OpenAI Unveils Custom ‘Jalapeño’ AI Chip Developed With Broadcom OpenAI’s push to control more of the hardware behind ChatGPT has moved from strategy to silicon, intensifying a broader industry race to loosen Nvidia’s grip on AI computing.

On June 22, OpenAI and Broadcom officially unveiled Jalapeño, describing it as the company’s first “Intelligence Processor,” an accelerator “architected around OpenAI’s vision for the future of LLM inference” and the first in a multi‑generation compute platform. Jalapeño is an ASIC tuned for running large language models rather than training them, built over nine months with Broadcom and manufacturing partner Celestica to optimize compute, memory, and networking for higher real‑world utilization.

OpenAI executives quickly framed the launch as a major efficiency win. President Greg Brockman highlighted that Jalapeño was “designed from scratch for LLM inference over nine months” and said performance per watt was “looking incredible.” The company says early tests suggest the chip will deliver “performance per watt substantially better than current state-of-the-art,” with a detailed technical report still to come. CEO Sam Altman offered a characteristically brief verdict on X: “team cooked, spicily.”

By June 24, external reporting filled in the competitive stakes. Axios noted that OpenAI has begun lab testing Jalapeño as the “first in a family of homegrown chips,” aimed at handling customer queries later this year, with commercial use at Microsoft and other partners expected by year-end and real volume in 2027. The company ultimately wants these custom chips powering 10 gigawatts of compute by 2029. Broadcom CEO Hock Tan argued the partnership proves that “you cannot, should not rely on some other third-party GPU” for such a key capability.

Coverage from The Verge and Ars Technica stressed how Jalapeño is meant to reduce dependence on scarce Nvidia GPUs by focusing on inference and matching or surpassing rival accelerators in efficiency. Both outlets highlighted OpenAI’s claim that this is just the “first step in a multi-generation compute platform” expected to be widely deployed by the end of 2026.

Industry analysts place the move in a larger realignment. TechCrunch framed Jalapeño as “Big Tech’s spiciest move away from Nvidia,” part of a trend that includes Google, Apple, Amazon, Meta, and even SpaceX building custom chips to escape “single-supplier risk” and gain Apple‑style performance and cost advantages from vertical integration.

Not everyone sees centralised data‑center chips as the long‑term destination. Hugging Face CEO Clément Delangue amplified a contrasting vision, reposting the claim: “I have never been more confident that Local and Opensource AI are going to win,” suggesting that even as hyperscalers design proprietary silicon, the race between cloud mega‑models and local, open‑source systems is far from settled.

Continue reading https://foxvector.com/stories/019f0979-e85a-360d-7079-158489ad8279