Introduction
On September 14, 2026, Fujitsu announced the commercial launch of the FUJITSU-MONAKA — a 144-core datacenter processor built on a 2nm process, designed and manufactured entirely in Japan. Sales begin in November 2026 across Japanese and European markets. Behind the hardware announcement lies a pointed commercial proposition: halve the server footprint required to run a given AI inference workload.
An Inference-First CPU Architecture Built for Production
The MONAKA is built on the ARM Armv9.3-A architecture, featuring 256-bit SVE2 extensions optimized for HPC and AI vector workloads. Its heterogeneous chiplet design pairs a compute die fabricated at 2nm with SRAM and I/O dies at 5nm. The result: 144 cores per socket, 12 DDR5 channels running at 8,800 MT/s, PCIe 6.0 connectivity with CXL 3.0 support — enough bandwidth to feed memory-intensive inference workloads without bottlenecks. Maximum clock speed reaches 3.8 GHz.
Fujitsu offers two SKUs: a high-efficiency variant at 350W and a high-performance variant at 500W. The company claims twice the AI inference throughput of a comparable CPU at half the power consumption under equivalent load. If those figures hold up in real-world conditions, the procurement implications are immediate and significant.
What This Means in Practice for Infrastructure Teams
For the same inference workload, organizations would need half the number of machines. In operational terms: fewer rack units, lower kilowatt-hour consumption, fewer maintenance contracts — and potentially server rooms capable of absorbing twice the workload without physical expansion.
The FUJITSU MONAKA server ships in 1U and 2U configurations for inference and AI agent deployments, with a 2U multi-node option for HPC and academic use cases. Critically, it runs on air cooling. For datacenters that have held back on GPU server adoption due to liquid cooling constraints, that removes a meaningful barrier to scaling AI infrastructure.
Supply Chain Traceability and Sovereignty: A Differentiator for European CIOs
The competitive gap with rival silicon doesn't come down to specs alone. The MONAKA is developed and manufactured at Fujitsu's Kasashima plant in Japan. The company highlights documented traceability of component origins and full manufacturing history — a rare attribute in an industry where supply chains routinely cross a dozen countries before a chip reaches a rack.
For organizations operating under technology sovereignty mandates — government agencies, critical national infrastructure operators, defense — this traceability offers a concrete answer to regulatory obligations that are tightening across Europe. Fujitsu's decision to launch first in Japanese and European markets is no accident: it is a deliberate positioning move.
A Signal on the CPU/GPU Balance in the Datacenter
The MONAKA enters a broader industry debate about the case for CPUs versus GPUs in inference workloads. GPUs dominate model training, but production inference is less demanding in raw compute and far more sensitive to energy efficiency and processing density per watt. Several hyperscalers and cloud operators are already exploring hybrid deployment strategies.
Fujitsu is not trying to displace the GPU for training — it is targeting the stage where most enterprises actually spend their AI budgets: running workloads in production, day in and day out.
For CIOs planning server investments on a twelve- to eighteen-month horizon, the MONAKA now belongs on the shortlist of options to evaluate seriously — particularly where supply chain sovereignty already features in procurement criteria.