OpenAI and Broadcom Unveil Jalapeño AI Inference Chip

OpenAI has officially unveiled its first custom AI processor, named Jalapeño, developed in collaboration with Broadcom. The Application-Specific Integrated Circuit (ASIC) is purpose-built to optimize inference workloads, allowing OpenAI to run large language models more efficiently while reducing its reliance on Nvidia’s high-cost graphics processing units.

Designed in just nine months, Jalapeño aims to address the growing demand for scalable compute power by balancing memory movement and networking resources. This launch represents a core pillar of OpenAI’s long-term full-stack infrastructure strategy, which seeks to make AI services more affordable and faster. Broadcom’s involvement underscores their position as the primary partner for hyperscalers aiming to create custom silicon to support massive, gigawatt-scale data center operations.

The move places OpenAI in direct competition with major players like Google, Amazon, and Tesla, all of whom have developed their own custom processors to navigate the supply constraints in the AI semiconductor market. By transitioning toward proprietary hardware, OpenAI expects to achieve better performance-per-watt efficiency compared to existing state-of-the-art GPUs. The company plans to release a detailed technical report on the chip's performance capabilities in the coming months as it prepares for large-scale production deployments scheduled to begin in 2026.