Report: Arm developing custom CPU for OpenAI's in-house accelerator — core IP would underpin 10GW of installed AI capacity

OpenAI is collaborating with Arm to develop a custom CPU for its in-house AI accelerator, which will support up to 10GW of AI capacity. Arm's involvement marks a significant move into the data center market, with plans to design a server-class CPU for OpenAI's AI racks. The chip, part of OpenAI's partnership with Broadcom, is tailored for inference workloads and expected to enter production in late 2026. This collaboration could potentially generate billions in revenue, benefiting SoftBank, which owns Arm. OpenAI's chip programs, including agreements with Nvidia and AMD, aim to reach a total installed base of over 26GW, impacting the AI training market and potentially alleviating GPU supply constraints.