Nvidia will supply its GB200 NVL4 systems to Riken for use in two new supercomputers. One machine will be an AI system for scientific work and the other will support quantum computing research.
A total of 2,140 Blackwell GPUs will go into the two systems: 1,600 GPUs in the AI machine and 540 in the quantum system. Both units will use Nvidia’s GB200 NVL72 platform and will be linked with Quantum‑X800 InfiniBand networking.
“Riken has long been one of the world’s great scientific institutions, and today it stands at the forefront of a new era in computing,” said Ian Buck, VP of hyperscale and high-performance computing (HPC) at Nvidia.
Nvidia said this build expands on a project announced in August 2025 with Riken and Fujitsu to design the follow-up to the Fugaku supercomputer, known as FugakuNEXT. FugakuNEXT will pair Fujitsu’s Arm-based Fujitsu‑MONAKA‑X CPUs with Nvidia GPUs. The MONAKA‑X chips, made on 2nm process technology and expected in 2027, are designed for tight chip packing and very low power use.
Riken plans to connect MONAKA‑X CPUs and Nvidia GPUs using NVLink Fusion, which joins custom CPU designs and Nvidia GPUs through a chiplet UCIe link. The two new systems are due to start work in spring 2026, while FugakuNEXT aims for service by 2030.
Nvidia revealed these updates at the Supercomputing 2025 event in St. Louis, where it announced several partnerships and system plans.
Leave a comment