San Francisco / London — OpenAI has confirmed a new multi-year computing agreement with Cerebras Systems, a specialised AI hardware start-up, marking one of the largest compute capacity deals in the artificial intelligence industry to date.
Under the arrangement, OpenAI will purchase up to 750 megawatts (MW) of high-performance computing power over the next three years to support its growing AI workloads. The deal is expected to accelerate the deployment of wafer-scale AI systems capable of handling large-scale model inference and training tasks.
Industry analysts say the agreement reflects the soaring demand for AI computing resources as companies race to build and operate more powerful artificial intelligence systems. By securing this capacity, OpenAI aims to ensure it has the infrastructure necessary to train advanced models and deliver services to a global user base.
The scale of the arrangement — equivalent to a substantial fraction of a major data centre’s power usage — highlights how compute-intensive modern AI development has become. This has broader implications for the hardware ecosystem, prompting chip designers and data-centre operators to adapt quickly to meet demand.
The deal follows a series of strategic partnerships OpenAI has forged with industry heavyweights, including long-term collaborations on data centre build-outs and cloud computing capacity. These initiatives are part of a wider effort to diversify OpenAI’s infrastructure footprint and avoid over-reliance on any single supplier.
While spending at this scale has prompted questions about long-term financing and cost management, OpenAI’s continued investment in compute capacity underscores its commitment to advancing the state of AI research and applications.
Author: Pishon Yip