Fujitsu announced on September 14, 2026, from Kawasaki, Japan, that it will begin global sales of FUJITSU-MONAKA, a next-generation CPU designed and built entirely in Japan, starting in November 2026. Alongside the chip, the company is launching the Fujitsu MONAKA Server, a made-in-Japan system the company is positioning as sovereign AI infrastructure for data center operators, enterprises, academic and HPC researchers, and defense agencies in Japan and Europe. The announcement, carried by PR Newswire and confirmed by outlets including Unite.AI and CybersecAsia, marks Fujitsu’s clearest attempt yet to sell Japan and Europe on domestically built AI hardware instead of relying on Nvidia-centric supply chains.
The pitch is specific and numerical: Fujitsu says FUJITSU-MONAKA delivers twice the AI inference throughput of comparable CPUs while consuming half the power for the same workload, which the company frames as letting customers run equivalent jobs on 50% fewer physical servers. That claim, paired with a 144-core Arm chip built on a 2nm compute die, puts Fujitsu in direct competition with Nvidia’s Grace CPU, AWS’s Graviton line, and Ampere Computing’s AmpereOne, but with a distinctly different sales pitch built around national security and supply-chain traceability rather than raw benchmark supremacy.
What Is FUJITSU-MONAKA? Inside the 2nm, 144-Core Arm Chip
FUJITSU-MONAKA is a 144-core Armv9-based server processor built from four 36-core compute chiplets. According to technical breakdowns published by Tom’s Hardware and ServeTheHome, the compute chiplets are fabricated on TSMC’s N2 (2nm) process and stacked face-to-face onto SRAM cache tiles built on TSMC’s N5 (5nm) process, joined with hybrid copper bonding rather than a conventional interposer. Fujitsu first demonstrated this 3D-stacked packaging approach at the Hot Chips 2026 conference, months before confirming a commercial launch date, which is why the deepest architectural details trace back to that earlier technical disclosure rather than the September 14 sales announcement.
The commercially available FUJITSU-MONAKA CPU tops out at a 3.8GHz clock speed with a memory transfer rate of 8,800MT/s, and it includes dedicated matrix-multiplication instructions alongside Arm’s SVE2 vector extensions, both aimed squarely at AI inference math rather than general-purpose compute. Notably, Fujitsu chose stacked cache dies and fast DDR5 memory over HBM, a departure from the GPU-style memory Nvidia and its rivals typically pair with AI accelerators. That decision keeps bill-of-materials costs down and simplifies manufacturing, though it also means MONAKA is not attempting to match GPU-class memory bandwidth. Instead, Fujitsu is betting that CPU-native inference, run at scale across many cores, can undercut GPU-based inference on cost per watt for specific workloads.
Fujitsu MONAKA Server: 1U, 2U, and a Multi-Node HPC System
The Fujitsu MONAKA Server ships in three configurations built around the new CPU. The 1U model is designed for AI inference and AI agent workloads in standard data center racks, a 2U model targets denser deployments, and a separate 2U multi-node system is planned specifically for academic and HPC environments that need to cluster many nodes together. All three are, per Fujitsu’s own description, “designed, developed, and manufactured in Japan, with component and manufacturing traceability,” a phrase the company has repeated across its sovereign-AI messaging this year.
One of the more unusual engineering details is cooling tolerance. The 1U server is rated to operate in ambient temperatures up to 40 degrees Celsius using air cooling alone, and up to 45 degrees Celsius when water-cooled, without requiring the specialized liquid-cooling infrastructure that GPU-dense AI racks typically need. Fujitsu says this cuts server cooling power consumption by as much as 80%, a meaningful figure for data center operators wrestling with power and thermal limits as AI infrastructure has driven a broader memory supply crunch across the industry this year. The server also adopts CDI/CXL (Composable Disaggregated Infrastructure and Compute Express Link) technology, letting operators pool memory and accelerators across physical server boundaries instead of stranding capacity inside individual boxes.
