RIKEN publishes the FugakuNEXT basic design: 1.4 nm MONAKA-X CPUs paired with NVIDIA GPUs under a 40 MW cap

RIKEN/FugakuNEXT

RIKEN released the technical report on FugakuNEXT's basic design, carried out with Fujitsu from June 2025 and NVIDIA from August 2025 through February 2026. It fixes the overall shape: MONAKA-X CPUs that can run Fugaku's programs, coupled to NVIDIA GPUs, under a 40 MW power limit. GPU memory, scale-up size and the CPU-GPU link are left for detailed design.

• Targets: 50+ EFLOPS effective AI performance with zetta-scale peak, and at least 5-10x faster existing simulation codes.
• MONAKA-X: 1.4 nm, 3D chiplets, many cores, vector (SVE2) and matrix (SME2) engines.
• GPU scale-up options compared: 72-GPU switched racks vs 4-GPU nodes.
• NVLink Fusion is the lead option for the CPU-GPU link; decision due by end-2027.
• Operation is still planned around 2030.