Meta, CoreWeave Seal $21 Billion AI Cloud Pact

Meta Platforms and CoreWeave have expanded their artificial intelligence infrastructure partnership with a new agreement valued at about $21 billion, running through December 2032. CoreWeave will supply dedicated AI cloud capacity to support Meta’s large-scale AI workloads, including inference tasks that run trained models at scale.

The latest agreement builds on a prior AI infrastructure deal signed in 2025, which was worth up to about $14 billion to $14.2 billion and extended through 2031. With the new commitment, Meta’s total contracted spend with CoreWeave now stands at approximately $35.2 billion, making it one of the largest single-customer infrastructure commitments reported in the current AI buildout.

Infrastructure under the deal will be deployed across multiple data center locations and is designed to optimize performance, resilience, and scalability for Meta’s AI systems. The rollout will include some of the first deployments of Nvidia’s (NVDA) Vera Rubin platform, giving Meta access to advanced GPU infrastructure for its next-generation models.

CoreWeave said the expanded Meta contract significantly increases its revenue visibility. The company’s total revenue backlog has risen to $66.8 billion, representing 342% growth from the start of 2025. CoreWeave generated $5.13 billion in revenue in 2025 and has guided for $12 billion to $13 billion in 2026, implying year-over-year growth of 134% to 153%.

The Meta win adds to a customer roster that includes Microsoft (MSFT), OpenAI, IBM, and Mistral AI. With the new contract, CoreWeave said no single customer will account for more than 35% of its sales, reducing its prior reliance on Microsoft (MSFT) for more than half of revenue. The company previously expanded its OpenAI agreement to about $22.4 billion.

To fund its aggressive build-out, CoreWeave is planning capital expenditures of $30 billion to $35 billion in 2026, up from $14.9 billion in 2025. The company announced a proposed $3 billion private offering of convertible senior notes due 2032 and carries $21 billion of long-term debt as of the end of 2025, alongside an $8.5 billion term loan facility closed in March 2026.

CoreWeave shares were volatile on the news. Reports noted the stock spiked as much as 8% in premarket trading before pulling back after the debt announcement. Around midday, shares were recently up between about 3.7% and 5.5%, trading in the low-$90 range. One report also cited intraday swings that briefly pushed the stock into negative territory before it recovered.

Analysts highlighted that CoreWeave’s model is highly capital intensive, with management projecting roughly $2.60 in capital spending for every $1 of new revenue in 2026. The company reported a 47.77% gross margin and adjusted EBITDA margin of 43% for 2025, but free cash flow remains negative and net losses widened to $1.167 billion in 2025.

Despite these pressures, commentators described the Meta contract as a foundational agreement that converts a significant portion of CoreWeave’s growth into contracted cash flows through 2032. CEO Michael Intrator said the deal is another example of leading companies choosing CoreWeave’s AI cloud for their most demanding workloads.

The expanded CoreWeave partnership coincides with Meta’s broader push into frontier AI systems. On April 8, Meta introduced Muse Spark, a natively multimodal large language model capable of processing text, images, voice, and video, and designed to tackle complex reasoning tasks using multiple AI agents.

Muse Spark now powers Meta’s digital assistant in a dedicated app and desktop site, with planned rollouts across Facebook, Instagram, WhatsApp, Messenger, and Ray-Ban Meta AI glasses. Meta has also published an Advanced AI Scaling Framework to guide risk evaluation and testing for its most capable models before deployment.

Meta has signaled substantial capital commitments behind this strategy, stating that AI capital expenditures in 2026 are expected to total between $115 billion and $135 billion. The company said that as it builds more capable and personalized AI, reliability, security, and user protections are increasingly important, requiring an “advanced approach to safety” that scales with the technology.