OpenAI and Broadcom today jointly unveiled their first in-house developed intelligent processor, "Jalapeño," an AI acceleration chip designed from scratch for large language model (LLM) inference scenarios, marking the cornerstone of both companies' multi-generation computing platform strategy. OpenAI CEO Sam Altman and President Greg Brockman received engineering samples from Broadcom CEO Hock Tan. The chip has successfully run ML workloads at production-targeted frequencies and power levels in laboratory settings, including GPT-5.3-Codex-Spark. From design to tape-out took just nine months, which OpenAI describes as arguably the fastest ASIC development cycle in the field of high-performance advanced semiconductors. Behind this accelerated timeline, OpenAI relied not only on Broadcom’s design and manufacturing capabilities but also leveraged its own models to optimize the chip design process—the same model that serves users also helps improve hardware running next generations. Early tests show Jalapeño significantly outperforms current state-of-the-art solutions in performance per watt. Its architecture focuses on reducing data movement while balancing compute, memory, and network resources, bringing actual utilization closer to theoretical peaks. A detailed technical report on performance will be released several months later. Jalapeño is no mere improvement over general-purpose accelerators; it was designed from scratch specifically for LLM inference based on system-level insights accumulated by OpenAI through ChatGPT, Codex, APIs, and future Agent products. It aims to strike a balance between the throughput capacity of top-tier AI accelerators and the low latency required by fast inference systems, making it well-suited for large-scale interactive LLM applications. The deployment roadmap is already defined: starting late 2026, gigawatt-class data centers will be built alongside partners such as Microsoft, expanding year by year. Broadcom's Tomahawk networking chips and Celestica's board-level and system integration capabilities collectively support mass production of the platform. According to Brockman, “The world is moving toward a compute-driven economy. Jalapeño represents part of our long-term full-stack infrastructure strategy—designing more layers of proprietary components to serve greater intelligence with higher efficiency, making AI accessible to more people.”