88Z AI, 25 June — OpenAI has unveiled Jalapeño, its first in-house inference chip, developed with Broadcom and Celestica under the leadership of a former Google TPU core member. Tailored for models like ChatGPT, it set a record for fastest ASIC tape-out and is scheduled for deployment by late 2026 with superior energy efficiency.
At a glance
- Led by a former Google TPU member, Jalapeño set a record with tape-out in just 270 days
- Optimized for models like ChatGPT, it will deploy in late 2026 with leading energy efficiency
The update
OpenAI has officially launched Jalapeño, its first self-developed inference acceleration chip. Led by Richard Ho, a former core expert of Google's TPU team, the project was manufactured in collaboration with Broadcom and Celestica. Jalapeño is custom-built for large models such as ChatGPT and Codex, aiming to improve actual utilization by optimizing data movement and resource balance. Taking only 270 days from design to tape-out, it set a new record for the fastest development of high-performance ASICs. The chip reportedly offers significantly better performance per watt than current mainstream solutions and is scheduled for deployment by late 2026 to support more efficient AI inference services.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 量子位Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.