Overview
- OpenAI announced Jalapeño on June 24 as a purpose-built processor for running large language models and real-time inference requests.
- The chip was designed from scratch in about nine months with Broadcom and Celestica and is already running production‑target workloads in OpenAI’s lab, including GPT‑5.3‑Codex‑Spark.
- OpenAI says early tests show substantially better performance per watt than current technology, but independent benchmarks and large-scale production results are not yet public.
- The move is intended to reduce OpenAI’s reliance on external GPU suppliers such as Nvidia and to anchor a planned multigenerational compute platform that will begin rolling out to Microsoft and other partners this year.
- If Jalapeño delivers at scale, operators could see lower inference costs and more available capacity for services like ChatGPT, a shift that follows similar custom‑chip strategies at Google, Amazon and Meta.