OpenAI and Broadcom unveil Jalapeño chip for AI data centers

What happened

OpenAI and Broadcom unveiled a new chip called Jalapeño, an application-specific integrated circuit (ASIC) built from scratch for LLM inference in large data centers. The two companies say this is the first generation of a long-term project, and the chip's design was informed by OpenAI's roadmap and took nine months to develop.

Why it matters

Current data centers rely on general-purpose chips that are not tailored to the specific demands of language models. Jalapeño is intended to be more specialized for these workloads, which may help companies reduce energy costs and improve efficiency when running AI systems at scale.

What to watch

OpenAI states that early testing shows Jalapeño will deliver performance per watt substantially better than current state-of-the-art, though the company notes it is still measuring performance and will present a detailed technical report in the coming months.