OpenAI has announced it signed a $10 billion deal with Cerebras Systems for about 750 megawatts of AI computing power, marking a major milestone for the chip startup in a market long dominated by Nvidia and the next step towards faster AI.
Cerebras was founded in 2015 by Andrew Feldman and former colleagues from SeaMicro, shortly after selling their previous company to AMD for $334 million. Rather than playing it safe, the team went on to bet big – building a single, dinner-plate-sized processor designed to handle massive AI workloads faster and more efficiently than GPUs.
For years, that idea was met with skepticism. The hardware was expensive, the engineering challenges enormous, and critics often dismissed the project as an impressive science experiment rather than a real business.
But the vision behind Cerebras powered through, raising more than $1 billion and landing a handful of large customers, including G42, which accounted for a significant share of its early revenue.
The bet that finally paid off
The shift came when Cerebras proved its technology could run large AI models in real time, with lower delay and lower cost per token. Tests revealed that open-source models ran more efficiently on Cerebras’ chips, especially for inference – the stage where AI systems generate answers quickly.
That proof is what caught OpenAI’s attention.
Under the deal, Cerebras will help power OpenAI’s AI systems by running data centers filled with its own custom chips, as the big draw for OpenAI is speed. New AI models don’t just answer right away – they often pause to “think,” and Cerebras’ newest chip’s design will help them do that faster by avoiding the slowdowns that come with stitching together lots of smaller chips.
The deal also changes Cerebras’ trajectory, reducing the company’s dependence on earlier overseas partners and raising its profile across the AI industry on its path toward a potential public offering.
