OpenAI Partners With Broadcom to Develop Custom Jalapeño AI Inference Chip

OpenAI has unveiled Jalapeño, a custom AI inference chip developed in partnership with Broadcom, marking a pivotal shift in the company’s infrastructure strategy. Designed as an application-specific integrated circuit (ASIC), Jalapeño focuses on accelerating the execution of large language models (LLMs) and improving performance per watt, aiming to reduce OpenAI’s long-standing reliance on Nvidia’s general-purpose graphics processing units (GPUs).

The collaboration leverages Broadcom’s deep expertise in silicon implementation, board systems, and networking, including its Tomahawk networking technology. By designing a custom platform tailored for inference—the process of running models on new data—OpenAI intends to optimize its stack for greater efficiency, reliability, and accessibility. The chip development, which was completed in nine months, follows broader industry trends where major hyperscalers are creating in-house silicon to manage the high costs and infrastructure demands of frontier AI models.

This development underscores a fundamental commitment to scaling AI infrastructure, with early plans for deployment in massive data centers starting in 2026. As OpenAI joins companies like Google and Amazon in producing custom processors, the move represents a critical step in the ongoing race to secure compute capacity, with the company continuing to work with Microsoft and other partners to satisfy the surging global demand for faster and more affordable AI intelligence.