OpenAI and Broadcom Unveil Jalapeno: A Game-Changer for AI Inference

OpenAI, partnering with Broadcom, introduces Jalapeno, its bespoke AI Processor for large language models, promising improved performance and energy efficiency. Developed in just nine months, the chip aims for deployment by 2026 at gigawatt-scale data centers, marking the start of a multi-generation AI computing platform.

OpenAI and Broadcom have joined forces to launch Jalapeno, a cutting-edge custom Intelligence Processor engineered specifically for large language model inference. This breakthrough chip marks OpenAI's first tailored processor and is set to transform AI efficiency and performance. Developed on an accelerated nine-month timeline, it's poised for widespread data center deployment by the close of 2026.

Hock Tan, President, and CEO of Broadcom emphasized the strategic importance of this collaboration, highlighting their dedication to enhancing AI infrastructure for the coming decade. The partnership with OpenAI represents a significant leap forward, co-developing a robust compute platform that integrates Broadcom's leading silicon technology with OpenAI's innovative AI models.

Jalapeno has been meticulously designed from inception to be the optimal solution for modern LLM inference, rather than adapting from pre-existing AI technologies. With a focus on reducing data movement and balancing computing resources, the platform leverages Broadcom's top-tier networking innovations like Tomahawk silicon. Early experiments have shown promising results, with engineering samples executing machine learning tasks effectively, including handling advanced models like GPT-5.3-Codex-Spark.