• OpenAI has signed a five-year, $11.9 billion agreement with CoreWeave, including $350 million in equity.
  • CoreWeave, backed by Nvidia, operates 32 data centers with over 250,000 GPUs and reported $1.9 billion in revenue in 2024, up from $228.9 million in 2023.

In a high-stakes move that reshapes the AI cloud landscape, OpenAI has signed a five-year, $11.9 billion agreement with CoreWeave, a GPU-intensive cloud provider. The deal includes OpenAI receiving $350 million in CoreWeave equity, though this private placement is separate from CoreWeave’s upcoming IPO.

CoreWeave, which filed for an IPO last week but has yet to price or schedule its public debut, has seen massive revenue growth. In 2024, the company reported $1.9 billion in revenue, an almost eightfold jump from $228.9 million in 2023. A key driver of this surge was Microsoft, which accounted for 62% of CoreWeave’s revenue last year.

By securing OpenAI as a direct customer, CoreWeave reduces its dependence on a single client—an aspect that could bolster investor confidence ahead of its public listing.

Backed by Nvidia, which owns a 6% stake, CoreWeave operates an AI-focused cloud infrastructure across 32 data centers with over 250,000 Nvidia GPUs as of late 2024. The company has since expanded its fleet, integrating Nvidia’s Blackwell chips designed for AI reasoning.

This deal highlights the evolving dynamic between OpenAI and Microsoft. While Microsoft remains a major backer of OpenAI, tensions have emerged as both companies compete for AI dominance.

Microsoft has been developing its own AI models, including the MAI family, and recently hired Mustafa Suleyman, co-founder of DeepMind, to lead its AI division. Meanwhile, OpenAI has been pushing to secure more computing power, with CEO Sam Altman recently stating that the company is “out of GPUs.”

Originally a crypto-mining operation, CoreWeave has transformed into a key player in AI cloud computing. As it eyes billions in IPO funding, the company must also navigate its staggering $7.9 billion debt load.

Edited by Harshajit Sarmah