Yes the lack of transparency on ARM architecture and the obfuscation of the RAM capacity with just the slot information is annoying. Not sure if it is deliberate to sell this at a premium to "mainframe" customers or whether its just their marketing folk not being on the ball. Fascinated by the fact Japan has a 2nm process node but I suspect that too is marketing. Perhaps the 2nm portion of the chip is done at TSMC and the rest of the package done in Japan or something like that. It's not like there's a plethora of 2nm Fabs around the planet to choose from.
It looks to me like it is an AI optimized HPC CPU and while HPC CPUs are actually quite fast, they are not particularly well suited for transformer based neural networks.
It's weird that they don't give out any of the interesting numbers like number of memory channels, how much SRAM they have (CPUs tend to have more of it) or what their expected performance is going to be.
No it is not at all Neoverse V2, which is obsolete and has much lower performance.
It is a custom Armv9.3-A design (same ISA like the Arm C1 CPUs from the flagship smartphones of 2026), but it has double-width execution units for SVE2 (i.e. 256-bit width, vs. 128-bit for the other Arm CPUs) and it has some ISA extensions for AI/ML, e.g. instructions for inference with FP8 (BF16 is already supported by the standard Arm ISA).
Despite the fact that Armv9.3-A may include SME (scalable matrix extension, like in the Apple CPUs and in the Arm C1 CPUs), Fujitsu did not mention SME, so I assume that they did not implement it and they rely on their enhanced SVE2 (which is not surprising, while the origin of Arm SME is at Apple, the origin of Arm SVE is at Fujitsu).
They actually do not in that article. I had to look elsewhere to find that out. Weird, one would expect that a big announcement like this would probably need to be vetted against licensing agreements with Arm that would probably want that fact to be prominently highlighted in press releases.
Yeah, searched the page for common architectures (x86,arm and Sparc since it was Fujitsu) since it'd seem silly to use a totally novel architecture but didn't find anything until noticing SVE2.
Thanks kinda how Japan works. Inventing something new is hard, but taking something that's a cool idea and making it actually usable is well within their wheelhouse.
I'm sure there's plenty of other stuff invented in Japan, but that's really beside the point.
The industry is now mature enough that nobody really wants the headache of a new architecture at this point in time, Risc-V got a pass for being a grassroots movement growing out off FPGA's since the IP situation was more or less clear or even predatory with all existing architectures, Power, Sparc,SuperH,68k,etc are buried for good reasons.
I had this idea that Fujitsu was one of those Japanese conglomerates that had operations in all kinds of markets, from electronics to heavy earth moving equipment to air conditioners to locomotives. But that doesn't appear to be the case so I'm wondering which company I'm confusing them with.
Sun invented SPARC, Fujitsu just happened to be the last producer since they built mainframes on it. Reading it seems like they plan to stop producing Sparc processors and machines in a near timeframe so that makes this announcement less surprising.
For many years the Fujitsu SPARC implementations were much faster than the Sun SPARC implementations.
During the decade 1995-2005, until the 64-bit AMD Opteron servers with Linux offered an alternative that crushed any kind of SPARC CPU with Solaris by a much higher performance combined with a much lower price, the SPARC Sun or Fujitsu servers dominated the market for servers used to host the CAD/EDA design tools used in electronics engineering, for the design of integrated circuits or electronic equipment. This means that all the EDA software from vendors like Mentor, Cadence, Synopsis etc. was available only for Solaris. That kind of software could not be run on Windows, due to the 32-bit memory limit. During that time, whoever got Fujitsu servers instead of the sluggish Sun servers, was very lucky.
Not mentioning the architecture is rather strange. Being Fujitsu I'd expect SPARC, but apparently not. Maybe we're reaching a point where some circles either don't care or just defaults to ARM?
Obviously they can just ignore the license in the future and continue development out of a local branch, but it's also a bit disingenuous to speak of "sovereign infrastructure" and then use licensed processor design.
Fujitsu has a long history of designing their own micro but using a standard ISA for HPC. Used to be SPARC, now it is ARM. IIRC Fujitsu were the main architects behind the SVE ARM extension.
Fun fact, Fujitsu used to run their own fabs in Japan until they were sold to UMC and now they're another UMC fab although they are stuck at like the 20nm node. So this is likely made at JASM.
The Phase 2 factory was under construction as of January 2025, next to the Phase 1 facility, and is expected to be completed by 2027.[3] Initially planned for manufacturing semiconductors using a 6 nm process, 3 nm was incorporated into production plans for 2028 in 2026
Doesn't really make sense to me. The whole "sovereign" thing is to basically say you won't be screwed if you go to war with a country tomorrow because you are dependent on their supply chain. Nothing about this changes that IMO. They're still getting the chips fabbed somewhere else and run the same risks.
True, but besides wars there are other dangers, e.g. USA does not declare war when they apply "sanctions" randomly, under the legal theory that they can dictate how anything that includes some piece of US origin may be used by its owner.
Since that theory began to be used, anyone outside US has become anxious to become "sovereign", i.e. to stop using anything about which a foreign state can claim to have rights.
All the component chips are made at TSMC, in various processes.
"Developed in Japan" means the same thing like "developed in USA" means for NVIDIA or AMD or Intel. You can design chips anywhere on the planet, as long as TSMC is willing to provide you the required documentation and EDA libraries (which it does only if you already are or they believe that you will be a big customer).
Indeed, it is extremely clear: by not saying "manufactured in Japan", the words "developed in Japan" tell you that they drew up the designs for the CPU in Japan, and then sent those designs off to a fab outside Japan for actual manufacturing. Then they bring those CPUs back to Japan, so that the server can be manufactured in Japan even if the server components are not.
