OpenAI Unveils Jalapeño AI Chip With Broadcom, Targeting End-2026 Deployment
Updated
Updated · The Verge · Jun 24
OpenAI Unveils Jalapeño AI Chip With Broadcom, Targeting End-2026 Deployment
3 articles · Updated · The Verge · Jun 24
Summary
OpenAI on Wednesday introduced Jalapeño, its first custom AI server chip, built with Broadcom to handle inference for ChatGPT and other large language models.
Jalapeño is an ASIC tuned for inference rather than training, part of OpenAI’s push to cut dependence on Nvidia GPUs, which have faced supply constraints.
Broadcom CEO Hock Tan told Reuters the chip matches Nvidia’s Blackwell and Google TPU performance, while OpenAI said early tests show substantially better performance per watt.
Nine months after disclosing the Broadcom partnership, OpenAI now calls Jalapeño the first step in a multi-generation compute platform it expects to deploy by the end of 2026.
The launch places OpenAI alongside Microsoft, Meta and Amazon in building in-house AI chips, even as Nvidia remains the overall performance leader.
With an $18 billion roadblock, will OpenAI's custom chip challenge Nvidia's dominance or become a costly misstep?
Is OpenAI's 'Intelligence Age' policy a path to democratized AI or a strategy for corporate control of the future?
OpenAI’s Jalapeño Chip: Redefining LLM Inference with Purpose-Built Silicon and Unprecedented Performance per Watt
Overview
On June 24, 2026, OpenAI unveiled Jalapeño, its first custom-designed AI inference chip, marking a pivotal step in the company’s strategy for advanced, vertically integrated AI infrastructure. Jalapeño is a blank-slate design, meticulously crafted to meet the demanding requirements of large language models and deeply informed by OpenAI’s operational systems like ChatGPT and Codex. While tailored for OpenAI’s internal needs, the chip is also designed to support current and future LLMs across the broader industry. Jalapeño aims to deliver high power and throughput, setting a new standard for AI hardware performance and scalability.