HPC and AI and Quantum-HPC in Japan

In this session, Satoshi Matsuoka of RIKEN outlines Japan's strategy at the intersection of HPC, AI, and quantum computing. He argues for moving from a 20th-century model of science toward an automated 'industrial revolution of science' in which principal AI scientist agents work alongside humans and domain-specific tools to drive discovery loops. Matsuoka stresses that AI for science extends well beyond hosting large language models to diverse non-lingual models, including simulation surrogates, physics-informed neural networks, generative models for terabyte-scale images, and graph neural networks. He describes about a billion US dollars in Japanese government funding, including Fugaku Next design and roughly 250 million dollars for domain-science programs, and a modest RIKEN cluster of about 2,000 Grace Blackwell GPUs (two GB200 NVL4 systems) running a hardened software stack derived from Virtual Fugaku plus AI components and the REVOLT secure agent sandbox. He highlights international collaboration, notably Japan becoming a primary partner in the US DOE Genesis mission with matched half-billion-dollar investments, and award-winning joint work with Oak Ridge. On quantum, Japan operates IBM Heron and Quantinuum systems in a hybrid quantum-HPC infrastructure, reporting results such as a 4Fe-4S iron-sulfur structure determination and a 12,635-atom protein simulation with IBM and Cleveland Clinic. Fugaku Next, developed with Fujitsu and NVIDIA, will use the 1.4nm Monaka X CPU with NVLink Fusion, targeting a 100x application speedup driven largely by software and algorithms, including FP8 techniques.