That manufacturing step may change in the future, but right now that CPU is not "Made in Japan", it's only "Designed in Japan".
They did a presentation at Hot Chips, with all these details.
The core die is made with TSMC-N2P, while the SRAM die and the peripheral die are made with TSMC-N5.
Their CPU is designed in Japan, like the AMD CPUs or the Intel CPUs are designed in USA (the latest Panther Lake Intel CPUs have returned for fabrication to USA in 2026, but they also cost double than the corresponding Arrow Lake models made at TSMC in 2025, and their GPUs are still made at TSMC).
The compute part must have been made in Taiwan because it uses TSMC N2. TSMC N2 is only available in Taiwan right now due to a law restricting the latest node to Taiwan for 1-2 years.
At the very top of the article:
"Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan"
"Japan-developed CPU..." plus the server itself (parts, power supplies, case, etc.) integrated, developed and manufactured in Japan. It sounds like they're weasel-wording it a bit, but I don't think the CPU itself is manufactured in Japan, only "developed."
According to a Feb press release, they're fabbed by TSMC:
For FUJITSU-MONAKA, the 2nm
semiconductors will tape out next year. Due to this, Rapidus would be unable to meet the
deadline, so TSMC is handling the manufacturing.
Their CPUs are faster at inference than many GPUs. These are server CPUs with 144 cores and wide memory interface.
Fujitsu has added some ISA extensions for AI inference to the standard Arm ISA, e.g. instructions for inference with FP8.
Even their previous CPU generation was for some time the champion in energy efficiency, with better performance per watt than the NVIDIA GPUs, until a newer generation of NVIDIA GPUs has leapfrogged them.
There are good chances that this new Fujitsu CPU might be again for some time the CPU with the best energy efficiency, but it remains to be seen how it compares with the recent GPUs.
The two made-in-Japan, premium grade, Fujitsu laptops I had to work with were by far the worse pieces of electronic I have ever used. Both exhibited identical defects: fans running continuously at maximum speed, and the batteries would completely die within hours of the devices being powered off. Windows or Linux. And don't talk about that tiny unresponsive trackpad to me. Never again.
The Fujitsu Lifebooks were legendary for their robustness back in the day, I'm genuinely sad you had that experience.
The only fujitsu laptops I used was back in 2011- it was a budget version for on-call and it was fine..
Sounds like the fans running 100% probably contributed significantly to the issue here. If I had to guess it was likely that the C-States were disabled in BIOS somehow so the CPU clock was running at full tilt the whole time, maybe combined with a bad thermal paste job.
I'm not here to defend Fujitsu but I've had really bad experiences with basically every major laptop brand (Dell: majorly bad coil whine and especially faulty soldered RAM, HP: keys vanishing from the keyboard and very weak hinges that break all the time, Apple: The GPU unsoldering itself and the butterfly keyboard shenanigans).
I don't really know any brand with a flawless track record sadly.
Before anyone brings up Thinkpads, they're trading on a reputation that hasn't been true for over a decade. If you have fond memories of a thinkpad it's most likely that you had it from before 2016, or it's "fine", but certainly not great.
I do recall that Fujitsu had a very... unique approach to exposing the Setup option in their firmware. It was just a UEFI boot entry. If that entry was removed, you couldn't get back into the firmware setup; you had to find the name of the setup EFI file and then run that from the EFI console.
Reaching further back, the Sylistic line was mostly quite good/durable, and I still use my Stylstic ST-4110 when I need to use an old scanner or control a CNC machine on my back deck when cutting tropical hardwoods.
Really miss the transflective display and wish that there were newer devices with such technology
What year? The ones I had were the best. Very expensive, but service at the office & at home, very sturdy, very long battery life. That was somewhere early 2000s.
I just have Fujitsu Lifebooks at Home since 2007. I think they are good, came with a bloat free Windows. I bought even one for my Mom over 10 years ago. I just changed once the HDD with a SSD. Now all my Lifebook run with Linux just flawless.
Their software is so poor that it caused probably the biggest miscarriage of justice in history. Well that and Fujitsu staff colluding with post office executives to put innocent people in prison rather than admit there was a problem
The 'Fujitsu' that made the shitty software was a British consulting company that was acquired by Fujitsu [1] to farm public contracts. This is made by the Japanese parent company.
You're right, but the scandal is so upsetting that I personally cannot ever look at Fujitsu the same way even. Also, I think you're neglecting that a large part of why it became so bad was that Fujitsu claimed there was no issue with their software. Even if it were built by a company they acquired, the response to the developing situation was something that the parent company should have stepped in to oversee.
I think what's most notable about this for me is the reduced environmental impact. The specifications list liquid-cooling as optional and it highlights the CPU as the main processor instead of a power hungry GPU. My takeaway is that the next generation of AI HPC's will focus more heavily on sustainable operations.
One problem we’re going to have with AI hardware is coming up with a standard set of specifications that are comparable. I don’t really care about the CPU GHz and the memory bandwidth, at least not directly. What I really want to know is how many tokens per second this will deliver, but that also depends on the model. We need a standard metric for that. Perhaps we agree on a specific open weight model (e.g. GLM 5.3 Flash or Qwen vWhatever) and then measure TPS on the hardware of interest.
I always heard of GB/s as most important number ... AI told me their 8800 MT/s on 8 Byte, 12 DDR5-Channels means 845 GB/s. A Nvidia RTX 4090 has 1008 GB/s. Nvidia B200 has 8000 GB/s.
Is this the right way of looking at it?
You have to load the model weights into VRAM over PCI-E (from RAM). So the (PCI-E) bandwidth strongly affects time to first token.
