OpenAI Unveils Jalapeno AI Chip to Rival Nvidia

OpenAI developed a new AI processor called Jalapeno in partnership with Broadcom Inc. to optimize the inference phase of AI models. In performance tests against the Nvidia Corp. GB300, the Jalapeno chip demonstrated superior response speeds and higher AI workload efficiency per unit of power.

The chip operates at a low voltage of 700 watts, a specification OpenAI expects will lower overall data center power expenses. While the processor can handle larger models than technology from providers such as Cerebras Systems Inc., it is not designed for AI training.

OpenAI plans to integrate the chips into its infrastructure later this year. This deployment will allow customers to choose between high-throughput options for lower costs or low-latency responses for increased speed. OpenAI chip chief Richard Ho stated that the company is pursuing in-house silicon to spur innovation and lower costs, though he noted that Nvidia remains a key supplier.