OpenAI and Broadcom have introduced a new AI chip called Jalapeño, designed to power what the companies describe as the next wave of large language models (LLMs), according to a Tech Bytes report carried by Yahoo Finance Australia.
Jalapeño is described as an inference chip. In plain terms, inference is the stage where a trained AI model actually does its job — answering questions, generating text, or responding to users in real time — as opposed to the separate, training phase where a model first learns. Chips tuned specifically for inference aim to deliver those answers faster and at lower cost when a service is running at large scale.
The project pairs OpenAI, the maker of ChatGPT, with Broadcom, a major designer of semiconductors and networking hardware. According to Memeburn, the Jalapeño chip signals OpenAI's broader "full-stack" AI ambition heading into 2026 — meaning the company is reaching beyond software and models toward the underlying hardware that runs them.
The source items announce the chip and its purpose but do not, in the material provided here, detail specifications, manufacturing arrangements, pricing, or a firm availability date. Those particulars remain to be confirmed.
Why it matters: most cutting-edge AI today runs on chips from a small number of suppliers, and an inference processor co-developed by OpenAI could reshape who controls — and profits from — the expensive infrastructure behind the AI boom.