OpenAI and Broadcom Unveil ‘Jalapeño’ AI Chip OpenAI’s first custom AI chip, “Jalapeño,” has quickly become a flashpoint in the race to control the hardware behind large language models, promising cheaper, more efficient AI while raising questions about industry dependence on a few chipmakers.

OpenAI and Broadcom first detailed Jalapeño on June 22, describing it as an “Intelligence Processor” and the “first AI accelerator in a multi-generation compute platform” built to make advanced AI “faster, more reliable, and more accessible to more people.” OpenAI said the ASIC was designed from scratch around its understanding of LLM fundamentals and future product needs, with early lab tests—running workloads including GPT‑5.3‑Codex‑Spark—showing performance per watt “substantially better than current state-of-the-art.”

Over the following days, human-focused coverage filled in commercial and competitive details. Axios reported on June 24 that OpenAI has begun testing Jalapeño in its labs for Codex‑like tasks and plans to handle customer queries later in 2026, part of a broader goal to power “10 gigawatts’ worth of compute by 2029.” The first chips are expected in commercial use at Microsoft and other partners by year’s end, with real volume next year.

The Verge framed the move as an attempt to ease reliance on scarce Nvidia GPUs, noting that Jalapeño is an inference-only ASIC for powering current and future LLMs and quoting Broadcom CEO Hock Tan’s claim that it matches Nvidia Blackwell and Google’s Tensor Processing Units in performance. Ars Technica emphasized that Jalapeño is meant for “LLM inference at scale” in data centers and that OpenAI hopes to “own the full stack” to reduce dependence on outside companies like Nvidia amid a global compute crunch.

OpenAI president Greg Brockman struck an optimistic tone, calling Jalapeño “designed from scratch for LLM inference over nine months” with “perf per watt looking incredible” in a post on X. But not everyone sees hyperscaler chips as the future; Hugging Face CEO Clément Delangue amplified a contrasting view, reposting that “Local and Opensource AI are going to win,” underscoring a parallel bet on decentralized, open alternatives to tightly controlled proprietary stacks.

Continue reading https://foxvector.com/stories/019efd43-3d8c-1476-706e-22264d70c554