Fujitsu is unveiling a global sovereign AI infrastructure built around the FUJITSU-MONAKA CPU, with availability kicking off in November 2026 and a broader rollout across data centers, enterprises, academia, HPC, and defense sectors in Japan and Europe.

Alongside the processor, Fujitsu introduces the FUJITSU-MONAKA Server, a Made-in-Japan server designed to deliver sovereign AI infrastructure with high reliability, traceability, and transparent supply chains through domestic manufacturing.

The MONAKA Server will be designed and manufactured in Japan at Kasashima Plant, offered in 2U dual-CPU and 1U single/double-CPU configurations starting in November 2026, optimized for AI agents, AI inference, and energy-efficient cooling.

1U models feature disaggregated memory and accelerators via CDI/CXL, operate in ambient temperatures up to 40C with air cooling and up to 45C with water cooling, and can cut cooling power by as much as 80%, addressing AI memory shortages.

The MONAKA processor supports confidential computing and Arm CCA, enabling secure AI use in multi-tenant environments and protecting data in memory from privileged software.

Fujitsu positions the product as a sustainable, sovereign AI infrastructure that supports autonomy and accountability in Japan and Europe for data centers, enterprises, academia, and HPC, contributing to national security and AI infrastructure expansion.

The MONAKA Server emphasizes space efficiency and power effectiveness, suitable for air-cooled data centers and scalable through future resource pooling and advanced cooling technologies.

Fujitsu plans a vertically integrated sovereign AI stack—Kozuchi AI platform, Takane, and industry-specific models—centered on traceability, security governance, and domestic data management.

The MONAKA initiative is linked to a NEDO-supported project, underscoring government-backed innovation and domestic development.

The FUJITSU-MONAKA CPU reaches up to 3.8 GHz with memory at 8800 MT/s, delivering twice the AI inference throughput and improved power efficiency versus competing CPUs, reducing required servers and power by about half for similar workloads.

This CPU uses a 3D-stacked design with a 2-nanometer core and 5-nanometer cache/I/O, delivering up to 3.8 GHz, 8800 MT/s memory, and hardware AI inference acceleration.

For academia and HPC, Fujitsu will offer 2U multi-node servers (two CPUs per node, four nodes total) in Japan and Europe for high-performance workloads like surrogate modeling and fluid analysis.