Fujitsu Limited announced on September 14, 2026, from Kawasaki, Japan, that it will begin global sales of a new Arm-based server processor called FUJITSU-MONAKA, paired with a companion system called the Fujitsu MONAKA Server. The company is positioning both as “made-in-Japan” sovereign AI infrastructure, built to give governments, enterprises, and research institutions a domestically manufactured alternative for running AI inference workloads without depending on offshore chip supply chains.
The announcement lands at a moment when sovereign compute has become a policy talking point as much as a technical one. Countries from Japan to the EU are weighing how much of their AI infrastructure and server hardware should sit on chips, software, and soil they fully control. Fujitsu’s answer is a 144-core chiplet CPU built on a 2nm process, manufactured at its Kasashima plant, with global sales opening in November 2026.
What Fujitsu Actually Announced on September 14
According to Fujitsu’s own release, the company will launch “the made-in-Japan Fujitsu MONAKA Server equipped with the FUJITSU-MONAKA CPU, offering high reliability and power efficiency as a sovereign AI infrastructure,” a description Fujitsu Limited repeated across its official press release. Two products carry the news: FUJITSU-MONAKA, the CPU itself, and the Fujitsu MONAKA Server, the rack-level system built around it.
Fujitsu says the server will “be made broadly available from November 2026 to data center operators, enterprises, and the academic and HPC sectors in Japan and Europe, contributing to national security,” a framing that puts the launch squarely inside the sovereign-AI narrative rather than a general-purpose server refresh. Order acceptance for the MONAKA Server also opens in November 2026, per the announcement.
Trade press moved fast on the story. StorageReview and HPCwire both picked up the release within hours, alongside PR Newswire, TechPowerUp, Telecompaper, Unite.AI, and XenoSpectrum, underscoring how closely the HPC and enterprise-server press is watching Japan’s push into domestic AI silicon.
Inside FUJITSU-MONAKA: 144 Cores on a 2nm Chiplet Stack
FUJITSU-MONAKA is an Arm-based processor built on the Armv9 architecture with 256-bit SVE2 vector extensions, packing 144 cores into a single socket. Instead of one monolithic die, Fujitsu built the chip as a 3D chiplet stack: compute cores sit on a 2nm process die, while SRAM and I/O logic live on a separate 5nm die bonded to it. That split lets Fujitsu use the most advanced, and most expensive, node only where it matters for raw compute density, while keeping cache and I/O logic on a cheaper, more mature process.
Memory bandwidth is handled through 12 channels of DDR5, and the chip exposes 96 lanes split between PCIe 6.0 and CXL 3.0 for external I/O and memory expansion. In its release, Fujitsu said FUJITSU-MONAKA “integrates Fujitsu’s extensive expertise in processor and server development, achieving world-class computational performance and data supply capabilities with a maximum operating frequency of 3.8GHz and memory transfer speed of 8800MT/s.”
That combination, a 144-core Arm chip with wide DDR5 memory channels and CXL 3.0 support, targets two workload classes explicitly: AI inference and traditional HPC. Fujitsu is not trying to compete with GPU accelerators head-on. It is building a CPU that can feed AI workloads with data fast enough that fewer accelerator-heavy racks are needed for the same throughput.
FUJITSU-MONAKA Technical Specifications
| Spec | FUJITSU-MONAKA |
|---|---|
| Architecture | Arm, Armv9-generation with SVE2 (256-bit vector width) |
| Cores per socket | 144 |
| Die construction | 3D chiplet: compute die on 2nm, SRAM/I-O die on 5nm |
| Max operating frequency | 3.8GHz |
| Memory | DDR5, 12 channels, 8800MT/s transfer speed |
| External I/O | 96 lanes, PCIe 6.0 and CXL 3.0 |
| Target workloads | AI inference, HPC |
| Manufacturing site | Kasashima plant, Ishikawa Prefecture, Japan |
Those numbers put FUJITSU-MONAKA in the same conversation as the highest core-count server chips shipping today. shattered.io covered Intel’s Xeon 6+ line reaching 288 cores on its 18A process earlier this year, and AMD’s Threadripper Halo Station workstation platform packing 96 cores with 576GB of HBM3E. FUJITSU-MONAKA’s 144-core count sits between those two, but its pitch is different: it is not chasing the highest core count on the market, it is chasing sovereign manufacturing and AI-inference efficiency per rack.
The Sovereign AI Argument, Explained
“Sovereign AI infrastructure” has become shorthand for a specific worry: that a country’s AI capacity, from chips to the software stack running on them, depends on foreign vendors it cannot fully audit or control in a crisis. Japan is not alone in raising this concern. The EU has debated similar cloud-sovereignty rules this year, and shattered.io reported on how that policy push is splitting vendors trying to serve both sovereign-cloud buyers and hyperscale customers at once.
