OpenAI signs $10B Cerebras deal

OpenAI said on January 14 that it had signed a multiyear agreement with Cerebras Systems to add 750 megawatts of AI compute, a deal that CNBC, Reuters and Bloomberg reported was worth more than $10 billion. OpenAI described the partnership as a way to add high-speed inference capacity for its products, while Cerebras said the deployment would roll out in stages beginning in 2026. The announcement did not frame the arrangement as a replacement for Nvidia; it added another infrastructure supplier as OpenAI expands capacity for ChatGPT and related services. ### When was this actually announced? January 14, 2026 was the public disclosure date. OpenAI published a post that day saying it was partnering with Cerebras, and Cerebras published a matching announcement on its site. Reuters reported the same day that OpenAI would purchase up to 750 megawatts of computing power over three years from Cerebras. May 23, 2026 social posts appear to have resurfaced the deal rather than broken it. (openai.com) The primary public record available online points to the January 14 company announcements and same-day media reports, not a new May 23 filing or press release. ### What exactly did OpenAI buy from Cerebras? OpenAI said the agreement would add 750 megawatts of high-speed AI compute. (openai.com) Cerebras said the companies had signed a multiyear agreement to deploy 750 megawatts of its wafer-scale systems to serve OpenAI customers, with capacity coming online in multiple stages through 2028. Reuters reported that OpenAI planned to use Cerebras systems to power its chatbot products. (openai.com) OpenAI’s own post said the partnership was aimed at reducing inference latency and making ChatGPT faster for real-time AI workloads. ### Where does the $10 billion figure come from? (openai.com) CNBC reported on January 14 that the arrangement was worth over $10 billion, citing people close to the company. Reuters reported the same day that the deal was worth more than $10 billion over the life of the contract, citing a source familiar with the matter. Bloomberg also reported a multiyear OpenAI-Cerebras deal for 750 megawatts of computing power. (money.usnews.com) OpenAI’s public announcement did not state a dollar value. Cerebras’ blog post also described the scale and rollout timetable without publishing a contract price. ### Is this about training, inference, or both? OpenAI’s announcement emphasized inference. The company said the Cerebras partnership would add high-speed compute to make ChatGPT faster for real-time workloads, and Cerebras called it a deployment of wafer-scale systems for OpenAI customers. (cnbc.com) Reuters said Cerebras Chief Executive Andrew Feldman told the news agency the companies began talks after Cerebras showed OpenAI’s open-source models could run more efficiently on its chips than on traditional GPUs. (openai.com) That description also points to serving model execution and response workloads, rather than a deal described primarily around training. ### Why does Cerebras matter to OpenAI’s supplier mix? Cerebras is positioning itself as an alternative AI hardware supplier to the Nvidia-centered market. Reuters described Cerebras as an Nvidia challenger, and CNBC said the OpenAI contract would help Cerebras diversify away from G42, which had accounted for most of its earlier revenue. (money.usnews.com) Andrew Feldman, Cerebras’ co-founder and chief executive, told CNBC the company was building a broader customer base as it pursued growth. OpenAI, for its part, has been signing multiple large infrastructure agreements as demand for AI compute rises. ### What should readers watch next? Through 2028, the main milestones are the staged rollout of Cerebras capacity and any further disclosures from OpenAI on how much of that compute is live. (money.usnews.com) Cerebras’ next formal detail is also likely to come through IPO-related filings and investor materials, where customer concentration, revenue mix and contract structure could be described in more detail. (cerebras.ai) (cnbc.com)