Fujitsu has unveiled its most ambitious offering yet for the artificial intelligence market. The Japanese company has confirmed that it will begin globally shipping its new Fujitsu Monaka CPU in November . This Arm-based processor with 144 cores is designed for AI inference workloads and environments that demand strict control over data and the supply chain. The announcement comes amid a race to reduce reliance on external accelerators and optimize energy consumption in data centers.
Fujitsu's offering goes beyond the chip. The company will also launch its own range of servers, the Fujitsu Monaka Server , designed and manufactured entirely in Japan, intended to cover tasks ranging from inference to advanced simulation, as well as areas such as defense and academic research. The initial timeline places Japan and Europe as the first markets for the servers, while the CPU will also be distributed in the United States and Asia-Pacific.
A CPU designed for inference without dedicated accelerators
The Monaka processor employs a 3D stacking architecture that combines a 2-nanometer manufacturing process for the cores with a 5-nanometer process for the cache and input/output functions. This separation optimizes costs and efficiency, according to the company. The maximum declared frequency is 3,8 GHz, with a memory transfer rate of 8.800 MT/s. The design also incorporates proprietary ultra-low voltage operating technology, a key element in minimizing power consumption.
The acceleration of matrix operations, essential in AI inference, is channeled through specific hardware instructions. This is complemented by SVE2 vector extensions and software optimization. Fujitsu claims that this combination allows for doubling inference performance compared to other CPUs and roughly halving both the number of servers and the energy required to process equivalent workloads. The company does not specify which benchmark systems it used to arrive at these figures, so we will have to wait for independent testing to validate the actual results.
One of Monaka's strengths is its ability to deliver practical inference performance using only the CPU, without requiring dedicated accelerators in all scenarios. This is particularly appealing to organizations that want to run models locally but whose facilities aren't equipped to accommodate high-density GPUs due to space, power, or cooling constraints. However, some server models will support GPU integration for expanded capabilities.
Security is also a central focus. Monaka incorporates confidential computing using Arm CCA, a technology that encrypts and isolates data and applications during in-memory processing, even from privileged software such as the operating system or hypervisor. This approach facilitates the execution of sensitive workloads in shared infrastructures or multi-tenant cloud environments, without relying solely on controls external to the processor.
1U, 2U and multi-node servers manufactured in Japan
Fujitsu will manufacture its Monaka servers at its Kasashima plant in Japan. The company explained that it will maintain detailed records of component origin and manufacturing history to ensure complete traceability. This transparency is a key advantage for organizations that need to know the provenance of their hardware or that handle confidential information subject to national regulations.
The initial range includes a 2U server with two CPUs and a configurable 1U model with one or two processors. Each CPU has 144 cores, and the configurations include RDIMM memory, SSD storage, and PCIe Gen6 expansion. The 2U system is designed for workloads such as AI agents, digital twins, and inference, while the 1U aims to offer more flexible scalability, from small deployments to larger installations.
The 1U server implements CDI/CXL technology, which combines disaggregated composable infrastructure with Compute Express Link. This allows memory and accelerators to be allocated beyond the physical limits of a single machine, helping to alleviate memory constraints that arise when running large models. Fujitsu aims to improve resource utilization and avoid siloed configurations.
Thermal management is another key strength. The 1U model can operate with air cooling at temperatures up to 40 degrees Celsius, and with water cooling up to 45 degrees Celsius when using liquid cooling. According to Fujitsu, its thermal technologies allow for up to 80% reduction in the energy used to cool the servers. However, this figure is an estimate and will depend on the specific configuration and the installation environment.
For the academic and HPC sectors, the company has developed a liquid-cooled, 2U multi-node system with four nodes, two CPUs per node, and eight processors per chassis. It is designed to concentrate computing power in a small footprint and run large simulations, surrogate models, or fluid dynamics analyses. Fujitsu plans to expand the product line in the future with high-density systems and rack-scale servers featuring automated operation and maintenance capabilities.
An integrated stack to keep data and operations under control
Monaka's hardware isn't coming alone. Fujitsu has announced that it will integrate it with its Kozuchi AI platform and Takane, its enterprise generative AI technology, as well as with models tailored to different sectors. The intention is to offer a vertical solution where processing, software, and the modeling layer coexist under a single data and operations management system.
This approach responds to the demands of digital sovereignty , a concept gaining traction in Europe and other regions. It's not just about running sensitive workloads in one's own facilities, but also about documenting the origin of components and limiting operational dependencies on third parties. Domestic manufacturing in Japan adds physical traceability, but it doesn't resolve all the sovereignty implications on its own; technical security, access policies, and model management also come into play.
For businesses, the decision to adopt Monaka will depend on several factors: software compatibility, actual equipment availability, operating costs, and the ability to integrate accelerators or shared resources. With the first phased shipments planned for April 2027 in Japan and Europe, Fujitsu will need to demonstrate in commercial configurations that its efficiency and inference figures hold up in practice.
In the short term, the launch of Monaka represents another step towards offering alternative processors in the server market, with a clear focus on AI inference and supply chain transparency. Its success will depend on Fujitsu's ability to convince data centers that a general-purpose CPU with dedicated acceleration can meet a significant portion of their needs, without requiring the installation of GPUs in every case.
Finally, availability in November 2026 will mark the starting point, but real adoption will be seen throughout 2027 as the servers are rolled out gradually and their performance can be compared with established solutions. The promise of doubling inference performance and halving power consumption will need to be validated with independent testing, something experts are eagerly anticipating.