OpenAI bets on Cerebras

OpenAI has agreed to spend more than $20 billion over three years on Cerebras-powered server capacity, widening its chip supply beyond Nvidia-heavy systems. (reuters.com) Reuters reported on April 16 that the arrangement could also give OpenAI warrants for a minority stake in Cerebras, with ownership rising as spending rises. The report said OpenAI may also provide about $1 billion to help fund data centers running Cerebras hardware. (reuters.com) The same Reuters report said OpenAI had already agreed in January to buy up to 750 megawatts of Cerebras computing capacity over three years in a deal worth more than $10 billion. The new commitment would roughly double that previously reported agreement. (reuters.com) What OpenAI is buying is inference capacity: the computing used after a model is trained, when it generates answers for users. Reuters said the deal reflects surging demand for inference as OpenAI tries to keep up with usage and push new products into market. (reuters.com) Cerebras sells a different kind of artificial-intelligence processor: one giant chip built from an entire silicon wafer instead of many smaller chips linked together. The company says its Wafer-Scale Engine 3 has 4 trillion transistors and is its flagship processor for training and inference. (cerebras.ai) Cerebras says the Wafer-Scale Engine 3 is 58 times larger than a leading graphics processing unit chip and can deliver inference up to 15 times faster on its benchmarks, while using less power per unit of compute. The company adds that performance can vary by workload, configuration, date, and model. (cerebras.ai) The timing lines up with Cerebras returning to the public markets. On April 17, the company said it filed a new Form S-1 for an initial public offering and plans to list on Nasdaq under the ticker CBRS. (cerebras.ai) CNBC reported from that filing that Cerebras revenue grew nearly 76% in 2025, and that OpenAI had loaned the company $1 billion and received a warrant to buy stock. Reuters separately reported that Cerebras’ tie-up with OpenAI is central to its push to go public. (cnbc.com) (reuters.com) OpenAI has been building out compute through several routes at once, not just one supplier. In January 2025, it announced Stargate, a project that said it intended to invest $500 billion over four years in United States artificial-intelligence infrastructure, and in July 2025 OpenAI said an Oracle partnership brought Stargate capacity under development to more than 5 gigawatts and over 2 million chips. (openai.com 1) (openai.com 2) Cerebras has also been lining up cloud partners beyond OpenAI. In March 2026, Amazon Web Services and Cerebras said they would combine Amazon Trainium servers for one stage of inference with Cerebras CS-3 systems for another, with deployment planned in Amazon Bedrock data centers. (cerebras.ai) OpenAI and Cerebras did not confirm the reported terms to Reuters, and Reuters said it could not independently verify the full arrangement. But if the spending, financing, and warrant pieces hold, OpenAI is not just renting more compute — it is tying part of its infrastructure expansion to a chipmaker now trying to go public. (reuters.com)