OpenAI will spend more than $20 billion on Cerebras Systems’ wafer-scale chips over the next three years, according to Cerebras’ S-1 filing with the SEC on April 17. The deal includes an equity stake for OpenAI and a $1 billion loan to fund data center buildout, making it the largest single compute commitment OpenAI has made outside its relationship with Microsoft.

The agreement calls for Cerebras to deliver 250 megawatts of computing power per year from 2026 through 2028, totaling 750 megawatts in the initial phase. OpenAI holds options to purchase an additional 1.25 gigawatts through 2030, according to CNBC’s reporting on the filing. In December, Cerebras issued OpenAI warrants to purchase up to 33.4 million shares of non-voting Class N stock. The warrants vest fully only if OpenAI buys 2 gigawatts of computing power.

The IPO Filing

Cerebras filed to go public on Nasdaq under the ticker “CBRS” on April 17, its second attempt after withdrawing a September 2024 filing. The company reported $510 million in revenue for 2025, up 76% from the prior year, with $87.9 million in net income, a sharp reversal from a $485 million net loss in 2024. Cerebras listed $24.6 billion in remaining performance obligations as of December 31, expecting to recognize 15% of that sum in 2026 and 2027, CNBC reported.

The target valuation is approximately $35 billion, up from the $23 billion valuation Cerebras achieved in its $1 billion Series H round in February 2026, according to The Next Platform’s analysis of the S-1. That follows a $1.1 billion Series G in October 2025 at an $8.1 billion valuation. Cerebras also closed an $850 million revolving credit facility with Morgan Stanley, Citi, Barclays, and others as joint lead arrangers, CWEB reported.

Customer Concentration and the UAE Question

Revenue concentration remains Cerebras’ most visible risk. In 2025, the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), a UAE government institution, accounted for 62% of revenue, while G42, another Abu Dhabi entity, contributed 24%, according to the S-1 filing. That means 86% of 2025 revenue came from two customers pulling from overlapping pools of Gulf sovereign capital, as The Next Platform noted.

The OpenAI deal changes that picture going forward. So does an Amazon partnership signed in March that allows AWS to offer cloud services on Cerebras hardware, with Amazon purchasing approximately $270 million in Class N stock, according to CNBC.

The Compute Diversification Bet

Cerebras’ core architectural bet is a single wafer-scale chip rather than clusters of thousands of smaller GPUs. That design trades manufacturing complexity for a memory bandwidth-to-compute ratio that favors low-latency inference, the kind of workload that matters most as AI shifts from chatbots to autonomous agents executing multi-step tasks in real time.

OpenAI’s commitment signals that the company sees a future where NVIDIA is not the only viable compute supplier at scale. With $20 billion flowing to Cerebras, separate custom ASIC development with Broadcom, and ongoing cloud commitments with Microsoft Azure, OpenAI is building a multi-vendor compute strategy designed to avoid single-supplier dependency as it scales agentic AI workloads into the hundreds of billions of tokens per day.

Cerebras described the OpenAI alliance as representing “a substantial portion of our projected revenues over the next several years” in its S-1. OpenAI retains the right to terminate part or all of the agreement if Cerebras fails to deliver computing power on schedule or falls below performance thresholds.