OpenAI unveils Jalapeño, its first custom AI chip

OpenAI on Wednesday unveiled Jalapeño, its first custom-designed artificial intelligence chip, developed in collaboration with Broadcom ↗ 4.00% and built specifically for the inference workloads that power ChatGPT and the company's broader product suite.U.S. News & World Report+1

The Jalapeño processor is designed to work quickly and efficiently with large language models, according to Richard Ho, OpenAI's hardware lead, who told Reuters that the chip is "crafted to operate swiftly and effectively" with the LLMs driving the company's AI applications. "We believe it will excel in all upcoming iterations of LLMs," Ho said. OpenAI reported that prototypes of the chip are already functioning in its laboratories, achieving targeted power and performance benchmarks.mezha

Broadcom CEO Hock Tan said the Jalapeño chip's performance rivals that of Nvidia ↗ 0.90% Blackwell chips and Google's Alphabet Inc. ↗ 0.84% tensor processing units. OpenAI expects to deploy Jalapeño at gigawatt scale through data center partners, including Microsoft ↘ 0.24%, beginning later this year as the initial phase of a multi-generation chip development initiative.tradingpedia+1

A Long Road to Silicon

The chip represents the culmination of a partnership first reported in October 2024, when Reuters revealed that OpenAI was working with Broadcom and TSMC Taiwan Semiconductor Manufacturing Company Limited ↗ 2.95% to build its first in-house chip focused on inference. OpenAI's hardware team, led by Ho — a former Google TPU engineer — grew from roughly 20 to 40 people as the design moved toward tape-out in early 2025.reuters.com+2

The companies formally announced a 10-gigawatt custom accelerator collaboration in October 2025, with deployments slated to begin in the second half of 2026 and conclude by the end of 2029. The chip is fabricated using TSMC's 3-nanometer process and employs a systolic array architecture with high-bandwidth memory.openai+3

Financing Challenges and Competitive Context

The path to production has not been without complications. In May, The Information reported that the project — internally codenamed "Project Nexus" — faced an $18 billion financing hurdle for its first phase, with Broadcom requiring Microsoft to purchase roughly 40 percent of the chips before agreeing to finance production. Wednesday's unveiling suggests those negotiations have advanced.Stocktwits

OpenAI's move to custom silicon mirrors efforts by Google, Amazon Amazon.com, Inc. ↗ 1.61%, and Meta ↘ 1.64%, all of which have developed proprietary chips to manage surging AI compute costs. OpenAI currently runs on more than 2 gigawatts of capacity and has committed to approximately 33 gigawatts across partnerships with Nvidia, AMD Advanced Micro Devices, Inc. ↗ 5.45%, Broadcom, and Amazon. The Jalapeño chip marks OpenAI's most direct step toward reducing its dependence on Nvidia's GPUs — hardware that has cost the company billions annually.maginative.com+2

On its official account on X, OpenAI described the chip as "purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products".x