OpenAI and Broadcom announce LLM inference chip

OpenAI and Broadcom have announced a chip designed for LLM inference at scale, targeting the workload that runs trained large language models in production. The collaboration points to growing pressure on AI infrastructure as demand for model-driven services continues to test available compute capacity.

The move places the companies in a widening silicon race, where specialized hardware is becoming central to supporting large-scale AI deployments. The announcement frames inference, rather than model training, as the focus, signaling attention on the operational side of serving LLMs to users at high volume.