Arm Hits 25% Share in Data Center CPUs

The rapid expansion of AI data centers is creating new opportunities in server computing, with Nvidia emerging as a key driver—not only leading GPU growth but also helping Arm-based CPUs gain traction in the market.

According to the latest report from Dell’Oro Group, Arm CPUs captured 25% of the server CPU market in Q2 2025, up from 15% a year ago. This surge is largely fueled by Nvidia's rack-scale compute platforms that combine Grace CPUs and Blackwell GPUs, including models like the GB200 and GB300 NVL72.

Each GB300 NVL72 rack, rated at 120 kW, is equipped with 72 Blackwell GPUs and 36 Grace CPUs. First launched in 2022, the Grace CPU is built on the Arm Neoverse V2 architecture and leverages Nvidia's custom NVLink-C2C interface to maximize data throughput across the system.

Analysts note that a year ago, Arm's server market presence relied almost entirely on in-house processors from major cloud providers like AWS Graviton. Today, revenue from Nvidia's Grace CPU rivals that of other cloud-focused Arm CPUs, signaling a broader adoption of Arm-based solutions in data centers.

Despite strong growth, the current 25% market share remains below the 50% target set by Arm's architecture head, Mohamed Awad, for the end of 2025. However, as more chip designers develop Arm-based server CPUs, market share is expected to continue rising.

Nvidia is already working on the next-generation Arm-based CPU, codenamed Vera, featuring in-house cores. Meanwhile, Qualcomm and Fujitsu are also developing or updating Arm server CPUs, both certified for NVLink Fusion technology. This opens the possibility for future "Superchip" architectures built on multiple Arm CPU variants, not just Nvidia's own processors.

Dell' Oro data further shows that the AI expansion cycle is pushing server and memory component markets upward, with Q2 annual growth at 44% and projected 46% growth for 2025.

At the same time, as AI clusters move toward Ethernet-based designs, sales of SmartNICs and DPUs have nearly doubled compared to last year. Custom AI ASIC shipments have now reached a scale comparable to GPUs, though GPUs still dominate AI accelerator revenue.