Fujitsu Monaka ARM Server CPU 2026: UK Buyer Guide

What's actually going on sale

Fujitsu first discussed Monaka in 2023 and has now set a sales date of November 2026. Fujitsu teased Monaka as being derived from the silicon it designed for the Fugaku supercomputer, which topped the Top500 list from 2020 to 2022 and still sits in ninth place six years on. Monaka itself has moved a long way from that heritage — it's built on Broadcom's 3D chip-stacking technology, runs the Armv9 instruction set, and packs four 2nm compute dies (36 cores each) plus four 5nm SRAM chiplets around a central I/O and memory die offering 12 channels of DDR5 and PCIe 6.0. That adds up to 144 cores per processor. Fujitsu will sell the chip to cloud operators and server manufacturers, as well as in its own servers.

This matters for UK buyers because it's the first time Fujitsu has put commercial terms — even without a public price — around a chip it has spent years describing only in roadmap terms.

Three server designs, one processor

Fujitsu has created three server models that use the 144-core CPU, instead of leaving all system integration to third-party vendors. A 1U box has room for one or two CPUs clocked at 2.1GHz or 2.9GHz, eight E3.S SSDs and two M.2 drives, and can run air-cooled up to 40°C or liquid-cooled to 45°C. A 2U model pairs two 2.1GHz CPUs with four E3.S disks and two M.2 drives, and stays air-cooled only. A second 2U design, intended for four-per-rack multi-node HPC and academic deployments, runs its CPUs at 2.9GHz, has two E1.S SSDs plus two M.2 drives, and requires liquid cooling.

The spread tells its own story: Fujitsu is hedging between general enterprise racks and dense, liquid-cooled HPC clusters, rather than betting everything on one form factor.

• •1U: 1-2 CPUs, up to 45°C liquid-cooled operation
• •2U enterprise: air-cooled only, four E3.S bays
• •2U multi-node: liquid-cooled, four nodes per rack, HPC-focused

Why the sales date moved forward

Fujitsu spent years telling the Arm-curious that Monaka wouldn't arrive until 2027. Bringing commercial sales forward to November 2026 is a meaningful acceleration, but UK buyers should read it as a market-entry milestone rather than a signal that volume supply is instantly deep. Earlier technical disclosures around the chip's 2nm compute dies and 5nm cache stack were still framed against a 2027 production ramp, so procurement teams should expect early allocation to favour hyperscalers, cloud operators and Fujitsu's own channel before broader server-maker stock normalises.

That's a familiar pattern for new silicon architectures, and it's worth building into any evaluation timeline rather than assuming shelf-ready volume from day one.

The inferencing and sovereignty pitch

Fujitsu is leaning hard on two arguments. First, Monaka is pitched at double the inferencing output of rival CPUs — a claim that will draw attention from AI teams and neoclouds hunting for compute outside the usual GPU supply constraints. Second, because the smaller server models are built in Japan, Fujitsu is positioning them as a sovereign-friendly option, a pitch it clearly expects to land in Europe given the region's growing appetite for non-US-controlled infrastructure.

Add in the ability to run air-cooled at 40°C, and the appeal for UK operators in warmer or space-constrained facilities becomes more concrete — less cooling capex per rack is a real line-item saving, not just a marketing claim.

How it stacks up against EPYC and Xeon

Arm vendors have promised power efficiency for the best part of a decade, yet mainstream server-makers have mostly offered token Arm support rather than full product lines. Monaka's dual-socket configurations reportedly scale to 288 cores per node, which puts it in the same conversation as Intel's newer 256-core Xeon 7 parts on raw core density, even before efficiency claims are tested independently. UK teams currently running the numbers should compare EPYC Turin vs Xeon 6 against what Fujitsu is proposing, rather than assume Arm automatically wins on cost-per-core.

The absence of a published price is the biggest gap in the pitch right now. Until Fujitsu or its server-making customers publish list pricing, any TCO comparison against x86 alternatives remains an estimate.

What UK buyers should do next

Treat November as the point evaluation becomes possible, not the point mass deployment becomes sensible. Cost-sensitive inferencing workloads, sovereign-data projects, and HPC/academic clusters willing to commit to liquid cooling are the most obvious early fits. For everyone else, the sensible move is to find the right server against your actual workload profile before locking into a new architecture, and to model acquisition against server finance options given pricing is still undisclosed.

Procurement teams should also press both Fujitsu and any third-party server-maker reseller on delivery windows explicitly, given the gap between the November sales date and the 2027 volume-production context still attached to this chip in earlier technical coverage.