FUJITSU-MONAKA CPU Specifications at a Glance
| Specification | Detail |
|---|---|
| Announcement date | September 14, 2026 (Kawasaki, Japan) |
| Architecture | Armv9-based cores, four 36-core compute chiplets |
| Process node | 2nm compute die (TSMC N2) + 5nm cache die (TSMC N5), 3D-stacked via hybrid copper bonding |
| Core count | 144 cores per chip |
| Max clock speed | Up to 3.8GHz |
| Memory transfer rate | 8,800MT/s (DDR5-based, no HBM) |
| AI performance claim | 2x inference throughput vs. comparable CPUs at half the power, per Fujitsu |
| Cooling | Air-cooled to 40°C, water-cooled to 45°C; up to 80% lower cooling power |
| Server form factors | 1U, 2U, and 2U multi-node (HPC/academic) |
| Global CPU sales start | November 2026 |
Why “Sovereign AI” Is the Whole Pitch
Fujitsu is not primarily selling FUJITSU-MONAKA on raw performance. The recurring phrase across the company’s materials is sovereign AI infrastructure, a category built around the idea that governments, militaries, and regulated industries want AI hardware whose design, fabrication, and supply chain they can trace and control. In its own announcement, Fujitsu stated: “Fujitsu MONAKA Server, powered by the FUJITSU-MONAKA CPU, enhances sovereign capabilities through domestic manufacturing, contributing to traceability and supply chain transparency,” according to the company’s official press release.
That framing lands at a moment when export controls, chip shortages, and geopolitical tension have made hardware provenance a boardroom and cabinet-level issue rather than a purely technical one. Reporting from Startup Fortune frames MONAKA as Fujitsu’s bet that banks, telecoms, and militaries want AI inference hardware that does not route through Nvidia at all, an alternative built on domestic manufacturing rather than a faster GPU. It’s a narrower market than general-purpose cloud AI, but one where price sensitivity matters less than auditability.
Target Markets: Data Centers, Defense, Academia, and HPC
Fujitsu says 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,” according to the company’s announcement. The standalone FUJITSU-MONAKA CPU is being sold more broadly, to cloud providers, data center operators, and third-party server vendors, positioning Fujitsu as both a systems builder and a component supplier in the same launch.
The explicit mention of the defense sector is notable. Most CPU launches lead with hyperscalers and cloud tenants. Fujitsu instead foregrounds government and military buyers in Japan and Europe, sectors where procurement rules increasingly require domestic or allied-nation hardware for sensitive workloads. That same dynamic has been reshaping sovereign cloud strategy among the big three hyperscalers, who have spent the past year building region-locked, government-only infrastructure to compete for the same procurement budgets Fujitsu is now targeting with hardware instead of cloud services.
Fujitsu’s Vertical AI Stack: Kozuchi, Takane, and MONAKA
Fujitsu is not selling MONAKA as a standalone chip so much as the base layer of a vertically integrated sovereign AI stack. Coverage from Unite.AI describes the company combining its Kozuchi AI development platform, its Takane enterprise generative AI models, and industry-specific AI models on top of MONAKA hardware, giving customers a single vendor relationship spanning silicon, servers, and software. That mirrors a broader industry pattern in which chipmakers increasingly bundle software and services around hardware to lock in enterprise customers, rather than compete purely on component specs.
For a company best known outside Japan for IT services and mainframes rather than cutting-edge silicon, owning the full stack from chip to application is a meaningful strategic shift. It also reduces Fujitsu’s exposure to the same AI chip supply constraints that have squeezed rivals this year, including the memory shortages tracked in reporting on record demand for advanced packaging and the surge in TSMC’s AI-driven revenue growth, since Fujitsu designs and integrates the full server rather than depending on third-party AI accelerator allocations.
MONAKA vs. Nvidia Grace, AWS Graviton, and Ampere
FUJITSU-MONAKA enters a crowded field of Arm-based server processors, but its positioning differs from each of its closest rivals. Nvidia’s Grace CPU pairs 72 Arm Neoverse V2 cores tightly with Nvidia’s own GPUs and HBM memory, built explicitly to feed data to accelerators rather than to run inference on its own. AWS’s Graviton line has scaled aggressively, jumping from 96 cores on Graviton4 to 192 cores on the newly unveiled Graviton5, according to Tom’s Hardware‘s coverage, though Graviton remains exclusive to AWS’s own cloud. Ampere Computing’s AmpereOne also reaches 192 cores in a single socket, aimed at cloud-native workloads for any operator willing to buy Ampere silicon directly.