You have to run inference on the GPU by reading and writing to VRAM. So TFLOPS of the compute matters, and bandwidth to the VRAM (Always integrated with the GPU, rarely a bottleneck), and this strongly affects tokens/s
If you're doing training workloads or offloading to system RAM, it gets more complicated. (And mostly bound up trying to feed compute on time)
So the (PCI-E) bandwidth strongly affects time to first token
On dedicated inference hardware I'd expect model weights to never leave the RAM, and you'd probably load them on startup before even starting to serve requests
I fail to see how they will take a significant market share or even break even on this venture.
It's an overcrowded market. Far better would be to focus on semiconductor supply chain, which Japan already supplies some elements, to sell to fabs.
Because the bottleneck is the fabs. If a new 2 nm fab came online today, it would immediately sell all its capacity to 2030 no problem, without Fujitsu trying this gambit.
It's not 'sovereign', the architecture is not designed in Japan, the silicon is not fabbed in Japan. This whole thing is sideways.
Every developed country is going to push for their own homemade chips. Japan used to make Sparc CPU's back in the day. Probably many others I'm not aware of.
Every developed country except Australia, perhaps. This government is atrocious. The record for highest number of bureaucrats per capita in the world is owned by Australia. All they know how to do is tax people. All the wealthy people have left, or are leaving.
I've been calling for high-multicore CPUs (at least 100 cores) with local memories for a quarter century now. No winners so far.
A Pentium 4 hit 3.8 GHz longer than that ago in 2004, so that's not special. And the estimated price will be $7,000-10,000, which isn't special either since that's about 10x more than it should be.
Hot take: GPUs disrupted the CPU industry to such a degree that CPUs never recovered, and like the k-shaped economy, the current status quo only serves a small fraction of customers. We can and should do better, but sadly we won't. Still, it's good that Fujitsu did this, for the competition if nothing else.
Unless they are used for multi-processing in classical UNIX fashion, or proper microkernels, most applications will hardly take advantage of them.
Managed languages runtimes are probably the ones that would be better equipped to take advantage of them, for distributed JIT, GC and asynchronous code.
The Connection Machine style with StarLisp.
Very few devs can write optimal multi-threaded code that explores the single digit count of cores on their laptops or phones already.
Back in 2008 Fujitsu has one of the best performing 10Gbps Switches. We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate. Fujitsu was way ahead of the pack. I always wondered what held them back from building a meaningful networking business in the US.
I always wondered what held them back from building a meaningful networking business in the US.
It was just timing. Their networking gear was built on some very impressive ASICs which became commodity chips a few years later before they could grow. Arista, Juniper, et al ate their lunch using other vendors’ IP. IIRC it was Broadcom and Fulcrum that released the chips that killed them.
We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate
I'm sure there were good reaasons for this, but man this just sounds bad. Like that's the spec I think NSA must give to their contractors outfitting 611 Folsom Street's Room 641A
There are companies that mirror their traffic to storage for a few days to be able to look back and troubleshoot issues. ALL their traffic, proactively.
For US federal government, "Office of Management and Budget" says to store 72 hours of packet capture, since 2021 (partially in response to SolarWinds attack). (M-21-31)
It does sound bad the way OP described it, but packet or HTTP mirroring is a standard feature of A/B testing and blue-green deployments of a very critical code. Mirror traffic between version 1 and 2, then compare response body, response time and return response A to the end users.
Port mirroring is the network engineer's equivalent of using breakpoints in a debugger for a programmer. It allows you to inspect what happens on the wire to get a clue for what's wrong.
Probably not, Japanese firms excel at hardware and generally build extremely poor software. This a hardware product, so consequently it is probably fairly good.
Honestly, no idea. In my career I have worked with a lot of Japanese firms and have always found their engineers to be excellent. I used to think it was a lack of consideration for UX, but that obviously can't be true when you see excellent examples of UX in hardware (e.g. Switch controls), even though its exceedingly rare on the software side.
If anyone on HN knows (there are several other commenters more deeply steeped in Japanese engineering culture), I would love to know why as well.
Japanese culture and work culture is very hierarchical/top down. They excel at waterfall but it's basically impossible to do agile in a Japanese office environment, and waterfall is suitable for hardware but terrible for building software.
I think Ruby would be a counterpoint - although it's the only one I know of.
I worked for a Japanese company although in a local office with local management. Through all those layers of protection I got the impression that they suffer from what all corporates suffer from - no software engineers that have made it to the top. Software companies run by people who have no clue.
In the US, for example, I think some people who have a clue or a partial clue have made it up the ladder in some places and therefore although everyone there has their horror stories of idiotic behavior it is probably somewhat better.
This is just my hypothesis, but I believe the language barrier is an issue. English is effectively a necessity for programming because real programming languages with ecoystems, error diagnostics, documentation, and useful articles/blogs/books are written in English. Japanese software developers know enough to get by, but they're still by-and-large insulated from the Anglosphere learning culture, where ideas and best practices rapidly propagate in the open. I'm not involved in hardware, but I'm guessing this is less of a barrier for hardware because hardware/documentation tends to be proprietary rather than open anyways, so whereas the Anglosphere gains a cumulative learning advantage from an open source + technical blogging culture in their fluent language, this does not manifest so much in hardware.
Conglomerates in general, and especially Japanese ones, are heterogeneous enough that those are basically two different companies. The British "Fujitsu" was an acquisition that I'm certain they now deeply, deeply regret.
I'm kind of a Fujitsu fanboy. Since i came across the fact that a pretty modern xeon ecc ram nvme Workstation boards can be optimized to only draw ~10 Watt idle is just impressive, that is less than a gaming router. Even older ones like D3417-B12 were crazy effizient.
