Nvidia and RIKEN launch 2,140-GPU quantum and AI supercomputers

Signals Inbox·July 16, 2026·Quantum Computing

Japan has switched on a 540-GPU quantum supercomputer and is preparing a second 1,600-GPU machine for AI-driven science. The real story is not just 2,140 Blackwell GPUs. It is that RIKEN is building the bridge between today’s quantum processors, classical simulation, and the much larger FugakuNEXT system planned for around 2030.

Q1What actually happened?

According to RIKEN’s official announcement, its new ROQUO quantum-HPC supercomputer has started operating in Kobe. ROQUO uses 540 NVIDIA Blackwell GPUs. A second machine called RIKYU, built for AI-driven science, adds another 1,600 GPUs. Together, the two systems contain 2,140 GPUs.

Q2Why is this more than a big GPU cluster?

ROQUO is designed to connect classical supercomputing with real quantum processors through NVIDIA CUDA-Q. Quantum computers are still too small and noisy to work alone on most useful problems. The practical near-term model is hybrid: GPUs handle simulation, optimization, and error work, while quantum chips run the small parts where they may eventually have an edge.

Q3How powerful is ROQUO today?

RIKEN measured 19.8 petaflops of double-precision performance across ROQUO’s 135 compute nodes and 540 GPUs. That is real operating performance, not a future target. It also exceeded the system’s original design goal, which matters because the project has already moved from announcement to working infrastructure.

Q4What will the larger RIKYU system do?

RIKYU is aimed at AI for Science. That means training scientific models, finding patterns in huge research datasets, and speeding up work in areas such as materials, drugs, biology, climate, and engineering. Its 1,600 GPUs make it roughly three times larger than ROQUO by GPU count, but its job is different: it is the AI engine, while ROQUO is the quantum-HPC bridge.

Q5Why does this matter right now?

Because Japan is not waiting for a useful fault-tolerant quantum computer to appear. It is building the software, networking, simulation capacity, and researcher workflows first. That lowers the risk that expensive quantum hardware arrives later with no practical system around it. ROQUO lets researchers test hybrid workloads now, while RIKYU brings AI into the same scientific stack.

Q6How does this connect to FugakuNEXT?

These two machines are stepping stones toward FugakuNEXT, Japan’s next flagship supercomputer planned for around 2030. FugakuNEXT is expected to combine Japanese CPUs, NVIDIA GPUs, AI, simulation, and quantum links in one national platform. ROQUO and RIKYU give RIKEN a smaller place to build and test that model before the much larger system arrives.

Q7So is Japan ahead in quantum computing?

Not automatically. The United States, China, and Europe all have stronger quantum hardware programs in different areas. Japan’s advantage here is integration. It is treating quantum processors as one part of a wider scientific computing system, not as a standalone magic box. The next proof point is whether researchers use ROQUO and RIKYU to produce faster or better scientific results, not just impressive benchmark numbers.