Fujitsu used September 14 to draw a line in the sand for the global AI chip race. The company confirmed the launch of its FUJITSU-MONAKA CPU and the accompanying Fujitsu MONAKA Server, both built entirely in Japan and aimed squarely at governments and enterprises that want AI infrastructure they fully control, according to Fujitsu’s PR Newswire announcement, with Fujitsu planning to provide the standalone MONAKA processor to cloud and data center operators and server vendors starting in the fourth quarter of fiscal year 2026, including the US and APAC regions. The bigger story isn’t the silicon alone. It’s where Fujitsu is choosing to fight, and who it’s picking as friends instead of targets along the way.
Rather than chasing Nvidia and AMD on raw AI accelerator throughput, Fujitsu is staking out a narrower lane: sovereign compute for data centers that can’t or won’t run on foreign-controlled infrastructure. That’s a different game, with different winners, and it changes how the rest of the industry should read this launch.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Fujitsu Announced on September 14
Fujitsu’s announcement centers on two products: the FUJITSU-MONAKA processor itself and the Fujitsu MONAKA Server platform built around it, which ships in both 1U and 2U rack configurations according to StorageReview’s September 2026 coverage of the launch. Per the company’s own release, the chip hits a maximum operating frequency of 3.8GHz and a memory transfer speed of 8800MT/s, numbers Fujitsu frames as delivering “world-class computational performance and data supply capabilities.” Fujitsu also claims AI-inference throughput and power efficiency at 2× competing CPUs, says equivalent workloads could require 50% fewer servers, and plans Fujitsu MONAKA Server for Japan and Europe from November 2026. Both products are pitched explicitly at sovereign AI infrastructure, meaning data centers where the operator, and often a national government, wants to keep design, manufacturing, and operational control inside its own borders.
That framing matters more than the clock speed. Plenty of chips run at 3.8GHz. Very few can be marketed with a straight face as “made in Japan” from architecture through assembly, and that’s the pitch Fujitsu is making to buyers who’ve grown wary of concentrating national AI infrastructure inside a handful of US-based suppliers.
Inside the FUJITSU-MONAKA Chip: 144 Cores, 2nm, Chiplets
Fujitsu has been previewing MONAKA’s architecture for nearly two years, and the specs that Tom’s Hardware documented from early mechanical samples largely hold up in this week’s launch. The chip packs 144 Armv9-A cores per die, built from four 36-core compute chiplets manufactured on TSMC’s 2nm-class N2 process, assembled using CoWoS system-in-package technology with separate cache and I/O dies built on a 5nm process and bonded via hybrid copper bonding, a split-node chiplet mix Fujitsu confirmed in its September 2026 launch materials.
Fujitsu also confirmed 12-channel memory support, a wide interface that’s typical of chips built to feed data-hungry AI inference workloads rather than general-purpose office servers, and October reporting reportedly added PCIe 6.0 and CXL 3.0 to the interface list alongside the 12 DDR5 memory channels. Dual-socket configurations reportedly scale to 288 cores per node. No public pricing has been disclosed.
| Spec | FUJITSU-MONAKA (2026 launch details) |
|---|---|
| Cores per die | 144 (four 36-core chiplets) |
| Architecture | Armv9-A |
| Process node | 2nm-class (TSMC N2) |
| Packaging | 3D chiplet, CoWoS SiP, hybrid copper bonding |
| Memory channels | 12-channel |
| Memory transfer speed | 8800 MT/s |
| Max operating frequency | 3.8 GHz |
| Dual-socket node core count | 288 cores |
| Release timeline | 2027 (server platform launched Sept. 14, 2026) |
| TDP / socket / pricing | Not disclosed as of Sept. 15, 2026 |
Why “Sovereign AI” Is the Whole Pitch
Every major cloud region on earth now runs AI workloads on chips designed in the US, largely fabricated in Taiwan, and controlled by a small set of American and Taiwanese firms. That’s fine for most customers. It’s a problem for governments, defense agencies, healthcare systems, and regulated industries that need a legally and politically defensible answer to “who controls this infrastructure.”
Fujitsu is answering that question directly. MONAKA is designed, and largely built, under Japanese industrial policy, with the underlying technology tied to a project subsidized by Japan’s New Energy and Industrial Technology Development Organization (NEDO), according to Fujitsu’s own press materials. That’s a fundamentally different sales pitch than “our chip is faster.” It’s “our chip is yours to control,” and it’s aimed at buyers who will pay a premium, or accept a performance tradeoff, for that guarantee.
Fujitsu’s Partner Stack: AMI, Nvidia, Scaleway, NEDO
What makes MONAKA notable isn’t that Fujitsu built it alone. It’s who signed up around it in the past year. Firmware specialist AMI announced in October 2025 that it had been selected as Fujitsu’s firmware supplier and development partner for the chip, calling it “an exciting new FUJITSU-MONAKA next-generation Arm-based processor,” per the Kyodo News release distributing AMI’s announcement. That’s a routine-sounding deal that actually matters: without a credible firmware partner, no hyperscale customer will put a new CPU architecture into production.
