Arm Joins Nvidia’s NVLink Fusion Ecosystem

At Supercomputing ’25, Arm announced that it has officially joined Nvidia’s NVLink Fusion ecosystem—a move that significantly reshapes the CPU-GPU integration landscape. With Arm on board, NVLink Fusion is now supported by two major microarchitecture developers and four CPU vendors, marking a major expansion of the technology’s reach.

What This Means for Arm

Arm-based processors will now be able to communicate directly with Nvidia GPUs and AI accelerators using NVLink’s high-bandwidth, cache-coherent interface. Until now, only Nvidia’s Grace CPUs offered such tight GPU integration. With NVLink IP integrated into Arm architectures, Arm licensees—including companies like Google, Meta, and Microsoft—can design custom data-center CPUs that connect to Nvidia GPUs without relying on PCIe or Grace.

Nvidia’s Data Center Product Marketing Director Dion Harris emphasized that embedding NVLink IP into Arm designs enables faster, cheaper, and less complex custom CPU-SoC development.

Strategic Benefits for Arm

Arm gains three major advantages:

1. A stronger data-center position: Arm becomes the only large-scale IP provider offering native compatibility with Nvidia GPUs over NVLink, making its architecture far more appealing to hyperscalers and cloud operators.
2. More competitive server CPUs: Future Arm server chips can directly compete with Intel Xeon within Nvidia-centric systems, just as Grace and Vera CPUs do today.
3. A stronger ISA ecosystem: Native NVLink support gives Arm a technological edge over x86 and RISC-V, which currently lack comparable GPU-coherent connectivity.

Why It Matters for Nvidia

Bringing Arm into the NVLink Fusion ecosystem allows Nvidia to expand the pool of NVLink-capable CPUs without needing to design them in-house. Any custom Arm SoC can now become a first-class participant in Nvidia GPU infrastructures. This broadens Nvidia’s server footprint and could diminish the appeal of competing initiatives such as UALink and acceleration efforts from AMD, Broadcom, and Tenstorrent.

The move also simplifies adoption for governments and cloud providers that prefer Arm CPUs for sovereignty or efficiency reasons.