半导体芯闻·2025-10-15 10:47Core Insights - OpenAI is collaborating with Arm to develop a new CPU for its custom AI accelerator designed in partnership with Broadcom, marking a significant step for Arm in the data center market [1][2] - The custom AI accelerator, expected to begin production by the end of 2026, aims to achieve approximately 10 gigawatts of computing power between 2026 and 2029 [1] - OpenAI's CPU project could generate potential revenues in the billions, providing substantial benefits to SoftBank, which holds nearly 90% of Arm's shares [2] Group 1 - Arm is not just providing architecture licensing but is also designing and manufacturing CPUs, viewing the collaboration with OpenAI as a key opportunity to expand its server business [2] - OpenAI's total planned data center computing power has reached 26 gigawatts, with potential investments exceeding $1 trillion for custom chip deployments, including partnerships with Nvidia and AMD [2] - The collaboration with Broadcom may enhance OpenAI's negotiating power with Nvidia, which currently dominates the AI training market with its H100 and upcoming Blackwell GPUs [3] Group 2 - The custom AI accelerator has been in development for about 18 months and is expected to alleviate GPU supply constraints faced by AI labs over the past year [3]