"Comparable Performance to Blackwell and TPU"

OpenAI has unveiled its proprietary artificial intelligence (AI) semiconductor, "Jalapeño," which was jointly developed with Broadcom. This move is interpreted as an effort to reduce its dependence on Nvidia by developing custom chips in-house.


Sam Altman, CEO of OpenAI, is giving a speech at an event in Tokyo, Japan, on February 3, 2025 (local time). Photo by Reuters Yonhap News

Sam Altman, CEO of OpenAI, is giving a speech at an event in Tokyo, Japan, on February 3, 2025 (local time). Photo by Reuters Yonhap News

View original image

OpenAI and Broadcom announced on June 24 (local time) that they have unveiled the jointly developed AI chip "Jalapeño" and plan to deploy it in actual data centers and other locations starting at the end of this year.


According to both companies, initial testing showed that its performance is comparable to Nvidia's Blackwell and Google's Tensor Processing Unit (TPU). The companies emphasized that the development cycle, from design to tape-out, took only nine months, making it one of the fastest among application-specific integrated circuits (ASICs).


Jalapeño was designed from the outset specifically for inference, based on experience gained from operating ChatGPT. Its key features include reducing data transfer bottlenecks and optimizing computing, memory, and network resources to maximize performance per watt of power consumed.


This chip is manufactured by the Taiwanese foundry TSMC, while Samsung Electronics and SK hynix supply the memory. Broadcom plans to release the next version in 2028 and introduce new products every year thereafter.



Hyperscaler companies are leading the way in developing their own chips to reduce reliance on Nvidia. Google uses its own TPU, and Anthropic is also reportedly considering developing its own chips.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing