OpenAI and Broadcom Unveil Jalapeño Custom AI Inference Chip

On June 24, 2026, OpenAI and Broadcom Inc. unveiled Jalapeño, the first custom artificial intelligence processor designed by OpenAI. Developed in just nine months, the application-specific integrated circuit (ASIC) focuses on inference for large language models, including ChatGPT and Codex. OpenAI utilized its own AI models to accelerate the chip's design and optimization process, moving from design to manufacturing tape-out in what partners described as the fastest cycle in advanced semiconductor history.

Broadcom managed the silicon implementation, high-performance networking, and production systems, while TSMC handled manufacturing and Celestica assisted with board and rack system integration. Early testing indicates the accelerator provides superior performance per watt and could reduce inference costs by approximately 50 percent compared to typical AI GPUs. While OpenAI will continue to use Nvidia hardware for model training, Jalapeño represents a strategic shift toward proprietary infrastructure to lower operational costs and latency.

The chip is the first phase of a multi-generation roadmap aiming for gigawatt-scale data center deployment by 2029. Initial deployment is planned for late 2026, with Microsoft agreeing to purchase 40 percent of the initial output. To support this infrastructure expansion, OpenAI recently raised $122 billion, while Broadcom established a financing vehicle with Apollo Global Management and Blackstone. Subsequent chip generations are planned for 2028 and annually thereafter.