OpenAI and Cerebras confirm 750MW AI inference deployment by 2028

OpenAI and Cerebras confirmed a multi-year partnership to deploy 750MW of Cerebras wafer-scale compute for ultra-low-latency inference on OpenAI's platform. A Master Relationship Agreement effective December 24, 2025 splits capacity into three 250MW segments through the end of 2028.

• 250MW by end of 2026, 500MW by end of 2027, 750MW by end of 2028
• Capacity is for inference, not model training
• Agreement includes possible extra capacity and a hardware purchase path
• An OpenAI Codex Spark model on Cerebras infrastructure appeared around February 12, 2026