Fujitsu’s pitch folds neatly into that anxiety. A domestically designed, domestically manufactured CPU, running in a domestically built server, sold first into Japan and Europe rather than the US or China, reads as a deliberate positioning choice. Fujitsu is betting that government and enterprise buyers in Japan and the EU will pay a premium, or at least prioritize procurement, for infrastructure they can trace end to end back to a single, allied manufacturing base.
That framing also explains why Fujitsu is emphasizing “reliability and power efficiency” rather than leading with raw benchmark numbers against Nvidia or Intel. The MONAKA Server is not being sold as the fastest AI system available. It is being sold as the AI system a sovereign buyer can fully account for, from the fab to the rack.
Made in Japan: The Kasashima Plant and Manufacturing Timeline
Fujitsu says it will manufacture “Made in Japan” sovereign AI servers at the Fujitsu Group’s Kasashima Plant in Ishikawa Prefecture, with manufacturing beginning in March 2026. Production of servers actually equipped with the FUJITSU-MONAKA CPU is set to start within fiscal year 2026, which for Fujitsu runs through March 31, 2027. Global sales of the standalone FUJITSU-MONAKA CPU begin in November 2026, the same month the MONAKA Server becomes broadly available and starts accepting orders in Japan and Europe.
That sequencing matters. Fujitsu is not announcing a chip and leaving customers to wait years for silicon to show up in a shippable product. The gap between the March 2026 manufacturing start and the November 2026 commercial rollout is about eight months, tight by server-industry standards, which suggests Fujitsu has been running MONAKA through validation well before this week’s public announcement.
Rollout Timeline at a Glance
| Milestone | Timing |
|---|---|
| Manufacturing of Made in Japan sovereign AI servers begins at Kasashima Plant | March 2026 |
| Production of servers equipped with FUJITSU-MONAKA CPU | Within fiscal year 2026 (through March 31, 2027) |
| Public announcement of FUJITSU-MONAKA and MONAKA Server | September 14, 2026 |
| Global sales of FUJITSU-MONAKA CPU begin | November 2026 |
| MONAKA Server order acceptance and broad availability (Japan, Europe) | November 2026 |
The AI Inference Efficiency Claim
Fujitsu’s headline performance claim is about efficiency, not peak throughput. The company said FUJITSU-MONAKA “delivers twice the AI inference throughput and superior power efficiency compared to other CPUs, effectively halving the number of servers and power consumption required for equivalent processing loads.” Fujitsu did not name the specific CPUs used as the comparison baseline in the public release, so that 2x figure should be read as a company-supplied benchmark rather than an independently verified one.
Still, the claim is consistent with the chip’s design choices. Wide DDR5 memory channels, high-speed CXL 3.0 links, and a high core count are exactly the ingredients that help a CPU keep AI inference pipelines fed with data instead of stalling on memory bandwidth. If Fujitsu’s numbers hold up under third-party testing after the November launch, the “half the servers, half the power” pitch could matter a lot to data center operators already squeezed by rising electricity costs and rack space constraints.
That squeeze is real right now. shattered.io has tracked the broader memory crunch hitting the server market this year, including a memory chip shortage that has pushed stockpiles below 10 days of supply industry-wide. A CPU that promises to cut the number of servers needed for a given AI workload is a direct answer to that pressure, whether or not Fujitsu intended the timing.
How FUJITSU-MONAKA Compares to Other High-Core-Count Server CPUs
Fujitsu is entering a server CPU market that already has several high-core-count Arm and x86 competitors, each chasing a different mix of core density, memory bandwidth, and workload specialization. The table below lines up FUJITSU-MONAKA against processors shattered.io has covered previously, using only the specs each vendor has published.
| Processor | Vendor | Cores | Process / Memory | Primary Positioning |
|---|---|---|---|---|
| FUJITSU-MONAKA | Fujitsu | 144 | 2nm compute die / 5nm I-O die, 12-ch DDR5 | Sovereign AI inference and HPC |
| Xeon 6+ | Intel | 288 | 18A process | General-purpose data center, cloud |
| Threadripper Halo Station | AMD | 96 | 576GB HBM3E | High-end workstation, local AI |
| Arm-based cloud CPUs (Graviton-class) | Various hyperscalers | Varies by generation | Custom, cloud-only | Hyperscale cloud instances, not sold as standalone hardware |
The comparison shows Fujitsu is not trying to win on core count alone. Intel’s Xeon 6+ nearly doubles FUJITSU-MONAKA’s core density, and AMD’s Threadripper Halo Station leans on HBM3E for bandwidth that DDR5 cannot match. What FUJITSU-MONAKA offers instead is a fully domestic supply chain, from a 2nm compute die to a Japan-built server chassis, something none of the above competitors currently sell as a packaged, sovereign-manufactured system.