I also liked their Primergy Servers, but never got one for a reasonable price to test it out.
Another thing was the Futro series (e.g. S930) that could be used as opnsense firewall or low budget proxmox host.
Unfortunately it was always hard to obtain the high quality stuff as a consumer and years ago they sold their mainboard section to kontron. So no new Fujitsu Mainboards for now... What a pity.
Press releases are sadly like that. Fujitsu's main page has it in the first line:
FUJITSU-MONAKA is 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.
For a single CPU:
844 GB/s memory bandwidth (12 channels of DDR5 RDIMM, 8800MT/s). 4.3-6 TFLOPS.
Basically comparable to a modern (although not top-end) GPU, so good for HPC & AI workloads. (though only 2 CPUs per node, with GPUs 4-8 is more common)
For anyone not familiar with Fujitsu's CPUs, worth noting their overall long history with HPC, and more recently, ARM. FugakuNEXT is "NEXT" because the current Fugaku (#1 on TOP500 for a time, and still in the top 10) also used an ARM CPU from Fujitsu.
Could someone knowledgeable please explain why an AI-focused CPU is superior to a GPU-based solution with an ARM CPU driving the high-level operations, please, especially when the RAM isn't on-chip?
I know that for GPU's the model weights have to be transferred over the bus initially, but that only has to occur once for inference use cases, so is the Fujitsu system more about training scenarios? Or is the focus more about efficiency, as these are ARM-based cores with AI additions?
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, as well
as to the defense sector,
contributing to national
security.
So, not in America. A sign of the geopolitical times? If so, a very unusual one to come out of Japan.
It's Arm based, although they barely mention that.
Thanks I was, of course, looking and very curious what their tool chain story will be.
Yes the lack of transparency on ARM architecture and the obfuscation of the RAM capacity with just the slot information is annoying. Not sure if it is deliberate to sell this at a premium to "mainframe" customers or whether its just their marketing folk not being on the ball. Fascinated by the fact Japan has a 2nm process node but I suspect that too is marketing. Perhaps the 2nm portion of the chip is done at TSMC and the rest of the package done in Japan or something like that. It's not like there's a plethora of 2nm Fabs around the planet to choose from.
It looks to me like it is an AI optimized HPC CPU and while HPC CPUs are actually quite fast, they are not particularly well suited for transformer based neural networks.
It's weird that they don't give out any of the interesting numbers like number of memory channels, how much SRAM they have (CPUs tend to have more of it) or what their expected performance is going to be.
They did a detailed presentation at Hot Chips.
See e.g.:
https://chipsandcheese.com/p/hot-chips-2026-fujitsus-monaka-...
They have provided far more details than companies like Apple or Qualcomm.
Reading of JASM (TSMC/Sony subsidiary) it seems they're planning on producing at least 3nm chips there in their expansion plans.
If they didn't mention SVE2, I wouldn't have known either... that would make it arm9 at the earliest...
Quick Google result is it’s basically a Neoverse V2 system, so yeah, armv9.
Edit: it’s funny because this article from 2021 mentioned Fujisu precisely https://siliconangle.com/2021/03/30/arm-unveils-armv9-archit...
That was regarding the HPC Fugaku , this is next generation
No it is not at all Neoverse V2, which is obsolete and has much lower performance.
It is a custom Armv9.3-A design (same ISA like the Arm C1 CPUs from the flagship smartphones of 2026), but it has double-width execution units for SVE2 (i.e. 256-bit width, vs. 128-bit for the other Arm CPUs) and it has some ISA extensions for AI/ML, e.g. instructions for inference with FP8 (BF16 is already supported by the standard Arm ISA).
Despite the fact that Armv9.3-A may include SME (scalable matrix extension, like in the Apple CPUs and in the Arm C1 CPUs), Fujitsu did not mention SME, so I assume that they did not implement it and they rely on their enhanced SVE2 (which is not surprising, while the origin of Arm SME is at Apple, the origin of Arm SVE is at Fujitsu).
Monaka-X will include SME2 but it didn't make the cut for first-gen Monaka
They actually do not in that article. I had to look elsewhere to find that out. Weird, one would expect that a big announcement like this would probably need to be vetted against licensing agreements with Arm that would probably want that fact to be prominently highlighted in press releases.
It is in the article - they cite "SVE2" support
Yeah, searched the page for common architectures (x86,arm and Sparc since it was Fujitsu) since it'd seem silly to use a totally novel architecture but didn't find anything until noticing SVE2.
Interesting to note that Fujitsu did this when the SPARC architecture was popular: they enhanced/developed a CPU for the same ABI.
Seems they have decided they don’t want to create their own architecture, but take the popular architecture and build on it.
Thanks kinda how Japan works. Inventing something new is hard, but taking something that's a cool idea and making it actually usable is well within their wheelhouse.
See: - Blue LEDs - Quartz Watches - Lithium-ion Batteries - Bidets
I'm sure there's plenty of other stuff invented in Japan, but that's really beside the point.
The industry is now mature enough that nobody really wants the headache of a new architecture at this point in time, Risc-V got a pass for being a grassroots movement growing out off FPGA's since the IP situation was more or less clear or even predatory with all existing architectures, Power, Sparc,SuperH,68k,etc are buried for good reasons.
UC Berkeley developed RISC-V, so no doubt their influence within the computer science academia sphere had something to do with it.
I had this idea that Fujitsu was one of those Japanese conglomerates that had operations in all kinds of markets, from electronics to heavy earth moving equipment to air conditioners to locomotives. But that doesn't appear to be the case so I'm wondering which company I'm confusing them with.
Sun invented SPARC, Fujitsu just happened to be the last producer since they built mainframes on it. Reading it seems like they plan to stop producing Sparc processors and machines in a near timeframe so that makes this announcement less surprising.
