OpenAI, Broadcom Unveil Jalapeño Inference Chip
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI and Broadcom unveiled Jalapeño, an LLM-optimized inference chip, developed from design to manufacturing tape-out in nine months
- Jalapeño's development was aided by OpenAI's own AI models — a self-referential design approach the company highlighted as speeding the timeline
- Early testing indicates the first-generation accelerator will deliver performance per watt substantially better than current state-of-the-art inference hardware
- Cross-outlet framing splits: Bloomberg and the NYT led on 'run models faster, cheaper'; Gizmodo, How-To Geek, and crypto.news cast it as a direct challenge to Nvidia
- TechRadar's title framed the move as an 'Apple-like...build the full stack' play — an angle most other coverage did not foreground
Why it matters: OpenAI's first custom inference chip, built with Broadcom in a nine-month cycle and designed in part by OpenAI's own models, signals a vertical-integration push to reduce Nvidia dependence. The performance-per-watt gains claimed in early testing could materially lower inference costs for OpenAI's frontier models, with deployment targeted by end of 2026.
