025 amid broader concerns about returns on AI investments; Following the announcement of the Meta expansion, it was trading in the range of $88 to $95.
The debt that finances everything
The 💜 of EU technology
The latest rumors from the EU tech scene, a story from our wise founder Boris and some questionable AI art. It’s free, every week, in your inbox. Register now!
Cloud AI infrastructure is expensive to build before contracts start generating revenue, and CoreWeave has funded its growth primarily through debt. Along with the announcement of the Meta deal, the company revealed plans to raise $4.25 billion in new financing: $3 billion in convertible senior notes due 2032, with a coupon of between 1.5% and 2%, with an option for investors to convert them into shares; and $1.25 billion in senior unsecured bonds due 2031 at about 10%, effectively junk bond prices. CoreWeave’s total debt load is around $30 billion, about triple what it was a year earlier. The company’s argument in favor of the debt structure is that its contracted revenue base, more than $66 billion in portfolio, provides sufficient visibility to meet obligations. Intrator has described CoreWeave as an “AI factory” whose capital costs are covered by long-term customer commitments before the infrastructure is built. The broader AI infrastructure financing environment has been characterized by similarly large-scale debt structures: SoftBank took out a $40 billion bridge loan to finance its $30 billion follow-on investment in OpenAI as part of the Stargate project.illustrating that the capital requirements of AI at scale are now large enough to require financing instruments that did not exist in this form even two years ago. The year 2025 consolidated AI infrastructure as the main competitive variable in the technology industry.and CoreWeave, a company that started as a closet of Ethereum mining rigs, has positioned itself as a loading pillar of that infrastructure, with a $21 billion commitment at a time.
