CoreWeave and Meta Expand AI Cloud Partnership to $35 Billion With New $21 Billion Infrastructure Agreement
CoreWeave secures a $21 billion expansion of its Meta partnership through 2032, bringing total commitments to $35.2 billion and reducing customer concentration risk ahead of NVIDIA Vera Rubin deployments.
Overview
CoreWeave and Meta announced on April 9, 2026, an expanded long-term agreement worth approximately $21 billion to provide dedicated AI cloud capacity through December 2032. The deal raises the total value of Meta’s commitments to the specialized cloud provider to roughly $35.2 billion, up from the prior $14.2 billion arrangement, according to BusinessWire.
The expansion marks a significant milestone for CoreWeave’s post-IPO trajectory and sends a clear signal about the scale of infrastructure investment that major technology companies consider necessary to support their AI ambitions.
What We Know
The dedicated capacity will be deployed across multiple locations and will include some of the initial deployments of the NVIDIA Vera Rubin platform, CoreWeave’s next-generation GPU architecture designed to optimize performance, resilience, and scalability for large-scale AI workloads, according to BusinessWire.
Michael Intrator, CoreWeave’s co-founder and CEO, described the agreement as further validation of the company’s position in the AI cloud market. “This is another example that leading companies are choosing CoreWeave’s AI cloud to run their most demanding workloads,” Intrator stated, as reported by BusinessWire.
The contract is strategically important for CoreWeave’s financial profile. Microsoft represented 62 percent of CoreWeave’s revenue in 2024, a concentration risk that investors flagged during the company’s March 2025 IPO. With the expanded Meta commitment, no single customer will account for more than 35 percent of total sales, according to CoinDesk. Meta now represents roughly 40 percent of CoreWeave’s pro-forma backlog, which stands at approximately $87.8 billion.
Meta has projected AI infrastructure spending of $115 billion to $135 billion in 2026, underscoring the company’s view that securing compute capacity is essential to its generative AI strategy, as reported by PYMNTS. The CoreWeave partnership helps Meta supplement its own data center buildout by providing flexible access to GPU clusters optimized for both training and inference workloads.
CoreWeave shares (CRWV) rose following the announcement, reflecting investor confidence in the company’s expanding revenue visibility. The deal also follows a separate multiyear agreement CoreWeave signed with Anthropic to power the Claude family of AI models, further broadening the company’s customer base, according to CoinDesk.
What We Don’t Know
Neither company disclosed the specific geographic locations where the new capacity will be deployed, nor the exact number of data centers involved. The financial terms beyond the headline $21 billion figure remain undisclosed, including pricing per GPU-hour or any volume discount structures.
It is also unclear how much of the commitment is tied specifically to NVIDIA Vera Rubin hardware versus existing GPU generations. The shift from training-heavy workloads toward inference could alter the hardware mix over the contract’s six-year duration, but neither party has addressed how the agreement accounts for evolving technology requirements.
Analysis
The deal reflects a broader industry pattern in which hyperscalers and major technology companies are turning to specialized “neocloud” providers rather than relying exclusively on traditional cloud giants. CoreWeave, which went public just over a year ago, has positioned itself as a purpose-built alternative to AWS, Azure, and Google Cloud for AI-specific workloads, offering dense GPU clusters with lower-latency interconnects optimized for large model training and inference.
The customer diversification achieved through this deal addresses one of the most persistent criticisms of CoreWeave’s business model. Moving from 62 percent Microsoft revenue concentration to a cap of 35 percent per customer within roughly two years represents a meaningful reduction in single-client risk.
For Meta, the agreement signals that even companies building massive internal data center networks still see value in maintaining relationships with external compute providers. With AI workloads increasingly shifting from one-time model training to sustained inference serving, the demand for GPU capacity is becoming more predictable and long-term, making multiyear contracts a practical approach to capacity planning.