OpenAI signs $12B five-year deal with CoreWeave, gets $350M equity stake ahead of IPO
Mar 11, 2025
Key Points
- OpenAI commits $12 billion over five years to rent AI servers from CoreWeave and receives a $350 million equity stake ahead of the cloud provider's IPO.
- The deal cuts Microsoft out as intermediary, giving OpenAI direct compute access and reducing CoreWeave's revenue concentration risk that could derail its public offering.
- Large AI companies are increasingly taking equity stakes in infrastructure vendors when committing to multi-billion-dollar consumption agreements, aligning incentives as annual capex spending approaches $250 billion.
Summary
OpenAI has signed a five-year, $12 billion deal to rent AI servers from CoreWeave, the Nvidia-backed cloud infrastructure provider filing to go public. OpenAI receives an equity stake in CoreWeave expected to be worth approximately $350 million at IPO.
OpenAI is establishing its own independent compute pipeline rather than relying solely on Microsoft as an intermediary. Previously, OpenAI could have accessed CoreWeave's capacity indirectly through Microsoft's partnership, which accounted for 62% of CoreWeave's revenue in 2024.
The timing addresses a key investor concern for CoreWeave's IPO filing: revenue concentration risk tied to Microsoft. The OpenAI deal materially diversifies that revenue base and reduces the single-customer dependency that would otherwise dominate the S-1 narrative. In return, OpenAI gets a meaningful equity upside and contractual compute commitment.
As the industry commits to roughly $250 billion in annual capex spending, large AI companies are increasingly securing equity stakes in infrastructure vendors when committing to multi-billion-dollar consumption agreements. This approach aligns vendor success with customer success and provides exposure to upside when vendors go public.
The Financial Times reported that Microsoft cut back some spending plans with CoreWeave due to delivery issues and missed deadlines. CoreWeave denied the report, saying there have been no contract cancellations or commitments walking away. The discrepancy underscores execution risk at the infrastructure layer, even as demand for AI compute accelerates.