Fujitsu has bagged the contract to design Japan’s next-gen supercomputer to succeed the Fugaku system, and it looks set to be another Arm-based behemoth, using a CPU derived from its upcoming MONAKA datacenter silicon.
According to the IT giant, Japan’s High-Performance Computing Infrastructure (HPCI) Program Steering Committee has chosen the RIKEN research institute as the primary entity responsible for developing the nation’s next-gen flagship super, which in turn has awarded Fujitsu the contract to design it.
Provisionally named “FugakuNEXT,” the contract covers the overall system, including compute nodes plus the processors used, and the basic design phase is scheduled to run until February 27, 2026.
No details of the contract or its value were specified. We asked Fujitsu for further information, and will update if we get a response.
Likewise, little has been disclosed about the forthcoming system, but the original Fugaku became the world’s most powerful supercomputer, leading the Top500 rankings from when it became operational in 2020, until it was overtaken by the US Frontier exascale computer in 2022. It seems likely that RIKEN and Fujitsu may also aim to produce an exascale unit.
What is known is that FugakuNEXT will build on Fujitsu’s established supercomputing expertise, with the processor incorporating technology from the company’s MONAKA general-purpose CPU currently under development and expected in 2027, which means the computer won’t be ready before then.
Tentatively named “MONAKA-X,” the processor intended for use in FugakuNEXT will not only inherit and accelerate “existing Fugaku application assets,” but also incorporate state-of-the-art AI acceleration capabilities to meet growing AI demands, Fujitsu said.
MONAKA is set to be fabricated using a 2nm production process, and will feature up to 144 ArmV9 cores with support for SVE2 vector extensions. It will employ a chiplet architecture with SRAM tiles stacked atop the processor tiles.
In contrast, the Fujitsu-made A64FX processor that powers Fugaku is a 48-core chip, but each has 32 GB of high-performance HBM2 stacked memory, and our guess is that MONAKA-X will use a newer HBM variant rather than SRAM.
The Rhea1 processor used in Jupiter, Europe’s first exascale super, features 80 Arm Neoverse V1 cores and HBM2 stacked memory.
“Fujitsu is determined to build a system that can dynamically meet customer needs, drawing on our invaluable experience from Fugaku and the cutting-edge technologies of MONAKA and MONAKA-X. By forging diverse partnerships, we will deliver the powerful computing infrastructure that society and industry need to succeed,” said Vivek Mahajan, Fujitsu Corporate VP and CTO in charge of System Platforms.
Mahajan added that Fujitsu would be “channeling the expertise cultivated through this project” into the development of next-generation Neural Processing Units (NPUs) and other advanced AI processors, “taking Made-in-Japan technology from Japan to the world and spearheading the adoption and acceleration of AI for businesses and society.” ®