For many years the Fujitsu SPARC implementations were much faster than the Sun SPARC implementations.
During the decade 1995-2005, until the 64-bit AMD Opteron servers with Linux offered an alternative that crushed any kind of SPARC CPU with Solaris by a much higher performance combined with a much lower price, the SPARC Sun or Fujitsu servers dominated the market for servers used to host the CAD/EDA design tools used in electronics engineering, for the design of integrated circuits or electronic equipment. This means that all the EDA software from vendors like Mentor, Cadence, Synopsis etc. was available only for Solaris. That kind of software could not be run on Windows, due to the 32-bit memory limit. During that time, whoever got Fujitsu servers instead of the sluggish Sun servers, was very lucky.
Not mentioning the architecture is rather strange. Being Fujitsu I'd expect SPARC, but apparently not. Maybe we're reaching a point where some circles either don't care or just defaults to ARM?
Obviously they can just ignore the license in the future and continue development out of a local branch, but it's also a bit disingenuous to speak of "sovereign infrastructure" and then use licensed processor design.
Fujitsu dropped SPARC for HPC ~seven years ago with their A64fx. That's where ARM SVE comes from in the first place.
Yes but it's their own design, not licensing Arm cores.
Thanks. Looked for that in the article and couldn't find it.
What are the GPU capabilities?
Fujitsu has a long history of designing their own micro but using a standard ISA for HPC. Used to be SPARC, now it is ARM. IIRC Fujitsu were the main architects behind the SVE ARM extension.
But Arm is owned by SoftBank, a Japanese conglomerate.
(I think Japanese sovereignty is the main point of the article.)
Iam very curious as to how well it'll perform outside of Asia.
And if they'll sell it, seeing as they market it as 'sovereign AI'.
Performance would probably be the same regardless of where you run the CPU.
Fun fact, Fujitsu used to run their own fabs in Japan until they were sold to UMC and now they're another UMC fab although they are stuck at like the 20nm node. So this is likely made at JASM.
I think they'd say if the CPU was fabbed in Japan, instead there's some very "careful" wording:
https://en.wikipedia.org/wiki/Japan_Advanced_Semiconductor_M...
Right below the headline it says:
Sounds pretty clear to me
So, server manufactured/integrated in Japan; actual chip manufactured elsewhere (but "developed" in Japan, whatever that means)?
Doesn't really make sense to me. The whole "sovereign" thing is to basically say you won't be screwed if you go to war with a country tomorrow because you are dependent on their supply chain. Nothing about this changes that IMO. They're still getting the chips fabbed somewhere else and run the same risks.
True, but besides wars there are other dangers, e.g. USA does not declare war when they apply "sanctions" randomly, under the legal theory that they can dictate how anything that includes some piece of US origin may be used by its owner.
Since that theory began to be used, anyone outside US has become anxious to become "sovereign", i.e. to stop using anything about which a foreign state can claim to have rights.
Isn't the supply chain on advanced semiconductors ludicrously extended?
Wafers rare Earth's, specialized equipment, etc?
They made a detailed presentation at Hot Chips.
All the component chips are made at TSMC, in various processes.
"Developed in Japan" means the same thing like "developed in USA" means for NVIDIA or AMD or Intel. You can design chips anywhere on the planet, as long as TSMC is willing to provide you the required documentation and EDA libraries (which it does only if you already are or they believe that you will be a big customer).
depends on whether you read it as
((2nm 3d stacked CPU) and server) (integrated, developed, and manufactured in Japan)
or
(2nm 3d stacked CPU) and (server integrated, developed, and manufactured in Japan)
Indeed, it is extremely clear: by not saying "manufactured in Japan", the words "developed in Japan" tell you that they drew up the designs for the CPU in Japan, and then sent those designs off to a fab outside Japan for actual manufacturing. Then they bring those CPUs back to Japan, so that the server can be manufactured in Japan even if the server components are not.
That manufacturing step may change in the future, but right now that CPU is not "Made in Japan", it's only "Designed in Japan".
They did a presentation at Hot Chips, with all these details.
The core die is made with TSMC-N2P, while the SRAM die and the peripheral die are made with TSMC-N5.
Their CPU is designed in Japan, like the AMD CPUs or the Intel CPUs are designed in USA (the latest Panther Lake Intel CPUs have returned for fabrication to USA in 2026, but they also cost double than the corresponding Arrow Lake models made at TSMC in 2025, and their GPUs are still made at TSMC).
The compute part must have been made in Taiwan because it uses TSMC N2. TSMC N2 is only available in Taiwan right now due to a law restricting the latest node to Taiwan for 1-2 years.
That's Taiwan's silicon shield which the taiwanese see as important. That's why japan has rapidus and the us has Intel
Yeah I had a slight hope that they'd be a one of the first real Rapidus customers. but apparently not (for this one at least).
It's 2nm, so it has to be manufactured by TSMC.
I would assume they want to switch to rapidus asap once it comes online
Where are the CPUs fabbed? They make a big deal about "sovereign" but is the CPU actually made in Japan, or do they rely on TSMC?
Also odd how big an emphasis they put on AI inference when they don't build the GPU?
The article says that it’s manufactured in Japan.
where exactly does it say that? I don't consider "designed and developed in Japan" to be the same as manufactured there
At the very top of the article: "Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan"
"Japan-developed CPU..." plus the server itself (parts, power supplies, case, etc.) integrated, developed and manufactured in Japan. It sounds like they're weasel-wording it a bit, but I don't think the CPU itself is manufactured in Japan, only "developed."
Yep. I'd say the CPU is manufactured outside, and the rest of the server in Japan.
