On the evening of Wednesday, June 24 (Beijing Time), OpenAI and Broadcom jointly announced the launch of their first custom AI chip, Jalapeño, specifically designed for large-scale language model (LLM) inference tasks. The chip took only 9 months from design to tape-out, with early tests showing significantly improved energy efficiency compared to current top-tier standards and cost savings of approximately 50%. OpenAI plans to integrate Jalapeño into data centers of partners such as Microsoft later this year and deploy gigawatt-scale computing infrastructure. This move marks OpenAI's further control over its computing supply chain, reducing reliance on NVIDIA GPUs while optimizing model operational efficiency and costs through vertical integration. As the first product in a multi-generation chip roadmap, subsequent iterations of Jalapeño will continue to strengthen OpenAI's computing advantage.