NVIDIA advances quantum-classical integration as more than a dozen scientific computing centers adopt NVQLink to connect quantum processors with the Grace-Blackwell accelerated computing platform. Centers across Asia, Europe, the Middle East, Australia, and the U.S. plan to deploy the open interconnect to run hybrid quantum-GPU workflows using the CUDA-Q platform. The move signals an industry-wide shift toward quantum-GPU supercomputers designed to handle real-time error correction, ultra-low-latency control loops, and scalable entanglement experiments.
NVIDIA says NVQLink delivers sub-4-microsecond latency and 400 Gbps throughput between QPUs and GPUs, enabling 40 petaflops (FP4) of AI performance for hybrid workloads. Early adopters include AIST G-QuAT, RIKEN, KISTI, NCHC Taiwan, Singapore’s NQCH, and Australia’s Pawsey Centre, alongside major European and Middle East HPC sites such as CINECA, GENCI, JSC, NQCC, PCSS, IT4Innovations, TII (UAE), and KAUST. They join U.S. national labs—including ORNL, LBNL, Sandia, Fermilab, and Brookhaven—that have already begun linking their quantum systems to NVIDIA GPUs.
Quantinuum demonstrated the first real-time scalable decoder for qLDPC quantum error-correction codes on its Helios quantum processor using NVQLink, reporting a 67-microsecond reaction time—32× faster than Helios’ required two-millisecond window. NVQLink’s Ethernet-based architecture also gives labs the option to scale classical compute clusters as QPU sizes increase, while maintaining a single programming model through CUDA-Q.
• NVQLink throughput of 400 Gbps with <4 µs latency for GPU–QPU links
• Integrated with NVIDIA CUDA-Q for real-time quantum error correction and hybrid workflows
• Supported across Asia, Europe, Middle East, Australia, and leading U.S. national labs
• Quantinuum’s Helios demonstrates first scalable real-time qLDPC decoding via NVQLink
• Ethernet-based architecture eases expansion of surrounding classical compute
• Designed to integrate diverse quantum processors and control systems
• Enables 40 petaflops FP4 AI performance in hybrid quantum-GPU pipelines
“In the future, supercomputers will be quantum-GPU systems — combining the unique strengths of each,” said Jensen Huang, founder and CEO of NVIDIA. “NVQLink with CUDA-Q is the gateway to that future.”
🌐 Analysis
NVIDIA’s NVQLink strategy positions CUDA-Q as the de facto programming environment for hybrid quantum-classical research, similar to how CUDA defined GPU computing in the 2000s. The broad uptake across national labs and quantum hardware developers—including Quantinuum’s Helios roadmap—signals an emerging consensus that tight GPU integration is essential for real-time decoding and control. Competing approaches include IBM’s Qiskit Runtime with classical acceleration, IonQ’s photonic-QPU network strategy, and Xanadu’s photonic workflows—none of which yet match NVQLink’s standardized GPU-QPU interface or its global supercomputing footprint.
Please see our feature article on NVQLink