MONAKA’s 144 cores sit between Grace and the 192-core chips on raw count, but Fujitsu isn’t trying to win a core-count contest. Its differentiation is the sovereignty and traceability pitch, the unusually high air-cooled thermal tolerance, and a memory architecture built around stacked cache and DDR5 instead of HBM, which keeps the platform cheaper to deploy at scale even if it can’t match GPU-class bandwidth. Whether that trade-off wins share from Nvidia depends heavily on how much buyers in defense and regulated industries are willing to pay for supply-chain certainty over peak performance. It’s a different playbook than the general-purpose custom silicon race playing out elsewhere in AI chips, where most competitors are still chasing raw throughput rather than provenance.
Sovereign and Hyperscale Arm Server CPUs Compared
| Chip | Maker | Cores | Primary positioning | Availability |
|---|---|---|---|---|
| FUJITSU-MONAKA | Fujitsu | 144 | Sovereign AI inference, air-cooled, made in Japan | Global sales from November 2026 |
| Grace CPU | Nvidia | 72 | Paired with Nvidia GPUs, HBM-coupled compute | Shipping |
| Graviton5 | AWS | 192 | Hyperscale cloud, AWS-exclusive | Unveiled late 2025, rolling out |
| AmpereOne | Ampere Computing | 192 | Cloud-native, single-socket, vendor-neutral | Shipping |
Historical Context: Japan’s Long Road Back to Chip Independence
Japan dominated global semiconductor manufacturing in the 1980s before losing ground to Taiwan, South Korea, and the United States over the following three decades. FUJITSU-MONAKA arrives as part of a broader, multi-year effort by Japanese industry and government to rebuild domestic chip capability, a push that has included new fab investment on Japanese soil and renewed state interest in supply-chain resilience following pandemic-era shortages and rising geopolitical tension over Taiwan-based manufacturing. That tension has also touched Japan’s neighbors, with Samsung and SK Hynix recently losing streamlined export authorization for chip tools tied to China, underscoring how quickly semiconductor supply chains have become geopolitical flashpoints across the region. Fujitsu itself has spent years building toward MONAKA, first detailing the chip’s 3D-stacked chiplet architecture at industry conferences well before committing to a commercial launch date.
That slow, staged rollout, technical preview first, commercial availability roughly a year later, contrasts with the compressed release cycles common among US chipmakers, and reflects the higher validation bar that government and defense buyers tend to require before adopting new silicon for sensitive workloads. It also means Fujitsu has had more time to line up initial customers in Japan and Europe ahead of the November 2026 sales date than a typical consumer or cloud-first chip launch would allow.
Market Impact: What This Means for the AI Infrastructure Race
The immediate market impact is narrower than a general-purpose Nvidia competitor would be. Fujitsu is not chasing hyperscaler training clusters, where GPU-dense systems remain entrenched, but the AI inference and government-procurement segment, where sovereignty requirements can matter more than benchmark leadership. If even a modest share of Japanese and European government, defense, and financial-services inference workloads shift to CPU-native, domestically traceable hardware, it would represent a meaningful crack in Nvidia’s near-total grip on AI infrastructure spending, a dominance underscored by Nvidia’s roughly 90% share of GPU shipments reported earlier this year.
It also puts pressure on rival hyperscalers and chipmakers to sharpen their own sovereignty stories. AWS, Microsoft, and Google have all built region-locked, government-only cloud offerings over the past year, but none of them yet ships a chip explicitly marketed with domestic manufacturing traceability the way Fujitsu is doing with FUJITSU-MONAKA. That gap could push competitors toward similar hardware-level sovereignty claims, or toward deeper partnerships with domestic manufacturers, as governments in Europe in particular continue tightening digital sovereignty requirements for critical infrastructure procurement.
What Fujitsu Is Saying
Fujitsu’s own statements, issued through its official September 14 press release, lay out the company’s positioning in its own words. On the launch itself, Fujitsu said: “Fujitsu today announced that it will begin global sales of its next-generation CPU, FUJITSU-MONAKA, designed and developed in Japan, starting November 2026.” On the server product, the company 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.”
On availability and target customers, Fujitsu stated: “Fujitsu 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.” And on the strategic rationale behind the entire launch, the company said: “Fujitsu is committed to supporting customers’ autonomy and accountability by providing a sustainable sovereign AI infrastructure with high operability.” Together, those four statements read less like a typical spec-sheet launch and more like a pitch aimed directly at procurement officers and national-security buyers.
