Arm Integrates Nvidia NVLink into Neoverse CPUs for AI

Arm Holdings announced on Monday at the Supercomputing 2025 conference that its Neoverse CPU designs will integrate Nvidia's NVLink Fusion technology, enabling direct and efficient connections between custom Arm-based processors and Nvidia graphics processing units in AI data centers.

The partnership allows hyperscalers such as Microsoft, Amazon, and Google to pair their tailored Arm CPUs with Nvidia GPUs without relying on traditional PCIe links, which often create bottlenecks in high-performance computing. This integration incorporates a new protocol into Neoverse chips for seamless data transfer, supporting configurations where a single CPU connects to multiple GPUs, typically up to eight, to handle intensive AI workloads like machine learning models. Arm, which licenses its instruction set architecture and provides chip designs but does not manufacture processors itself, stated that the move extends its platform to deliver "Grace Blackwell-class performance, bandwidth, and efficiency" for AI systems.

Nvidia's NVLink technology, introduced over a decade ago, facilitates high-speed communication among chips in AI servers. Previously, full NVLink capabilities were limited to...