OpenAI introduces Jalapeño, its first custom AI processor for ChatGPT

In short: OpenAI says it has created its first custom processor, called Jalapeño, to help run ChatGPT and other AI tools in its servers.

What happened

OpenAI announced a new “intelligence processor” chip made with Broadcom. The chip is called Jalapeño, and it is meant to be used inside the company’s server computers that handle ChatGPT requests.

Jalapeño is an ASIC, which means a chip built for one main job instead of many (like a kitchen tool made only for slicing bread). OpenAI says Jalapeño is designed for AI “inference”, which is the step where an AI system answers your question or carries out a task. This is different from “training”, which is when the AI learns from huge amounts of data before it can answer at all.

The Verge notes this announcement comes about nine months after OpenAI said it would work with Broadcom on custom chips. The goal is to rely less on Nvidia’s widely used AI chips, which have been hard to get in large quantities. Broadcom CEO Hock Tan told Reuters that Jalapeño matches the performance of Nvidia’s Blackwell chips and Google’s TPU chips.

OpenAI says Jalapeño is the first step in a multi-generation plan, and it expects to start deploying it by the end of 2026. OpenAI also says early tests show better “performance per watt”, meaning it can do more work using less electricity.

Why it matters

Most people will not see this chip, but it could affect the speed, cost, and energy use of AI services like ChatGPT. If OpenAI can run more requests with less power and fewer supply bottlenecks, it may help the company keep up with demand as more people use AI tools.

Source: The Verge AI