More striking is Fujitsu’s parallel expansion of its relationship with Nvidia, announced in October 2025, covering a joint AI agent platform for healthcare, manufacturing, and robotics, plus AI computing infrastructure that integrates MONAKA CPUs with Nvidia GPUs over NVLink Fusion. Fujitsu also confirmed in earlier 2024 disclosures that MONAKA-based platforms would support AMD GPUs as well. In other words, Fujitsu isn’t trying to replace Nvidia or AMD’s silicon. It wants to be the sovereign CPU layer sitting underneath whichever accelerator a customer already trusts.
The Scaleway Deal: Europe’s Data-Sovereignty Test Case
Fujitsu’s clearest bet outside Japan is a strategic collaboration with Scaleway, the French cloud provider that markets itself as Europe’s answer to AWS and Google Cloud on data residency grounds. The two companies announced in December 2025 that they’d build a joint proof-of-concept environment for AI inference on MONAKA-based infrastructure in the second half of 2026, with pilot environments for actual customers under consideration starting in 2027, according to Fujitsu’s German press materials.
That timeline lines up almost exactly with the European Union’s ongoing push for “digital sovereignty” in cloud and AI infrastructure, a policy priority that has already reshaped how AWS, Microsoft, and Google market their EU-region services. Scaleway is a small player next to the American hyperscalers, but it’s precisely the kind of customer that gives Fujitsu a foothold in a market that badly wants an alternative to US-controlled silicon.
Nvidia and AMD Ties: Complement, Not Compete
It’s tempting to frame MONAKA as “Fujitsu vs. Nvidia” or “Fujitsu vs. AMD,” but the partnership structure argues against that reading. Nvidia’s Grace CPU line and Fujitsu’s MONAKA occupy adjacent, not identical, ground: Grace is built to be paired tightly with Nvidia’s own GPUs inside a vertically integrated stack, while MONAKA is explicitly designed to sit underneath GPUs from either Nvidia or AMD depending on what a sovereign customer already runs.
That’s a hedge, not a challenge. Fujitsu doesn’t need to beat Nvidia’s GPU roadmap. It needs governments and regulated enterprises to see MONAKA as the safe, domestically controllable CPU choice for AI infrastructure, regardless of which accelerator sits next to it in the rack. AMD is moving in a similar rack-scale direction: on October 12, 2026 it introduced Helion, an agentic-AI rack blueprint based on an Open Compute Project design, with configurations planned to become available in November 2026.
Fujitsu MONAKA vs Nvidia Grace vs AMD EPYC: Where Each One Plays
The three companies aren’t really fighting over the same customer in 2026. Here’s how the positioning breaks down based on each company’s public strategy and Fujitsu’s own MONAKA disclosures.
| Vendor | Architecture | Primary angle | Sovereignty model |
|---|---|---|---|
| Fujitsu MONAKA | Arm (Armv9-A) | Sovereign AI CPU, pairs with any GPU | Designed and controlled in Japan; NEDO-linked development |
| Nvidia Grace | Arm | Vertically integrated CPU+GPU AI systems | US-designed, globally sold |
| AMD EPYC 9006 Series (6th Gen) | x86 | General-purpose and AI-adjacent server compute | US-designed, globally sold |
The practical takeaway: MONAKA isn’t trying to win a head-to-head benchmark fight in 2026. Fujitsu is trying to own a category, sovereign-controlled Arm silicon for AI infrastructure, before Nvidia, AMD, or another entrant defines it first.
From Fugaku to MONAKA: Fujitsu’s Supercomputing Pedigree
Fujitsu isn’t a newcomer trying to bluff its way into chip design. MONAKA builds directly on the company’s A64FX processor, the Arm-based chip that powered Japan’s Fugaku supercomputer, which topped global supercomputer performance rankings after its 2020 debut. The architecture has shifted along the way, though: Tom’s Hardware noted in August 2026 that MONAKA trades A64FX’s single 512-bit SVE vector unit for dual 256-bit SVE2 units, a design choice that favors flexibility and chiplet scaling over raw per-core vector width. That history still matters commercially: Fujitsu has already proven, at national scale, that it can design competitive Arm silicon rather than simply integrating chips designed elsewhere.
MONAKA carries that lineage into a commercial data-center product instead of a one-off supercomputer. The chiplet packaging, the wide memory interface, and the power-efficiency focus all echo lessons Fujitsu learned building Fugaku, now aimed at a market Fujitsu hopes is large enough to justify the R&D: sovereign AI infrastructure buyers who need HPC-grade engineering credibility, not just a marketing claim.
