Arm to Design Custom Server CPU for OpenAI

According to a report by The Information, Arm is developing a server-grade CPU designed to power OpenAI's next-generation AI infrastructure, marking one of the company's most ambitious moves into the data center market so far.

The new CPU will serve as a key component of OpenAI and Broadcom's joint plan to build custom AI accelerators for massive data centers with a target capacity of 10 gigawatts (GW). It will work alongside OpenAI's in-house AI chips, which are focused primarily on inference workloads. Broadcom has been co-developing these AI chips with OpenAI for the past 18 months, aiming for mass production in the second half of 2026, followed by expansion through 2029 to achieve the 10 GW compute target. The companion Arm-based CPU is expected to be ready around the same time.

Unlike its traditional approach of licensing CPU architecture and IP to partners, Arm is now directly designing its own processors, signaling a shift in strategy toward vertical integration. Sources familiar with the matter said OpenAI could use Arm's CPUs not only with its own AI chips but also in systems running NVIDIA and AMD hardware.

This collaboration could generate multi-billion-dollar revenue for Arm—a major boost for its parent company SoftBank, which holds nearly 90% of Arm's shares and has heavily leveraged that stake. SoftBank has also pledged tens of billions of dollars to support OpenAI's data center expansion and to purchase its AI technologies, potentially accelerating Arm's entry into high-performance computing markets.

With additional chip supply agreements involving NVIDIA and AMD, OpenAI's total planned data center capacity may reach nearly 30 GW. Including its in-house chip efforts, the company's overall infrastructure investment could exceed $1 trillion, underscoring the scale of its long-term AI ambitions.