FugakuNEXT: Japan’s $750M Zetta-Scale Supercomputer Blends Nvidia GPUs and Fujitsu Arm CPUs for the Future of AI-Driven Science

Japan's FugakuNEXT: Pushing Supercomputing Beyond Exascale

Japan is set to advance its position in high-performance computing with FugakuNEXT, a hybrid AI-HPC supercomputer targeting performance beyond exascale into the zetta-scale range. Announced on August 22 in Tokyo, this project is a collaboration among RIKEN, Fujitsu, and Nvidia, combining Japan’s domestic CPU technology with U.S.-based GPUs to tackle scientific and industrial challenges anticipated over the next decade.

Legacy of Japanese Supercomputing

Fugaku, deployed in 2020, held the world’s fastest supercomputer title for two years and was instrumental during the COVID-19 pandemic, supporting vital simulations that informed research and policy. Although currently ranked seventh globally, Japan is already preparing the next generation. FugakuNEXT is planned to be operational by 2030 at RIKEN’s Kobe campus, with a development budget exceeding 110 billion Yen (approximately $740 million USD).

This project continues a tradition of flagship Japanese supercomputers, following milestones such as the Earth Simulator (2002), the K computer (2011), and Fugaku itself, each setting records in computational speed during their time.

Technical Innovations in FugakuNEXT

Unlike earlier systems, FugakuNEXT will integrate GPUs at its core. Fujitsu is developing the MONAKA-X Arm CPUs, while Nvidia provides GPUs and co-designs the interconnect fabric via NVLink Fusion, a high-bandwidth link between CPU and GPU. This architecture is intended to efficiently support both conventional simulations and advanced AI workloads.

RIKEN highlights that FugakuNEXT marks Japan’s first flagship supercomputer to incorporate GPUs, reflecting a growing global trend of merging HPC and AI computing technologies.

Performance and Efficiency Goals

FugakuNEXT aims for a fivefold increase in raw hardware performance over Fugaku, combined with a twentyfold improvement from software and algorithmic optimizations such as mixed-precision computing and physics-informed neural networks. Together, these are expected to enable a hundredfold increase in real-world application performance.

Targeted peak performance reaches approximately 600 exaFLOPS in FP8 sparse precision, potentially making FugakuNEXT the world’s first zetta-scale supercomputer. Notably, this significant performance gain will occur within the same 40-megawatt power envelope as Fugaku, emphasizing energy efficiency.

AI Integration and Scientific Impact

FugakuNEXT is positioned as an integrated AI-HPC platform that supports “AI for Science.” It intends to automate key parts of the research process, including hypothesis generation, experiment simulation, and validation. The system is expected to accelerate discovery across various fields such as climate modeling, drug development, disaster resilience, and advanced manufacturing.

Nvidia’s software stack will be fully integrated, including CUDA-X libraries for quantum simulation and data science, TensorRT for AI inference, and NeMo for large language model training. This integration underlines AI’s expanding role as a scientific tool.

Strategic Importance for Japan

For Japan, FugakuNEXT represents both a scientific resource and a strategic investment. The government views it as vital for strengthening domestic semiconductor technology and maintaining leadership in AI and HPC innovation.

At the launch, Satoshi Matsuoka, director of RIKEN’s Center for Computational Science, described the partnership with Nvidia as a “major strategic move” that will enhance Japan’s capabilities and encourage wider adoption of Japanese CPU technologies globally.

Looking Ahead

While the project’s budget exceeds $750 million, FugakuNEXT’s objectives are ambitious: to pioneer large-scale integration of HPC and AI, serve as a national research and industrial platform, and establish Japan as a key player in the emerging zetta-scale era. Success will redefine the role of national supercomputers in advancing scientific research and societal progress.

For professionals interested in AI and HPC advancements, FugakuNEXT exemplifies the trajectory of supercomputing infrastructure adapting to the growing demands of AI-driven science. More on AI course offerings and training for emerging technologies can be found at Complete AI Training.