OpenAI Launches ‘Jalapeño,’ Its First Custom AI Chip Developed with Broadcom

OpenAI has officially entered the AI semiconductor market with the launch of Jalapeño, its first custom-built artificial intelligence chip developed in partnership with Broadcom. The chip is designed primarily for AI inference, the process of delivering responses from AI models such as ChatGPT to users in real time.

According to OpenAI, Jalapeño was built from the ground up specifically for large language model (LLM) workloads rather than being adapted from existing accelerator architectures. The company said the chip was designed using insights gained from operating ChatGPT, Codex, the OpenAI API, and future AI agent products.

OpenAI President Greg Brockman stated that building more of the AI technology stack internally will help the company deliver advanced AI capabilities more efficiently while expanding access to intelligent systems worldwide.

The project was developed in collaboration with Broadcom and manufacturing partner Celestica, which assisted with chip implementation, hardware integration, networking infrastructure, and large-scale production systems.

OpenAI said engineering samples of Jalapeño are already running machine-learning workloads, including support for GPT-5.3 and Codex Spark, while achieving target production performance and power-efficiency levels.

The company described Jalapeño as a flexible inference platform capable of supporting both current and future generations of large language models. By improving compute efficiency and reducing operational costs, OpenAI aims to build faster, more capable, and more affordable AI systems while continuing to invest in future AI infrastructure.

The launch marks a significant step in OpenAI’s broader full-stack AI strategy and positions the company alongside major technology firms developing custom AI hardware to support next-generation artificial intelligence applications.