NVIDIA's CUDA-on-RISC-V and NVLink Fusion Enable Non-x86, Non-ARM GPU Hosts

At Hot Chips 2026, NVIDIA detailed CUDA-on-RISC-V requirements and NVLink Fusion, which licenses NVLink interconnect IP to third-party chips. Together, a RISC-V server CPU with a licensed NVLink controller could host NVIDIA GPUs over a coherent fabric, running CUDA workloads without x86 or ARM. NVIDIA did not address whether any RISC-V vendor currently ships qualifying silicon.