Japan’s Bigger Chip Strategy: NEDO, Rapidus and the Sovereignty Push
MONAKA doesn’t exist in isolation. It’s one piece of a broader Japanese push to rebuild domestic semiconductor capability after decades of losing ground to Taiwan, South Korea, and the US. Rapidus, Japan’s state-backed effort to build a leading-edge logic foundry, is the manufacturing side of that push. Fujitsu’s MONAKA, tied to NEDO-subsidized development, is the design and product side.
Put together, Japan is trying to build an end-to-end domestic AI compute stack: foundry capacity through Rapidus, CPU design through Fujitsu, and now customer relationships through partners like Scaleway and Nvidia. None of these pieces individually threatens Nvidia’s or TSMC’s market position in 2026. Collectively, they represent Japan’s clearest attempt in over a decade to reduce its dependence on foreign AI infrastructure.
The backdrop here is decades long. Japan dominated global semiconductor manufacturing through the 1980s before losing ground to South Korean and Taiwanese foundries and, later, to US-designed GPU and AI accelerator supply chains. Rapidus and MONAKA together represent Tokyo’s most concrete attempt yet to reverse that slide, backed by direct government funding rather than left purely to the private sector.
The Sovereign AI Chip Race Beyond Japan
Japan isn’t alone in chasing compute sovereignty. The European Union has pushed sovereign cloud requirements onto AWS, Microsoft, and Google throughout 2025 and 2026, and Fujitsu’s Scaleway partnership is a direct product of that pressure. Comparable domestic AI chip programs are reportedly underway in other regions, though specific technical or commercial details on India’s or China’s sovereign AI silicon efforts were not part of this week’s Fujitsu disclosures, and any direct comparison would be premature without verified specs from those programs.
What’s clear is the pattern: no government wants its national AI capacity sitting entirely on infrastructure it doesn’t control. Fujitsu is simply the most concrete, most publicly documented example of that anxiety turning into an actual announced product in September 2026.
What Fujitsu Hasn’t Said: Pricing, TDP, and the 2027 Gap
For a company calling this a launch, Fujitsu has left significant gaps in the public record. There’s no published thermal design power figure, no confirmed socket designation, and no pricing, list or otherwise. The company’s own MONAKA technology page describes the chip as “set for release in 2027,” a full year after this week’s announcement, according to Telecompaper’s coverage of the launch.
That gap between launch and release is worth sitting with, though it narrowed somewhat after Fujitsu disclosed that Fujitsu MONAKA Server is planned for customers in Japan and Europe beginning in November 2026, with standalone processor sales planned for the fourth quarter of fiscal 2026 and larger-scale server shipments to Japan and Europe planned to begin gradually in April 2027. Fujitsu appears to be announcing a platform commitment and a partner ecosystem well ahead of general availability, a strategy that makes sense for a sovereign-infrastructure sales cycle, since governments plan procurement years out, but it leaves outside observers without the benchmark data needed to judge whether MONAKA can actually compete on performance once it ships at scale.
Where the MONAKA Rollout Stands Now
Fujitsu MONAKA Server is reported as planned for availability to customers in Japan and Europe beginning in November 2026. Larger-scale shipments to Japan and Europe are reported as planned to begin gradually in April 2027, following initial adoption during the second half of fiscal 2026. The names currently reported are FUJITSU-MONAKA for the processor and Fujitsu MONAKA Server for the server line.
Fujitsu’s own technology page describes FUJITSU-MONAKA as a Next-Gen Arm-based processor, set for release in 2027, designed to address the challenges of next-generation data centers with its unmatched performance and power efficiency. The footnote behind its performance claims reads: “Based on our performance estimates for release in 2027. Performance may vary depending on usage conditions, configuration, and other factors.” For readers, that means the headline comparisons are estimates tied to a future release, not results measured on hardware that customers can buy today.
Fujitsu exhibited at the 2026 OCP Global Summit, and a Fujitsu press release dated February 12, 2026 is titled “Fujitsu Group starts manufacturing sovereign AI servers in Japan to enhance digital sovereignty.” Anyone budgeting for a sovereign deployment should confirm delivery windows, pricing and thermal specifications directly with Fujitsu before committing, since the public record still leaves those open.
Market and Industry Reaction
No verified stock price movement for Fujitsu (Tokyo: 6702) tied specifically to the MONAKA announcement had been reported as of publication. That’s not unusual for a chip whose server line is only planned for customers in November 2026, with larger shipments planned to begin gradually in April 2027, and it underscores that MONAKA is, for now, more of a strategic and policy story than a near-term earnings event for Fujitsu.
The clearest signal of industry confidence so far is the partner list itself. AMI’s willingness to commit engineering resources as firmware supplier, and Nvidia’s decision to expand its own collaboration with Fujitsu around MONAKA rather than treat it as a rival platform, both suggest the ecosystem sees a viable product rather than a paper launch. That’s a meaningfully different reception than most first-generation server CPU architectures get from established GPU and firmware vendors.
