The brief

Analyst view

• Operates Fugaku in Kobe (442 PFLOPS on the TOP500 benchmark), world number one in 2020 and 2021.
• FugakuNEXT, with Fujitsu CPUs and NVIDIA GPUs, targets 600+ EFLOPS for AI in FP8 around 2030.
• Published the FugakuNEXT basic design report on 29 May 2026; GPU, network and link choices come in detailed design.

Technical approach

As reported

Apps first

Designed around real science codes

Hardware options are scored by how much they speed up real weather, earthquake, materials and drug codes (co-design), not by peak benchmarks.

Japanese CPUs

Own Arm processors

Fujitsu designs the CPUs (A64FX in Fugaku, MONAKA-X next). MONAKA-X is meant to run Fugaku's compiled programs as they are, and keeps chip know-how in Japan.

AI + simulation

Mixing simulation and AI

FugakuNEXT adds GPUs so physics simulations can be sped up or replaced in parts by AI models trained on them.

Power cap

A fixed ~40 MW budget

FugakuNEXT must fit about 40 MW, as Fugaku did, so gains must come from more efficient chips and smarter algorithms, not more electricity.

Show all 4