OpenAI and Broadcom have announced Jalapeño, a custom silicon chip built specifically for large language model inference in data centers.
This is generation one of a stated long-term co-development roadmap. Both companies have signaled iterative refinement over time, meaning Jalapeño is a starting point, not a finished product. The choice of Broadcom matters: this is not a startup partnership. Broadcom moves serious silicon volume.
The full article is worth reading for the architectural details and what inference-specific design actually means for deployment at scale. The gap between a general-purpose GPU running inference and a chip purpose-built for it is where the real story lives.
[READ ORIGINAL →]