It also puts pressure on rival Arm server CPU efforts to clarify their own sovereignty story. Buyers evaluating AI infrastructure for the next decade now have a concrete, publicly documented example of what a “sovereign-first” CPU pitch looks like, complete with a named firmware partner, a named cloud partner, and a named GPU partner. Competitors that can’t point to an equivalent list will have a harder time making the same argument to sovereignty-focused procurement teams in Europe and Asia.
5 Predictions for MONAKA and the Sovereign AI Chip Race
None of the following are confirmed by Fujitsu or its partners. They’re informed projections based on the partnership timeline, the 2027 release target, and the broader sovereign-compute trend documented above.
- Fujitsu will likely disclose TDP, socket, and initial pricing tiers sometime in 2027 as the Scaleway proof-of-concept moves from lab testing toward pilot deployment.
- Expect additional sovereign-cloud partnerships beyond Scaleway to surface in Europe and Asia-Pacific before MONAKA’s 2027 general availability, following the same playbook of policy-driven, not benchmark-driven, customer wins.
- Nvidia and AMD will keep treating MONAKA as complementary infrastructure rather than a competitive threat through at least 2027, since neither company’s core GPU revenue depends on which CPU sits next to it in a sovereign deployment.
- Japan’s broader chip-sovereignty push, spanning Rapidus manufacturing and Fujitsu’s CPU design, will face its real test only once Rapidus’s own leading-edge foundry output and MONAKA’s commercial shipments arrive in the same window, likely 2027 into 2028.
- Pricing transparency will remain limited until MONAKA reaches general availability, since sovereign-infrastructure deals of this kind are typically negotiated government-to-vendor rather than sold off a public price list.
Quick Reference: MONAKA Launch Timeline
2024 (Nov-Dec): First MONAKA mechanical samples previewed; 144-core, 2nm, chiplet design disclosed
2025 (Oct 3): Fujitsu expands NVIDIA collaboration - AI agent platform + NVLink Fusion integration
2025 (Oct 16): AMI announced as MONAKA firmware supplier and development partner
2025 (Dec 4-8): Fujitsu and Scaleway announce sovereign AI inference collaboration
2026 (Sept 14): FUJITSU-MONAKA CPU and MONAKA Server officially launched
2026 (H2): Scaleway proof-of-concept environment planned
2027: FUJITSU-MONAKA general release; Scaleway customer pilots under considerationFrequently Asked Questions
What is the FUJITSU-MONAKA CPU?
It’s Fujitsu’s next-generation, Arm-based server processor, designed and largely manufactured in Japan, built for sovereign AI infrastructure where governments and regulated enterprises need domestically controlled compute.
When will FUJITSU-MONAKA actually be available?
Fujitsu launched the CPU and server platform on September 14, 2026, with Fujitsu MONAKA Server planned for customers in Japan and Europe from November 2026, but the company’s own technology page lists 2027 as the general release timeline, with a Scaleway proof-of-concept planned for the second half of 2026.
How many cores does FUJITSU-MONAKA have?
144 Armv9-A cores per die, built from four 36-core chiplets, with dual-socket server nodes reportedly scaling to 288 cores.
Is Fujitsu MONAKA competing directly with Nvidia and AMD?
Not directly. Fujitsu has expanded its collaboration with Nvidia around MONAKA-based infrastructure using NVLink Fusion, and has said MONAKA platforms will also support AMD GPUs. MONAKA is positioned as the sovereign CPU layer underneath GPUs from either company, not a replacement for their accelerators.
What is sovereign AI infrastructure, and why does it matter?
Sovereign AI infrastructure refers to compute that a country or organization designs, builds, and operates under its own legal and political control, rather than depending on foreign-owned hardware and cloud platforms. It matters most for governments, defense, healthcare, and other regulated sectors handling sensitive data.
What process node is FUJITSU-MONAKA built on?
A 2nm-class process, manufactured on TSMC’s N2 node, assembled using CoWoS system-in-package chiplet technology with hybrid copper bonding.
Who are Fujitsu’s key MONAKA partners?
Confirmed partners include AMI as firmware supplier, Nvidia for AI agent platform and GPU integration work, Scaleway for European sovereign-cloud deployment, and Japan’s NEDO, which subsidized underlying development work.
Does Fujitsu MONAKA connect to Japan’s Rapidus chip foundry project?
They are separate but complementary efforts within Japan’s broader chip-sovereignty strategy: Rapidus is building domestic leading-edge foundry capacity, while Fujitsu’s MONAKA is a CPU design and product effort. No manufacturing partnership between the two has been confirmed for MONAKA’s current 2nm production, which uses TSMC’s N2 process.
