Fujitsu revealed a detailed roadmap for its new Arm architecture Monaka processor at the "Technology Update 2025" technology update event, clearly positioning the Monaka series as the company's next-generation critical data center CPU platform. The first batch of products is planned to be mass-produced in 2027 and adopt a 2 nm process technology.
The first Monaka processor is based on the Armv9-A instruction set architecture and adopts a 3D Chiplet stacking design. The core die is packaged with independent SRAM die and I/O die in the same package. Each processor integrates 144 CPU cores. Through dual-channel configuration, it can achieve a maximum scale of 288 cores on a single node. The platform supports 12-channel DDR5 memory, PCIe 6.0 interface, and CXL 3.0 interconnect, and is equipped with Arm SVE2 vector extensions for accelerating loads such as AI inference and high-performance computing (HPC). Fujitsu positions it as an energy-efficient data center processor for scenarios such as AI inference, numerical simulation, and large-scale data processing.
On this basis, Fujitsu has planned a more specialized follow-up product form. The next stage of the roadmap is Monaka-X, which is expected to be launched at the end of 2029. This is a CPU-only product that will use a 1.4 nm-level process technology. Fujitsu said that Monaka-X will be the first product to implement Arm SME (Scalable Matrix Extension) on server CPUs, aiming to further enhance matrix computing capabilities to serve AI and HPC applications. The processor continues the 3D multi-chip multi-core layout and emphasizes the tightly coupled design with the GPU. At the same time, hardware security capabilities related to confidential computing are integrated into the platform as standard configuration to meet the data center's needs for security isolation and privacy computing.
The next step on the roadmap is the "Monaka-X with NPU" version planned for the second half of 2030. This version adds an in-package NPU (Neural Network Processing Unit) to the CPU, targeting local acceleration for workloads such as medium-scale large language models (LLM). Fujitsu is exploring the introduction of reconfigurable computing engines and dedicated quantization accelerators into the platform to better support the requirements for different accuracy and model structures during the inference process, and to strike a balance between energy efficiency and throughput.
According to the established route, Monaka-XX will be ushered in 2031. This generation is described as a “complete CPU–NPU fusion chip”. It also plans to use 1.4 nm or more advanced process nodes to deeply integrate general computing and intelligent acceleration on a single platform. Through this series of product iterations, Fujitsu aims to further advance in the direction of high core count and high energy efficiency in the Arm server market. At the same time, it continues to enhance the native AI acceleration capabilities at the architectural level, making AI inference and high-performance computing capabilities a basic feature of the data center platform rather than a plug-in option. The overall roadmap shows that Fujitsu is using a new generation of process technology, 3D Chiplet packaging and Arm ecological expansion to establish an Arm server product line from general-purpose efficient computing to deep AI integration between 2027-2031.