Fujitsu’s Longer History Building Its Own Silicon
Fujitsu is not a newcomer to custom Arm processors. The company has spent more than a decade designing its own chips for supercomputing and enterprise servers rather than buying merchant silicon off the shelf, work that grew out of its long partnership with Japan’s national research institute RIKEN on Arm-based supercomputer processors. FUJITSU-MONAKA extends that in-house design lineage into the AI-inference era, applying the same design-our-own-silicon-for-our-own-sovereign-use-case playbook that shaped Fujitsu’s earlier HPC processors.
That history is part of why the sovereign framing is credible rather than purely marketing. Fujitsu already runs its own chip design teams, has an existing manufacturing relationship through the Kasashima plant, and has shipped custom Arm silicon into government and research procurement before. MONAKA is a continuation of a strategy the company has followed for years, not a pivot invented for this announcement.
What Industry Voices Are Saying
Fujitsu built its public case for FUJITSU-MONAKA around a small set of specific technical and strategic claims, repeated across its official channels and picked up by wire services. On the positioning of the new server line, Fujitsu Limited said: “Additionally, the company will launch the made-in-Japan Fujitsu MONAKA Server equipped with the FUJITSU-MONAKA CPU, offering high reliability and power efficiency as a sovereign AI infrastructure” (Fujitsu press release via PR Newswire).
On the availability window, Fujitsu Limited said the MONAKA Server “will be made broadly available from November 2026 to data center operators, enterprises, and the academic and HPC sectors in Japan and Europe, contributing to national security” (Fujitsu press release via PR Newswire).
On the chip’s technical profile, Fujitsu Limited described FUJITSU-MONAKA as integrating “Fujitsu’s extensive expertise in processor and server development, achieving world-class computational performance and data supply capabilities with a maximum operating frequency of 3.8GHz and memory transfer speed of 8800MT/s” (Fujitsu press release via PR Newswire).
And on the efficiency argument that underpins the whole sovereign-AI pitch, Fujitsu Limited said the chip “delivers twice the AI inference throughput and superior power efficiency compared to other CPUs, effectively halving the number of servers and power consumption required for equivalent processing loads” (Fujitsu press release via PR Newswire). Those four statements make up the entirety of Fujitsu’s public messaging so far; independent benchmarks from reviewers will not be possible until hardware ships in November.
Arm’s Role: Confidential Computing on MONAKA
Arm itself has published technical detail on how Fujitsu implemented confidential computing on FUJITSU-MONAKA using Arm’s Confidential Compute Architecture (CCA), in a post on Arm’s developer community blog. That collaboration signals FUJITSU-MONAKA is not a one-off Fujitsu design running on a generic Armv9 license. Fujitsu worked directly with Arm on hardware-level confidential computing features, which matters for the sovereign-AI pitch: a chip built for government and defense-adjacent workloads needs memory encryption and isolation guarantees that go beyond what a standard server CPU offers.
Confidential computing support also positions FUJITSU-MONAKA to compete for workloads where data cannot leave an encrypted enclave even from the cloud operator’s own administrators, a requirement increasingly common in financial services, healthcare, and defense procurement across both Japan and the EU.
Market Reaction and Early Industry Discussion
Community reaction moved quickly in hardware forums. A thread on TechPowerUp’s forums picked apart the chiplet architecture within hours of the announcement, with enthusiasts comparing the 2nm-on-5nm die split to similar approaches used elsewhere in the industry. That kind of fast technical scrutiny is normal for a chip announcement with real specifications attached, rather than a vague roadmap teaser.
The wider AI-infrastructure press treated the story as part of a bigger pattern: national and regional AI sovereignty projects moving from policy papers into shipping hardware. Fujitsu’s timing puts it ahead of several government-backed chip efforts still in earlier design stages, giving it a head start in a category that, until this week, existed mostly as an argument rather than a product line.
What This Means for Enterprise and Government Buyers
For data center operators evaluating AI infrastructure purchases in 2026 and 2027, FUJITSU-MONAKA adds a genuinely new option rather than a marginal update to an existing line. Buyers in Japan and the EU now have a domestically traceable alternative to US- and Taiwan-sourced AI hardware, with a concrete November 2026 order date rather than a vague future promise.
That said, buyers outside Japan and Europe should not expect access on day one. Fujitsu’s initial rollout is explicitly scoped to Japan and Europe, so enterprises in North America, Southeast Asia, or elsewhere will likely need to wait for a second wave of availability, if Fujitsu chooses to expand distribution at all. The company’s public messaging has so far stayed focused on national-security-adjacent procurement, which suggests government and defense-linked buyers in the initial markets get priority.