Combining info from https://news.ycombinator.com/item?id=49740907 and https://news.ycombinator.com/item?id=49740901, it seems that they're relying on TSMC for the first batch and are planning to move production to JASM once that's ready?
Pretty sure it's worded in a way that it applies only to the server, so it should be read as two distinct things:
- Japan-developed 2nm 3D-stacked CPU
- server integrated, developed, and manufactured in Japan
Ever seen the phrase "diffused in Taiwan, assembled in Malaysia"?
GPU in AI world is essentially a set of specific matrix calculations that this CPU supports on hardware-level. Thought memory speed seems low.
According to a Feb press release, they're fabbed by TSMC:
Source: https://global.fujitsu/en-global/pr/news/2026/02/12-01?utm_s...
According to Rapidus webpage, they are explicitly starting mass production of their 2nm in 2027.
Source: https://www.rapidus.inc/en/iim/?utm_source=chatgpt.com
Their CPUs are faster at inference than many GPUs. These are server CPUs with 144 cores and wide memory interface.
Fujitsu has added some ISA extensions for AI inference to the standard Arm ISA, e.g. instructions for inference with FP8.
Even their previous CPU generation was for some time the champion in energy efficiency, with better performance per watt than the NVIDIA GPUs, until a newer generation of NVIDIA GPUs has leapfrogged them.
There are good chances that this new Fujitsu CPU might be again for some time the CPU with the best energy efficiency, but it remains to be seen how it compares with the recent GPUs.
Would be cool to get ahold of one of these in the US.
I do find it odd that the 2U model has less storage capacity than the 1U on their chart. That doesn't make much sense to me.
The two made-in-Japan, premium grade, Fujitsu laptops I had to work with were by far the worse pieces of electronic I have ever used. Both exhibited identical defects: fans running continuously at maximum speed, and the batteries would completely die within hours of the devices being powered off. Windows or Linux. And don't talk about that tiny unresponsive trackpad to me. Never again.
The Fujitsu Lifebooks were legendary for their robustness back in the day, I'm genuinely sad you had that experience.
The only fujitsu laptops I used was back in 2011- it was a budget version for on-call and it was fine..
Sounds like the fans running 100% probably contributed significantly to the issue here. If I had to guess it was likely that the C-States were disabled in BIOS somehow so the CPU clock was running at full tilt the whole time, maybe combined with a bad thermal paste job.
I'm not here to defend Fujitsu but I've had really bad experiences with basically every major laptop brand (Dell: majorly bad coil whine and especially faulty soldered RAM, HP: keys vanishing from the keyboard and very weak hinges that break all the time, Apple: The GPU unsoldering itself and the butterfly keyboard shenanigans).
I don't really know any brand with a flawless track record sadly.
Before anyone brings up Thinkpads, they're trading on a reputation that hasn't been true for over a decade. If you have fond memories of a thinkpad it's most likely that you had it from before 2016, or it's "fine", but certainly not great.
I do recall that Fujitsu had a very... unique approach to exposing the Setup option in their firmware. It was just a UEFI boot entry. If that entry was removed, you couldn't get back into the firmware setup; you had to find the name of the setup EFI file and then run that from the EFI console.
Their consumer electronics arm is completely unrelated and also majority owned by Lenovo
I was about to say if they hadn't said it was a Fujitsu laptop I would have guessed it was a ThinkPad. Describes my experience with a P1 exactly.
Reaching further back, the Sylistic line was mostly quite good/durable, and I still use my Stylstic ST-4110 when I need to use an old scanner or control a CNC machine on my back deck when cutting tropical hardwoods.
Really miss the transflective display and wish that there were newer devices with such technology
What year? The ones I had were the best. Very expensive, but service at the office & at home, very sturdy, very long battery life. That was somewhere early 2000s.
2012 and 2017 Celsius H. Very expensive and both maxed up but those fans always running even with limited usage made them the worse to use.
I just have Fujitsu Lifebooks at Home since 2007. I think they are good, came with a bloat free Windows. I bought even one for my Mom over 10 years ago. I just changed once the HDD with a SSD. Now all my Lifebook run with Linux just flawless.
https://en.wikipedia.org/wiki/British_Post_Office_scandal
Their software is so poor that it caused probably the biggest miscarriage of justice in history. Well that and Fujitsu staff colluding with post office executives to put innocent people in prison rather than admit there was a problem
I wouldn't trust that this CPU can do maths properly.
The 'Fujitsu' that made the shitty software was a British consulting company that was acquired by Fujitsu [1] to farm public contracts. This is made by the Japanese parent company.
[1] https://en.wikipedia.org/wiki/International_Computers_Limite...
You're right, but the scandal is so upsetting that I personally cannot ever look at Fujitsu the same way even. Also, I think you're neglecting that a large part of why it became so bad was that Fujitsu claimed there was no issue with their software. Even if it were built by a company they acquired, the response to the developing situation was something that the parent company should have stepped in to oversee.
It is ultimately Fujitsu responsibility.
Not trying to be obnoxious, but it's still Fujitsu.
I'm surprised at the downvotes for this, does anyone think I'm misrepresenting the facts? I genuinely don't think I am.
Never buy products of this dodgy company.
https://en.wikipedia.org/wiki/British_Post_Office_scandal
I think what's most notable about this for me is the reduced environmental impact. The specifications list liquid-cooling as optional and it highlights the CPU as the main processor instead of a power hungry GPU. My takeaway is that the next generation of AI HPC's will focus more heavily on sustainable operations.
