OpenAI tests Broadcom AI chip that cuts inference costs by 50%
What's next: Microsoft data centers and other partner sites are set to start integrating the finalized Jalapeno chips later this year.
OpenAI has begun testing its first custom artificial intelligence chip, built with Broadcom, as the ChatGPT maker pushes to lower the cost of running AI models. The chip, called Jalapeno, has delivered roughly 50% savings versus typical AI graphics processors in early inference tests, Broadcom CEO Hock Tan said. OpenAI said it has received first samples and is evaluating performance, including energy efficiency. The finalized chips are due to be installed later this year in large data centers operated by Microsoft and other partners. OpenAI still depends heavily on Nvidia, but it has been widening its supplier base with deals involving AMD and Cerebras Systems as demand for AI services keeps rising.