Japan is preparing to launch a new national supercomputer, FugakuNEXT, developed by Fujitsu, NVIDIA, and the RIKEN Research Center. The system, expected to be operational in 2030, will combine modeling and artificial intelligence into a single platform. Graphics accelerators will be used for the first time in a project of this magnitude.
Fujitsu will reportedly build the CPU and integrate the system, NVIDIA will focus on GPU architecture and infrastructure, and RIKEN will focus on software and algorithms. Future NVIDIA Feynman GPUs are expected to be used as accelerators around 2028. For the processor, Fujitsu is preparing the MONAKA-X, a successor to the MONAKA chip with a higher core count and expanded AI task capabilities.
FugakuNEXT's performance in FP600 should exceed 8 EFLOPS, making it the most powerful AI supercomputer announced. It is also planned to accelerate applications nearly 100 times faster than the current Fugaku accelerator while maintaining power consumption at 40MW.
To achieve these targets, developers will not only use new hardware but also innovative methods, ranging from surrogate models and hybrid computational precision to neural networks that consider the physical properties of the process. The project is positioned as an "AI-HPC platform" for science, industry, and research, and could become the foundation for the next generation of supercomputing.
