Cerebras Systems powers OpenAI's GPT-5.6 Sol with 14x speed boost

Cerebras Systems has been building its business around a distinct approach to chip design: its Wafer-Scale Engine combines computing cores, memory, and communication capabilities into a single piece of silicon, reducing the need to move data between separate components. That architecture is aimed at alleviating bottlenecks that can slow AI workloads, particularly for inference—the step where an AI model answers questions, writes code, or executes tasks for users in real time.

The OpenAI relationship marks a significant validation. In August, Cerebras announced it is powering the Ultrafast mode for OpenAI's GPT-5.6 Sol, and the two companies had previously announced a multi-year agreement for 750 megawatts of compute capacity, with deployment planned in stages through 2028. This means Cerebras's technology is already part of a commercial service from a leading AI model provider, not just a lab experiment.

Beyond that deal, Cerebras is also expanding its hardware lineup. The recently announced CS-4 system is part of a broader platform that the company says will allow it to upgrade computing, cooling, and power delivery more independently, which could matter as AI data centers become larger and more complex. For investors comparing it with Nvidia and Broadcom, the key question may be whether speed-focused inference becomes a durable advantage as AI applications grow more interactive.