One problem we’re going to have with AI hardware is coming up with a standard set of specifications that are comparable. I don’t really care about the CPU GHz and the memory bandwidth, at least not directly. What I really want to know is how many tokens per second this will deliver, but that also depends on the model. We need a standard metric for that. Perhaps we agree on a specific open weight model (e.g. GLM 5.3 Flash or Qwen vWhatever) and then measure TPS on the hardware of interest.
There is a fairly direct link between the two numbers. You can predict the latter from former reasonably well
I feel like even TPS is becoming less of a good metric as we're seeing certain models handle similar problems while burning far fewer tokens.
I always heard of GB/s as most important number ... AI told me their 8800 MT/s on 8 Byte, 12 DDR5-Channels means 845 GB/s. A Nvidia RTX 4090 has 1008 GB/s. Nvidia B200 has 8000 GB/s. Is this the right way of looking at it?
You have to load the model weights into VRAM over PCI-E (from RAM). So the (PCI-E) bandwidth strongly affects time to first token.
You have to run inference on the GPU by reading and writing to VRAM. So TFLOPS of the compute matters, and bandwidth to the VRAM (Always integrated with the GPU, rarely a bottleneck), and this strongly affects tokens/s
If you're doing training workloads or offloading to system RAM, it gets more complicated. (And mostly bound up trying to feed compute on time)
(Edits for clarity.)
On dedicated inference hardware I'd expect model weights to never leave the RAM, and you'd probably load them on startup before even starting to serve requests
Wow a corporate TLD in a wild, never seen this outside of the erstwhile domains.google
there's quite a few
like:
- home.kpmg - global.honda
and probably more I forgot about
but as you can see, they're pretty terrible for replacing .com domains
Funny they have money to buy TLD or even develop such a fancy CPU, but don't have money to compensate people affected by Post Office scandal.
I fail to see how they will take a significant market share or even break even on this venture.
It's an overcrowded market. Far better would be to focus on semiconductor supply chain, which Japan already supplies some elements, to sell to fabs.
Because the bottleneck is the fabs. If a new 2 nm fab came online today, it would immediately sell all its capacity to 2030 no problem, without Fujitsu trying this gambit.
It's not 'sovereign', the architecture is not designed in Japan, the silicon is not fabbed in Japan. This whole thing is sideways.
Doesn't matter. You have to start somewhere and this is a good start.
Japan historically behaves as if they have nukes and need a totally sovereign HPC capability to support their stockpile.
It's a custom Fujitsu micro-architecture.
The text says "next-generation CPU, FUJITSU-MONAKA, designed and developed in Japan".
It's ARM.
Every developed country is going to push for their own homemade chips. Japan used to make Sparc CPU's back in the day. Probably many others I'm not aware of.
Every developed country except Australia, perhaps. This government is atrocious. The record for highest number of bureaucrats per capita in the world is owned by Australia. All they know how to do is tax people. All the wealthy people have left, or are leaving.
You have been convinced by fake news, please recheck this claim, its false.
*rich people
Oh god you're one of those. People who complain about taxes have historically been on the wrong side of arguments.
I've been calling for high-multicore CPUs (at least 100 cores) with local memories for a quarter century now. No winners so far.
A Pentium 4 hit 3.8 GHz longer than that ago in 2004, so that's not special. And the estimated price will be $7,000-10,000, which isn't special either since that's about 10x more than it should be.
Hot take: GPUs disrupted the CPU industry to such a degree that CPUs never recovered, and like the k-shaped economy, the current status quo only serves a small fraction of customers. We can and should do better, but sadly we won't. Still, it's good that Fujitsu did this, for the competition if nothing else.
Unless they are used for multi-processing in classical UNIX fashion, or proper microkernels, most applications will hardly take advantage of them.
Managed languages runtimes are probably the ones that would be better equipped to take advantage of them, for distributed JIT, GC and asynchronous code.
The Connection Machine style with StarLisp.
Very few devs can write optimal multi-threaded code that explores the single digit count of cores on their laptops or phones already.
IMO we are still far from discovering a truly usable concurrency paradigm (if one even exists).
Maybe if designing software like digital circuits, which are more parallel in execution.
However one could argue that FPGAs are already that, and still we come back to a skill issue.
I am also not an expert, cannot make heads or tails about SIMD algorithms, and in what GPUs goes, only traditional shading languages.
Let alone algorithms that on top of that, should take optimal advantage of all available CPUs.
Back in 2008 Fujitsu has one of the best performing 10Gbps Switches. We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate. Fujitsu was way ahead of the pack. I always wondered what held them back from building a meaningful networking business in the US.
It was just timing. Their networking gear was built on some very impressive ASICs which became commodity chips a few years later before they could grow. Arista, Juniper, et al ate their lunch using other vendors’ IP. IIRC it was Broadcom and Fulcrum that released the chips that killed them.
Another broadcom related tragedy
Having a us based vendor. Fujitsu is huge in Asia for ISPs and so is Mitsubishi and Sumitomo (yes they made or used to make network equipment too)
I'm sure there were good reaasons for this, but man this just sounds bad. Like that's the spec I think NSA must give to their contractors outfitting 611 Folsom Street's Room 641A
There are companies that mirror their traffic to storage for a few days to be able to look back and troubleshoot issues. ALL their traffic, proactively.
I do that with very, very low traffic industrial automation systems.
For US federal government, "Office of Management and Budget" says to store 72 hours of packet capture, since 2021 (partially in response to SolarWinds attack). (M-21-31)
https://www.cisa.gov/sites/default/files/2023-02/TLP%20CLEAR...
Nah, packet mirroring is a standard networking feature and ideally every feature is line rate.
Indeed. Juniper had this over 20 years ago.
It does sound bad the way OP described it, but packet or HTTP mirroring is a standard feature of A/B testing and blue-green deployments of a very critical code. Mirror traffic between version 1 and 2, then compare response body, response time and return response A to the end users.
