OpenAI and Broadcom have jointly developed a custom artificial intelligence chip designed to enhance the efficiency and reduce the operational costs of large-language model inference. Named Jalapeño, the chip was created specifically to support AI workloads, marking a strategic step by OpenAI to build a more integrated technology stack for its AI models and services.
Announced Wednesday, the Jalapeño chip is optimized around the unique memory, networking, and computing requirements of large-language models rather than being adapted from existing general-purpose hardware. Early testing indicates that the chip delivers significantly improved performance per watt compared to current top-tier alternatives, although the companies have not yet released precise performance metrics. A comprehensive technical report detailing the chip’s capabilities is expected in the coming months.
The collaboration reflects OpenAI’s broader initiative to reduce dependence on external hardware providers and tailor its infrastructure to better serve the growing demand for AI applications. “The world is moving to a compute-powered economy,” said Greg Brockman, OpenAI’s President and Co-Founder. He added that designing more components internally allows the company to expand AI access while improving operational efficiency.
Broadcom, which contributed silicon design and networking technology for the project, described the effort as the beginning of a multigenerational partnership intended to build large-scale AI infrastructure. This follows an announcement last fall that the two companies planned to deploy up to 10 gigawatts of custom AI chips and computing systems over four years.
The joint effort positions OpenAI among an increasing number of technology firms developing proprietary chips to better manage the cost and performance demands of AI workloads. By investing in specialized hardware, OpenAI aims to support the rapid growth of its AI offerings while advancing broader access to intelligent technologies.