RIKEN ROQUO Supercomputer Shows What Hybrid Quantum Computing Actually Needs

RIKEN’s ROQUO GPU supercomputer is now operating in Kobe, giving Japan a classical HPC layer for hybrid quantum computing research rather than a standalone quantum breakthrough. Japan’s latest quantum-computing milestone is mostly made of classical hardware. RIKEN’s Center for Computational Science has brought ROQUO, a GPU supercomputer for its quantum-HPC hybrid platform, into operation at its Kobe facility. The system reached 19.8 petaflops of FP64 performance in a full-system HPL benchmark, giving Japan a classical GPU platform for quantum simulation, algorithm development, and future integration with systems including Fugaku, IBM Quantum System Two “ibm_kobe,” and Quantinuum’s Reimei. ROQUO puts the classical stack first: GPUs, networking, cooling, and software before large-scale quantum-classical workflows. The launch does not prove quantum advantage, but it shows the infrastructure work required before hybrid quantum computing can become useful outside controlled research settings. Built primarily by DTS Corporation to RIKEN’s specifications, ROQUO has 135 NVIDIA GB200 NVL4 compute nodes, 540 NVIDIA Blackwell GPUs, 270 NVIDIA Grace CPUs, and NVIDIA Quantum-X800 InfiniBand networking at up to 3.2 terabits per second. RIKEN said the system reached 19.8 petaflops of measured FP64 performance, exceeding the design target for the core computing resource of its Quantum-HPC Hybrid Platform. ROQUO connects through the SQC Interface, a JHPC-quantum project API built to coordinate work between quantum computers and HPC systems. RIKEN has named the quantum systems ROQUO is expected to work with, including ibm_kobe and Reimei, but it has not yet shown public quantum-enabled scientific results at scale. For now, the HPL benchmark confirms that the classical GPU infrastructure is operational and has exceeded its design target. External access is expected through the JHPC-quantum test user program, so users should not read the launch as broad commercial availability. The GB200 NVL4 choice is important because RIKEN presents it as a four-GPU configuration suited to scientific and technical workloads, unlike the larger GB200 NVL72 configuration specialized for generative AI training and inference. Fault-tolerant quantum processors are not ready for broad production use, so GPU-heavy classical systems still handle the simulation, algorithm testing, and high-throughput workloads that future quantum-classical systems may require. ROQUO also gives RIKEN operational experience before FugakuNEXT, Japan’s next flagship supercomputer. RIKEN says FugakuNEXT is being developed with Fujitsu and NVIDIA, targets operation around 2030, and is designed for the convergence of AI, quantum computing, and HPC. That roadmap puts ROQUO in the same regional debate over compute capacity, vendor dependence, and technology sovereignty. ROQUO does not give APAC institutions a template to copy. It gives them a regional reference point for the infrastructure questions that come before useful hybrid quantum computing: low-latency networking, GPU capacity, software portability, power and cooling planning, and access rules. HPC operators will need to know whether existing networks, schedulers, and GPU clusters can support future quantum-classical workloads without forcing a complete rebuild. Japan’s phased approach also gives policymakers a model for building classical infrastructure now, connecting it to national systems, and using the interim period to develop software and operating expertise. ROQUO is benchmarked and operating, but its value will depend on what comes next: access rules, software documentation, workload portability, and published quantum-classical results. Japan’s approach raises a practical planning question across APAC: whether compute roadmaps account for the classical infrastructure hybrid quantum systems will require, from GPU capacity and cooling to AI infrastructure dependencies. Also read: Reliance’s AI roadmap ties India-scale AI services to local compute, telecom infrastructure, and enterprise compliance questions. If you can only read one tech story a day, this is it. With Qubitra, they intend to drive practical value by applying quantum approaches to some of the finance sector’s most complex problems. Australia could become the world’s first quantum computing nation — but it won’t come without controversy. We use cookies and other data collection technologies to provide the best experience for our customers. You may request that your data not be shared with third parties here: Do Not Sell My Data.