Port mirroring is the network engineer's equivalent of using breakpoints in a debugger for a programmer. It allows you to inspect what happens on the wire to get a clue for what's wrong.
Will it lock up thousands of innocent self-employed post office contractors as part of its instruction set?
https://en.wikipedia.org/wiki/British_Post_Office_scandal
Probably not, Japanese firms excel at hardware and generally build extremely poor software. This a hardware product, so consequently it is probably fairly good.
What caused this?
Honestly, no idea. In my career I have worked with a lot of Japanese firms and have always found their engineers to be excellent. I used to think it was a lack of consideration for UX, but that obviously can't be true when you see excellent examples of UX in hardware (e.g. Switch controls), even though its exceedingly rare on the software side.
If anyone on HN knows (there are several other commenters more deeply steeped in Japanese engineering culture), I would love to know why as well.
Japanese culture and work culture is very hierarchical/top down. They excel at waterfall but it's basically impossible to do agile in a Japanese office environment, and waterfall is suitable for hardware but terrible for building software.
I think Ruby would be a counterpoint - although it's the only one I know of.
I worked for a Japanese company although in a local office with local management. Through all those layers of protection I got the impression that they suffer from what all corporates suffer from - no software engineers that have made it to the top. Software companies run by people who have no clue.
In the US, for example, I think some people who have a clue or a partial clue have made it up the ladder in some places and therefore although everyone there has their horror stories of idiotic behavior it is probably somewhat better.
msgpack, mold, and lld are a few other examples of exemplary software.
This is just my hypothesis, but I believe the language barrier is an issue. English is effectively a necessity for programming because real programming languages with ecoystems, error diagnostics, documentation, and useful articles/blogs/books are written in English. Japanese software developers know enough to get by, but they're still by-and-large insulated from the Anglosphere learning culture, where ideas and best practices rapidly propagate in the open. I'm not involved in hardware, but I'm guessing this is less of a barrier for hardware because hardware/documentation tends to be proprietary rather than open anyways, so whereas the Anglosphere gains a cumulative learning advantage from an open source + technical blogging culture in their fluent language, this does not manifest so much in hardware.
Conglomerates in general, and especially Japanese ones, are heterogeneous enough that those are basically two different companies. The British "Fujitsu" was an acquisition that I'm certain they now deeply, deeply regret.
(Which in no way excuses their utter evil.)
I'm kind of a Fujitsu fanboy. Since i came across the fact that a pretty modern xeon ecc ram nvme Workstation boards can be optimized to only draw ~10 Watt idle is just impressive, that is less than a gaming router. Even older ones like D3417-B12 were crazy effizient.
I also liked their Primergy Servers, but never got one for a reasonable price to test it out.
Another thing was the Futro series (e.g. S930) that could be used as opnsense firewall or low budget proxmox host.
Unfortunately it was always hard to obtain the high quality stuff as a consumer and years ago they sold their mainboard section to kontron. So no new Fujitsu Mainboards for now... What a pity.
If anyone is curious and want to skip all the PR talk:
It’s ARMv9.
I wonder how many cores they plan to cram in there, at least like 256 right?
144 per the article
I hoped it would be SPARC.
I hoped it would be Alpha.
Why did they bury the lede so much on what the ISA is?
Press releases are sadly like that. Fujitsu's main page has it in the first line:
https://global.fujitsu/en-global/technology/research/fujitsu...
It is no secret it is Arm; MONAKA was announced in late 2024.
https://www.techpowerup.com/329761/fujitsu-previews-monaka-1...
Funny first thing I did was to search 'arm' and I got 1 result at the bottom of the page.
Hipe they'll be affordable
Here's the previous generation, two years ago: https://news.ycombinator.com/item?id=42415754
And here's more technical information about this generation: https://news.ycombinator.com/item?id=49443040
If they can’t build real enterprise sales and software support outside Japan, it’s just another cool Arm chip no one can actually buy
More detailed presentations about the Monaka CPU:
wccftech summary: https://wccftech.com/fujitsus-monaka-chip-3d-stacks-2nm-cpu-...
2026: https://global.fujitsu/-/media/Project/Fujitsu/Fujitsu-HQ/te...
2023: https://global.fujitsu/-/media/Project/Fujitsu/Fujitsu-HQ/te...
FugakuNEXT, the supercomputer it will be used in, also some details about the next-gen Monaka-X: https://global.fujitsu/-/media/Project/Fujitsu/Fujitsu-HQ/te...
For a single CPU: 844 GB/s memory bandwidth (12 channels of DDR5 RDIMM, 8800MT/s). 4.3-6 TFLOPS.
Basically comparable to a modern (although not top-end) GPU, so good for HPC & AI workloads. (though only 2 CPUs per node, with GPUs 4-8 is more common)
Thanks for clarifying why this makes sense for inference.
For anyone not familiar with Fujitsu's CPUs, worth noting their overall long history with HPC, and more recently, ARM. FugakuNEXT is "NEXT" because the current Fugaku (#1 on TOP500 for a time, and still in the top 10) also used an ARM CPU from Fujitsu.
Could someone knowledgeable please explain why an AI-focused CPU is superior to a GPU-based solution with an ARM CPU driving the high-level operations, please, especially when the RAM isn't on-chip?
I know that for GPU's the model weights have to be transferred over the bus initially, but that only has to occur once for inference use cases, so is the Fujitsu system more about training scenarios? Or is the focus more about efficiency, as these are ARM-based cores with AI additions?
Anyone have idea the likely cost range?
So, not in America. A sign of the geopolitical times? If so, a very unusual one to come out of Japan.