Availability, Pricing, and the Rollout Timeline
Fujitsu has been precise about timing and vague about cost. Global CPU sales begin in November 2026, and the MONAKA Server family launches in Japan and Europe on the same timeline, with standalone processors also planned for availability to cloud and data center operators in additional markets during that quarter. No public pricing has been disclosed for either the CPU or the server configurations, which is typical for enterprise and government-focused hardware sold through direct sales relationships rather than retail channels. Prospective buyers will need to work through Fujitsu’s enterprise sales teams to get quoted pricing, and given the defense and critical-infrastructure focus, contract terms are likely to vary significantly by customer and region.
Predictions: Where MONAKA Goes From Here
A handful of things seem likely to follow from this launch over the next 12 to 24 months. First, expect Fujitsu to announce specific pilot customers in Japanese and European government or defense agencies within the next two to three quarters, since sovereign-AI marketing tends to need named early adopters to gain credibility. Second, expect competitors, particularly in Europe, to respond with their own domestic-manufacturing partnerships or chip investments rather than simply matching MONAKA on specs, since the entire premise of Fujitsu’s pitch is supply-chain origin rather than performance.
Third, Fujitsu is already looking past this generation. Coverage from WCCFTech has referenced a next chip, Monaka-X, reportedly eyed for around 2029 on a more advanced 1.4nm process with NVLink Fusion support, which would suggest Fujitsu wants closer interoperability with Nvidia’s ecosystem even while marketing MONAKA as an alternative to it. Fourth, expect scrutiny of the “made in Japan” claim to intensify, since neither Fujitsu nor the outlets covering the launch have named the specific fabs producing the 2nm compute dies, and sovereignty claims tend to draw closer inspection once large government contracts are on the line. Fifth, expect the air-cooling and cooling-power claims to become a bigger part of Fujitsu’s pitch than the AI performance numbers themselves, since data center operators are increasingly power- and thermal-constrained rather than compute-constrained.
Frequently Asked Questions
What is FUJITSU-MONAKA?
FUJITSU-MONAKA is Fujitsu’s next-generation server CPU, a 144-core Armv9-based chip built on a 2nm compute die stacked with 5nm cache dies, announced September 14, 2026, and designed and manufactured in Japan for AI inference workloads.
When does FUJITSU-MONAKA go on sale?
Fujitsu says global sales of the standalone FUJITSU-MONAKA CPU begin in November 2026, with the Fujitsu MONAKA Server launching in Japan and Europe on the same timeline.
What is the Fujitsu MONAKA Server?
It’s a made-in-Japan server family equipped with the FUJITSU-MONAKA CPU, available in 1U, 2U, and 2U multi-node HPC configurations, marketed as sovereign AI infrastructure for data centers, enterprises, academia, and defense buyers.
How much does FUJITSU-MONAKA cost?
Fujitsu has not published pricing for the CPU or the MONAKA Server. Enterprise and government buyers will need to work with Fujitsu’s sales teams for quotes, which is standard for infrastructure hardware sold outside retail channels.
Does FUJITSU-MONAKA use HBM memory like Nvidia’s AI chips?
No. Fujitsu chose stacked cache dies combined with fast DDR5 memory instead of HBM, a design decision that keeps costs and manufacturing complexity lower, though it trades away some of the raw memory bandwidth that HBM-equipped AI accelerators offer.
How does FUJITSU-MONAKA compare to Nvidia Grace and AWS Graviton?
FUJITSU-MONAKA packs 144 Arm cores, more than Nvidia’s 72-core Grace CPU but fewer than AWS’s 192-core Graviton5 or Ampere’s 192-core AmpereOne. Fujitsu’s differentiator isn’t core count, it’s the sovereign, made-in-Japan manufacturing story and unusually high air-cooled thermal tolerance.
What does “sovereign AI infrastructure” actually mean?
It refers to AI hardware and systems whose design, manufacturing, and supply chain a government or regulated organization can trace and control domestically, reducing dependence on foreign suppliers for sensitive workloads like defense and critical infrastructure.
Is FUJITSU-MONAKA only for defense and government buyers?
No. Fujitsu is selling the standalone CPU to cloud providers, data center operators, and server vendors broadly, while marketing the defense, academic, HPC, and enterprise sectors as priority customers for the sovereign-branded MONAKA Server.
