OpenAI and Broadcom Unveil Jalapeño Chip for Large Language Model Inference
OpenAI and Broadcom team up on custom AI silicon
OpenAI and Broadcom have announced a new chip designed specifically for large language model inference in data centers, marking another sign that the AI industry is moving deeper into custom hardware.
The chip, called Jalapeño, is being pitched as the first generation in a longer-term collaboration between the two companies. Broadcom said the ASIC was designed from scratch for LLM inference, drawing on “detailed insights” from discussions with OpenAI researchers and the company’s roadmap for future models and products. Broadcom said the chip was designed and built in nine months.
OpenAI said early testing suggests Jalapeño will deliver “performance per watt substantially better than current state-of-the-art,” though it added that testing is still ongoing and a more detailed technical report will follow in the coming months.
The effort reflects OpenAI’s broader ambition to control more of the infrastructure behind its models and products, reducing reliance on outside suppliers such as Nvidia. It also comes as AI companies continue to look for ways to stretch limited compute resources amid a global data center crunch.
Broadcom, already a major chip supplier for infrastructure builders, has been expanding its custom silicon business as hyperscalers and frontier model developers seek specialized chips tailored to their own workloads.
Both companies said Jalapeño chips are expected to be deployed in data centers by the end of this year.
Sources:
Read more tech news on the Doppler VPN Blog.