OpenAI and Broadcom Unveil 'Jalapeño' AI Chip

OpenAI and Broadcom Unveil ‘Jalapeño’ AI Chip OpenAI’s decision to design its own “Jalapeño” AI chip with Broadcom marks a pivotal shift in the power dynamics of the AI hardware market, aiming to ease a global compute crunch while subtly challenging Nvidia’s dominance.

From concept to silicon

OpenAI and Broadcom first unveiled Jalapeño on June 22 as an “Intelligence Processor” built around OpenAI’s vision for the future of large language model (LLM) inference. OpenAI says the ASIC was “designed from scratch for LLM inference over nine months,” with engineering samples already running workloads, including future models like GPT‑5.3‑Codex‑Spark, at target power and frequency.

On social media, OpenAI president Greg Brockman framed the project as a rapid, model-informed hardware effort: “Introducing Jalapeño — designed from scratch for LLM inference over nine months, accelerated by our models. Perf per watt looking incredible.” CEO Sam Altman added a characteristic, if cryptic, victory lap: “team cooked, spicily.”

Testing, deployment, and goals for scale

By June 24, reporting indicated OpenAI had begun testing Jalapeño in its labs on tasks similar to answering Codex queries, with better-than-expected thermal performance. Broadcom and OpenAI expect the first chips to enter commercial use at Microsoft and other partners by the end of the year, with real volume ramping in 2027 as part of OpenAI’s ambition to power 10 gigawatts of compute by 2029.

Coverage from The Verge emphasized Jalapeño as a custom inference ASIC meant to reduce reliance on scarce Nvidia GPUs, with Broadcom’s CEO claiming performance on par with Nvidia’s Blackwell and Google’s TPUs, while OpenAI touts “performance per watt substantially better than current state-of-the-art.”

Industry context and competing visions

Analyses from Ars Technica and TechCrunch place Jalapeño within a broader wave of custom AI silicon by Google, Amazon, Microsoft, Apple, and even SpaceX, all seeking more control and capacity amid a “compute-powered economy.” Broadcom CEO Hock Tan argues leaders “cannot, should not rely on some other third-party GPU” for such a critical component.

Not everyone sees hyperscaler chips as the endgame. Hugging Face CEO Clément Delangue amplified the opposite bet, retweeting that “Local and Opensource AI are going to win,” suggesting that open, decentralized models running on diverse hardware will counterbalance vertically integrated giants like OpenAI.

Continue reading https://foxvector.com/stories/019f0640-b82d-19fb-70d8-120529544230