OpenAI's Jalapeño chip posts inference wins over Nvidia Blackwell — but not on cost

AI/OpenAI/
OpenAI's Jalapeño chip posts inference wins over Nvidia Blackwell — but not on cost
Illustration

OpenAI published the first benchmarks for Jalapeño, the inference chip it designed with Broadcom, claiming 1.5x-1.9x more work per watt and materially lower latency than commercial Nvidia Blackwell systems on open models. Analysts at SemiAnalysis, who ran the comparison, note cost per token lands roughly level with Nvidia's Vera Rubin, and Jalapeño is still an engineering sample, ramping through 2027.

4 sources

  1. 01TechCrunch — OpenAI's Jalapeño chip is built for fast inference at scale, benchmarks showtechcrunch.com
  2. 02SemiAnalysis — OpenAI Jalapeño: Better Than Nvidia Blackwellnewsletter.semianalysis.com
  3. 03CNBC — OpenAI's Jalapeño AI chip brings new threat to Nvidia margins as custom silicon gains groundcnbc.com
  4. 04The Decoder — OpenAI's first custom chip Jalapeño reportedly beats Nvidia's Blackwell and Rubin in inference benchmarksthe-decoder.com