Fujitsu will begin selling its next-generation processor to the wider market in November, and the pitch is as much political as technical. FUJITSU-MONAKA, an Arm-based CPU designed in Japan, anchors a server line the company calls sovereign AI infrastructure and will be built at home.
The silicon pairs a 2nm compute core with 5nm cache and I/O dies in a 3D stack, clocks up to 3.8GHz and moves memory at 8800MT/s. Dedicated matrix instructions, SVE2 vector operations and software tuning handle inference work; Fujitsu claims double the AI throughput of competing CPUs and says that cuts server count and power draw roughly in half for the same job. Arm CCA confidential computing comes included for shared clouds.
Assembly happens at the Kasashima plant, which built the Fugaku supercomputer, giving buyers a traceable record of where parts came from and how they were made. Japan and Europe get first shipments, and the defense sector appears on the customer list. Air cooling keeps the 1U model running in ambient air up to 40C, with cooling power down by as much as 80%, so the systems skip water infrastructure.
Fujitsu is chasing organizations that want inference on their own floors and out of foreign supply chains, extending a domestic AI server effort it started earlier in 2026.
