Fujitsu launches MONAKA chip for sovereign AI servers

Fujitsu has launched the FUJITSU-MONAKA processor and Fujitsu MONAKA Server for sovereign AI infrastructure.

It plans to sell the processor globally as a stand-alone product to cloud operators, data centre operators and server vendors, while also offering rack-mounted servers built around the chip. The server line is aimed at customers seeking AI systems built on hardware designed, developed and manufactured in Japan.

The launch puts Fujitsu into more direct competition for AI infrastructure spending as data centre operators face pressure over electricity use, cooling demands and hardware availability. It also reflects a broader push among governments and regulated industries for tighter control over where AI systems are built, how data is handled and how supply chains are monitored.

The processor uses a 3D-stacked design, with a 2nm process for the core and 5nm for the cache and input-output sections. Fujitsu said the chip reaches a maximum operating frequency of 3.8GHz and a memory transfer speed of 8800MT/s.

According to Fujitsu, the processor delivers twice the AI inference throughput of other CPUs and greater power efficiency, allowing users to halve the number of servers and the electricity needed for the same workloads. The CPU includes dedicated instructions for matrix operations used in AI inference, alongside SVE2 vector operations and software optimisation.

That focus on inference is central to the launch. While much of the market's recent investment has centred on training large AI models, companies are increasingly focused on the cost of running those models in production, particularly for internal systems that need predictable power use and tighter operational control.

Cooling limits

The MONAKA Server is designed to address one of the practical constraints in AI deployment: cooling. Fujitsu said its 1U server can operate in ambient temperatures of up to 40C with air cooling, avoiding the need for specialised liquid-cooling infrastructure in some installations.

It also said cooling power consumption can be cut by as much as 80%. If that is borne out in customer deployments, it would matter for operators trying to add AI systems to existing facilities where power and floor space are already constrained.

The server range will include 2U systems with two CPUs and 1U systems with either one or two CPUs. Fujitsu also plans to offer 2U multi-node servers for academic and high-performance computing use, with four nodes in total and two CPUs per node.

For the 1U server, Fujitsu is using CDI and CXL technology so memory and accelerators can be allocated beyond the physical limits of a single server. It said this approach is intended to reduce memory bottlenecks in AI inference and make it easier to adapt systems to larger models and changing workloads.

Sovereignty focus

A key element of the launch is its emphasis on domestic production. Fujitsu said the MONAKA Server is designed, developed and manufactured in Japan, with production at its Kasashima Plant. That allows the origin of components and manufacturing history to be tracked, which the company sees as important for supply chain transparency.

That message is likely to resonate most strongly with public-sector buyers, defence users and operators in regulated industries. The server will be made broadly available to data centre operators, enterprises, and academic and HPC customers in Japan and Europe, as well as to the defence sector.

The processor also includes confidential computing, which encrypts data and applications while they are running in memory. Fujitsu said this is intended to support secure AI use in sensitive environments, including multi-tenant cloud systems.

Fujitsu tied the hardware launch to a broader strategy around sovereign AI systems. It said the server will sit at the core of an integrated model combining its own AI software, including Fujitsu Kozuchi and Takane, with industry-specific models. That would give Fujitsu a fuller stack to offer customers seeking both infrastructure and AI tools under a more controlled operating model.

The move comes as Japanese and European technology groups seek to carve out positions in AI infrastructure markets dominated by US chipmakers and hyperscale cloud groups. Rather than competing solely on raw scale, suppliers such as Fujitsu are increasingly focusing on energy consumption, manufacturing location, auditability and security features.