Enterprises weighing a MONAKA Server purchase against established options should also account for software ecosystem maturity. A brand-new Arm core, even one built on proven Armv9 SVE2 foundations, means AI frameworks, compilers, and container orchestration tooling will need validation against FUJITSU-MONAKA specifically, work that has had months, not years, to mature ahead of the November ship date.
Predictions: Where Sovereign AI Silicon Goes From Here
A few things look likely to play out over the next 12 months as FUJITSU-MONAKA moves from announcement to shipping product.
- Independent benchmarks will arrive within weeks of the November 2026 launch, and Fujitsu’s “2x AI inference throughput” claim will get tested against real workloads rather than company-supplied comparisons.
- Other governments watching Japan’s move will accelerate their own sovereign-chip announcements; expect at least one additional national or EU-backed sovereign AI CPU project to go public before mid-2027.
- Fujitsu will likely expand MONAKA Server availability beyond Japan and Europe once the initial fiscal-year-2026 production run proves stable, though a firm date for other regions has not been announced.
- Confidential computing support built with Arm’s CCA will become a bigger differentiator than raw core count, as defense and financial-sector buyers prioritize hardware-level data isolation over peak throughput.
- The memory bandwidth and CXL 3.0 support in FUJITSU-MONAKA will draw comparisons to how well it holds up against the ongoing industry-wide memory shortage shaping server pricing through the rest of 2026.
The Bigger Picture: Sovereignty as a Chip Design Constraint
FUJITSU-MONAKA is a useful marker for how far sovereign AI has moved from talking point to product roadmap. A year ago, sovereign compute was mostly discussed as a policy goal, something governments wanted but had not yet turned into shipping silicon. Fujitsu’s announcement, alongside its March 2026 manufacturing start and November 2026 commercial launch, shows a major vendor treating national manufacturing control as a core design constraint, not an afterthought bolted onto marketing copy.
Whether that translates into meaningful market share against entrenched US chip vendors will depend on execution over the next several quarters: whether Fujitsu hits its November ship date, whether independent benchmarks back up the 2x inference throughput claim, and whether the software ecosystem catches up fast enough for enterprise buyers to deploy MONAKA Servers without friction. The specs are real and dated. What happens after November 2026 is the part still being written.
Frequently Asked Questions
What is FUJITSU-MONAKA?
FUJITSU-MONAKA is a next-generation, Arm-based server CPU from Fujitsu Limited, built on the Armv9 architecture with 256-bit SVE2 vector extensions and 144 cores per socket. It uses a 3D chiplet design with a 2nm compute die and a 5nm SRAM/I-O die, and it targets AI inference and HPC workloads.
What is the Fujitsu MONAKA Server?
The Fujitsu MONAKA Server is the rack-level system built around the FUJITSU-MONAKA CPU. Fujitsu describes it as a made-in-Japan sovereign AI infrastructure product, manufactured at the company’s Kasashima plant in Ishikawa Prefecture, aimed at data center operators, enterprises, and the academic and HPC sectors.
When can I buy FUJITSU-MONAKA or the MONAKA Server?
Fujitsu says global sales of the FUJITSU-MONAKA CPU begin in November 2026. The MONAKA Server becomes broadly available and starts accepting orders the same month, initially in Japan and Europe.
What does “sovereign AI infrastructure” mean in this context?
It refers to AI computing hardware and software that is designed, manufactured, and operated within a single country or allied bloc’s own supply chain, rather than depending on foreign-made chips and cloud infrastructure. Fujitsu is marketing MONAKA specifically around this idea, tying it to national security in its own public statements.
How does FUJITSU-MONAKA compare to Intel and AMD server CPUs?
FUJITSU-MONAKA’s 144 cores sit below Intel’s Xeon 6+, which reaches 288 cores on Intel’s 18A process, and its DDR5 memory setup differs from AMD’s Threadripper Halo Station, which pairs 96 cores with 576GB of HBM3E. FUJITSU-MONAKA’s differentiator is not core count but its fully domestic Japanese design and manufacturing chain, paired with Arm confidential computing support.
Where is FUJITSU-MONAKA manufactured?
Fujitsu manufactures the associated MONAKA Server at its Kasashima plant in Ishikawa Prefecture, Japan. Manufacturing of Made in Japan sovereign AI servers at that plant began in March 2026, with production of CPU-equipped servers following within Fujitsu’s fiscal year 2026, ending March 31, 2027.
Does FUJITSU-MONAKA support confidential computing?
Yes. Fujitsu worked with Arm to implement confidential computing on FUJITSU-MONAKA using Arm’s Confidential Compute Architecture (CCA), according to a technical post on Arm’s developer blog, giving the chip hardware-level memory isolation features relevant to government and regulated-industry buyers.
Will FUJITSU-MONAKA be available outside Japan and Europe?
Fujitsu’s initial announcement scopes availability to Japan and Europe starting November 2026. The company has not published a timeline for availability in other regions.
