Core Viewpoint - OpenAI is reportedly developing custom AI accelerators with the help of Broadcom to reduce reliance on Nvidia and lower costs for its GPT series models [1][2]. Group 1: OpenAI and Broadcom Collaboration - Broadcom's CEO revealed that OpenAI is a $10 billion customer, indicating a significant partnership focused on AI technology [1]. - OpenAI is rumored to be developing a chip to replace Nvidia and AMD GPUs, expected to be unveiled next year for internal use only [1][2]. - Broadcom is currently serving three XPU customers and anticipates improved AI revenue prospects for fiscal year 2026 due to substantial orders [1]. Group 2: Technical Aspects of the Chip - The architecture of OpenAI's chip may resemble AMD's MI300 series, utilizing advanced stacking techniques for high-performance computing [5]. - Broadcom's 3.5D XDSiP technology is likely a candidate for OpenAI's chip, supporting extensive AI computing systems [5][7]. - The chip will require high-performance matrix multiplication units and sufficient high-bandwidth memory to optimize AI workloads [7]. Group 3: Broader Industry Context - OpenAI's investment in AI infrastructure aligns with broader trends, as other companies like Apple are also rumored to collaborate with Broadcom for custom AI accelerators [9][10]. - Apple has committed to investing $500 billion and expanding its domestic manufacturing capabilities, indicating a competitive landscape in AI chip development [10].