OpenAI has unveiled Jalapeño, a custom processor designed specifically for AI inference and already being tested with its GPT-5.3-Codex-Spark model.
Developed with Broadcom and manufactured by TSMC, the chip is part of OpenAI's strategy to reduce dependence on Nvidia while improving performance and energy efficiency.
Jalapeño is the first step in a multi-generation custom silicon roadmap aimed at making ChatGPT and future AI services faster, cheaper and more reliable.
Meet Jalapeño: The AI Chip OpenAI Built to Power ChatGPT
OpenAI has unveiled Jalapeño, its first in-house AI chip designed for inference, marking a major step towards reducing dependence on Nvidia and lowering the cost of running advanced AI models
X
Summary
Advertisement
OpenAI has unveiled Jalapeño, its first custom artificial intelligence chip built specifically for inference — the process of an AI model answering user queries after its initial training.
The hardware challenges Nvidia's historical dominance as the primary supplier of processors for ChatGPT and Google's AI systems, reports said. Jalapeño is already running OpenAI's GPT-5.3-Codex-Spark model inside the company's labs.
Early tests indicate the processor delivers significantly better performance per watt than leading AI hardware, according to OpenAI. The company said this efficiency could reduce the enormous costs of running generative models.
Designed With Broadcom
Engineers completed the processor in just nine months. OpenAI designed the hardware in partnership with Broadcom, using artificial intelligence tools to accelerate portions of the design process.
Engineers sent the finalised blueprints to Taiwan's TSMC for manufacturing. The resulting silicon directly competes with existing high-end hardware in the sector.
The Broadcom CEO said the chip performs "on par with Nvidia's latest Blackwell processors and Google's Tensor Processing Units (TPUs)".
Advertisement
A Multi-Generation Strategy
OpenAI expects to begin deploying Jalapeño by the end of 2026. Celestica, a Canadian manufacturer, will build the server systems required to house the new processors.
The rollout is the first step in a multi-generation hardware strategy that will expand over the coming years, OpenAI said. The firm has been exploring custom silicon since 2023, following tech giants like Google and Amazon in reducing reliance on external suppliers.
Competitors are adopting similar manufacturing tactics. Anthropic, OpenAI's main rival, is also planning to build its own custom AI chip, Reuters reported in April.
Every improvement in speed, cost, and reliability can directly translate into faster ChatGPT responses, more capable coding assistants, cheaper AI services, and fewer outages during periods of high demand, OpenAI said.