Nvidia Teams With RIKEN and Fujitsu on FugakuNEXT Supercomputer Push

Japan’s RIKEN research institute said on 22 August that U.S. chipmaker Nvidia will join RIKEN and long-time partner Fujitsu in designing and building the successor to the Fugaku supercomputer. The project, code-named “FugakuNEXT,” is the first time a foreign vendor has been invited to co-develop Japan’s flagship high-performance computing (HPC) system. FugakuNEXT will pair Fujitsu’s forthcoming MONAKA-X Arm-based central processors with Nvidia graphics processing units linked by the companies’ new NVLink Fusion interconnect. RIKEN expects the hybrid architecture to deliver more than a five-fold improvement in raw hardware performance and as much as a 100-fold boost in real-world application speeds versus Fugaku, which led the TOP500 list in 2020 but now ranks seventh. The system targets more than 600 exaFLOPS in FP8 precision—an early step toward zettascale computing—and is scheduled to enter service in Kobe around 2030. The institute says the new machine will be optimised for artificial-intelligence training as well as traditional simulation workloads, supporting research in drug discovery, advanced manufacturing and climate modelling. Adding Nvidia’s GPU technology signals Japan’s shift toward AI-centric HPC design rather than merely chasing top LINPACK benchmark scores. Investors welcomed the collaboration. Fujitsu shares jumped to their highest level in 25 years after the announcement, even as other semiconductor-related names such as Advantest fell in Tokyo trading.