Brief: CoreWeave announced a multi-year deal with Anthropic on April 10, 2026, providing Nvidia GPU capacity across US data centers for AI workloads. Financial details were undisclosed. The agreement follows CoreWeave’s $21 billion expansion with Meta and brings Anthropic into a client base that covers nine of the top ten AI model providers. CoreWeave’s 2025 revenue was $5.13 billion, projected to exceed $12 billion in 2026, supported by a backlog beyond $66 billion.
Ex-crypto miner pivots to AI landlord role
CoreWeave, founded in 2017 as Atlantic Crypto, initially operated as an Ethereum mining entity, purchasing Nvidia GPUs in mass to mine cryptocurrency and rent spare capacity. As crypto margins shrank in 2019, it transitioned to CoreWeave, focusing on GPU cloud services. This shift coincided with a surge in AI model training starting in 2023, elevating CoreWeave’s Nvidia hardware into a strategically crucial infrastructure asset. The company went public on Nasdaq with the ticker CRWV on March 28, 2025, at $40 per share, securing $1.5 billion and a valuation of about $23 billion. CoreWeave runs 32 data centers with over 250,000 GPUs and 1.3 gigawatts of power capacity. Its 2025 revenue of $5.13 billion was a 168% increase from the previous year, with projections for over $12 billion in 2026 revenue and a backlog exceeding $66 billion.
CoreWeave’s growth involves a concentration risk, with Microsoft comprising about 67% of its 2025 revenue. Analysts noted this leading up to the IPO, Microsoft’s AI model development introduces another strategic factor, questioning if Microsoft’s computing needs might move in-house rather than relying on third-party GPU cloud rentals. The Anthropic deal, following the $21 billion Meta expansion, is CoreWeave’s apparent move to diversify and reduce reliance on any single hyperscaler.
Anthropic’s strategic investment
Anthropic’s compute strategy complexity has increased with its revenue growth. Its annual revenue run rate exceeded $30 billion in early April 2026, up from $9 billion in late 2025. This growth, fueled by enterprise Claude adoption and Claude Code’s rise, necessitated expanding infrastructure across various chip architectures. Primarily, workloads run on AWS Trainium hardware through Project Rainier, a vast cluster of AI chips across US data centers. Just days prior to the CoreWeave agreement,Anthropic’s Google and Broadcom deal secured about 3.5 gigawatts of tensor processing unit compute for 2027. CoreWeave offers a third option: Nvidia GPU capacity for enterprise-level Claude deployments. Anthropic’s $100 million Claude partner network investment earlier this year hinted at expanding developer and enterprise ecosystems, which is now impacting compute procurement strategies like this one.
CoreWeave CEO Michael Intrator presented the deal as more than mere infrastructure. “AI now involves platforms turning models into real-world effects,” he said. “We’re thrilled to work with Anthropic where model performance materializes. It’s precisely the deployment scenario CoreWeave was built for.” Anthropic’s compute deployment follows a phased infrastructure approach, with expansion options. The specific Nvidia chip architectures remain undisclosed, but CoreWeave