Arm custom chips get a boost with Nvidia partnership (www.cnbc.com)

Arm announced that Neoverse-based CPUs will be able to link directly with Nvidia GPUs using NVLink Fusion, and that Neoverse custom chips will include a new protocol to move data seamlessly with GPUs. Technically this opens NVLink’s high-bandwidth, low-latency interconnect to Arm-licensees via an agreed interface rather than requiring Nvidia’s own CPUs; it complements existing Nvidia configurations (like the Grace Blackwell Arm-based CPU) and mirrors Nvidia’s recent moves to enable NVLink on other vendors’ CPUs (e.g., its $5B Intel investment). Arm itself continues to supply ISAs and reference designs for partners to build custom SoCs. For AI/ML infrastructure—where GPUs and other accelerators dominate—this matters because it lowers the friction for hyperscalers and custom-hardware builders to stitch Arm Neoverse CPUs into multi-GPU servers (commonly up to eight GPUs per host). The result should be more heterogeneous, competitively supplied AI servers, potentially better latency and bandwidth between CPU and accelerator, and greater design freedom for companies that prefer custom SoCs. Strategically, it signals Nvidia is broadening NVLink’s ecosystem through partnerships rather than vertical lock-in, which could accelerate adoption of diverse CPU-GPU pairings across the AI cloud and edge markets.