OpenAI has signed a major agreement with artificial intelligence (AI) chip startup Cerebras Systems valued at more than $10 billion, one of the largest deals in the AI space as companies scramble to secure computing resources.
The ChatGPT maker will purchase up to 750 megawatts of computing capacity over three years from Cerebras, whose specialized chips promise to deliver AI responses significantly faster than traditional processors. The partnership addresses OpenAI’s critical need for computing power as it serves more than 900 million weekly users while facing what executives describe as a severe capacity shortage.
The deal, first reported by the Wall Street Journal, is layered with intrigue: OpenAI CEO Sam Altman personally invested in Cerebras years ago, and the companies first explored collaboration in 2017. Cerebras CEO Andrew Feldman confirmed that recent negotiations began last fall, with a term sheet signed by Thanksgiving.
Cerebras designs dinner plate-sized processors called Wafer-Scale Engines that can run AI models up to 15 times faster than conventional GPU-based systems, according to company claims. This speed advantage proved decisive for OpenAI, particularly for applications like coding assistance where rapid responses enhance user experience.
“The biggest predictor of OpenAI revenue is how much compute is there,” Sachin Katti, an OpenAI infrastructure executive, told the Journal. He noted the company has tripled both its computing capacity and revenue annually over the past two years.
The partnership is a crucial win for Cerebras, which has struggled to challenge NVIDIA Corp.’s dominance in the AI chip market. When Cerebras filed to go public in 2024, it said 87% of its revenue came from a single customer, UAE-based AI firm G42. The company ultimately withdrew its IPO plans and raised $1.1 billion privately instead.
Mitch Ashley, vice president and practice lead, DevOps and AppDev, at the Futurum Group, said the deal “signals a move from opportunistic GPU sourcing to long-term capacity locking as a control point for AI platforms. The scale of the agreement reflects how acute the compute constraint has become when serving hundreds of millions of active users.”
“More broadly, this points to a structural change in the AI hardware/model stack where access to specialized silicon and power becomes a competitive differentiator, rather than an implementation detail,” Ashley said. “As model usage scales and response latency increasingly matters, vendors like OpenAI will increasingly couple model innovation with infrastructure strategies, while players such as Cerebras gain leverage by offering purpose-built alternatives to traditional GPU supply chains.”
Cerebras recently signed deals with IBM Corp. and Meta Platforms Inc., diversifying its customer base. Reports indicate Cerebras is currently seeking $1 billion in new funding at a $22 billion valuation, nearly triple its previous valuation of $8.1 billion.
For OpenAI, the Cerebras deal is part of a broader strategy to reduce dependence on NVIDIA’s pricey chips while securing sufficient computing resources. The company is simultaneously developing a custom chip with Broadcom Inc. and has agreed to use AMD Inc.’s forthcoming MI450 processors.
However, concerns persist about OpenAI’s ability to finance its computing commitments. Despite generating approximately $13 billion in revenue last year, the company has signed nearly $600 billion in cloud contracts with Oracle Corp., Microsoft, and Amazon.com Inc. Altman maintains these phased agreements will be covered by future revenue growth.
OpenAI is reportedly pursuing another massive fundraising round that could value the company at $830 billion before new investment, with reports suggesting it may seek at least $10 billion from Amazon. The computing capacity from Cerebras will begin rolling out in 2026, with full implementation expected by 2028.
