OpenAI–Cerebras Deal: 750 MW of Staged Inference Capacity

Signed capacity, staged deployment

OpenAI and Cerebras signed a multi-year agreement for 750 megawatts of wafer-scale inference systems. Cerebras said rollout would begin in stages in 2026. A signed commitment does not establish that all systems were available that day. [1]

OpenAI described the capacity as a low-latency complement to its existing compute portfolio. Its announcement scheduled deployment in stages through 2028, connecting the agreement to faster model responses rather than a completed replacement of other infrastructure. [4]

Price reporting and evidence limits

The Register reported a value above $10 billion, citing sources familiar with the deal. That figure is a reported estimate, not a price disclosed in the companies’ announcements. The announcement record establishes planned capacity and timing, not full delivery or independently measured performance across customer workloads. [3][1][4]