The Speed of Thought: OpenAI Inks $10B Deal for 15x Faster AI Responses

OpenAI and the American AI semiconductor unicorn Cerebras have formally announced a landmark three-year strategic alliance under which OpenAI will deploy a staggering 750MW of Cerebras’ Wafer-Scale Systems, in a move described as designed to construct the world’s most expansive high-speed AI inference platform. Sources familiar with the matter indicate the transaction is valued in excess of $10 billion, with deployment slated to commence in 2026 and complete by 2028 through several developmental phases.

Cerebras’ architecture diverges from traditional GPU designs, and its wafer-scale approach aims to remove bottlenecks that typically impede inference velocity, claiming in terms of large language model responsiveness that the Cerebras system outperforms GPU-based alternatives by a factor of fifteen. OpenAI has previously disclosed a broader strategy to diversify its computational supply chain, including partnerships with Broadcom for custom silicon development and an agreement to use AMD’s MI450 chips.

The relationship is longstanding, with OpenAI CEO Sam Altman noted as a personal investor in Cerebras, and formal negotiations for the accord having commenced last autumn, culminating in a signed Letter of Intent prior to Thanksgiving.