At Hot Chips 2026, Fujitsu detailed Monaka, a 144-core Arm server CPU that stacks a 2nm compute die on a separate 5nm cache/IO die. It succeeds the A64FX chip used in Japan's Fugaku supercomputer and targets AI/HPC data centers, with sampling underway and volume production planned for 2027.
Fujitsu used its Hot Chips 2026 presentation to give the first detailed technical breakdown of Monaka, its next-generation 144-core Arm server processor. According to Tom's Hardware, lead architect Ryohei Okazaki described it as a 'made-in-Japan CPU, specifically engineered for AI performance and power efficiency,' developed with backing from Japan's NEDO for so-called green AI data centers. The chip is the successor to the A64FX, the Arm processor that powered Japan's Fugaku supercomputer.
The standout design choice is Monaka's 3D chiplet structure: a 2nm compute die built on TSMC's N2P process sits on top of a separate 5nm die that holds all of the last-level cache along with I/O, connected via hybrid bonding according to Wccftech. Fujitsu says splitting the cache and I/O onto a 5nm die while keeping the compute die on 2nm cuts total silicon area on the expensive 2nm die by about 30%, and the placement of the hotter compute die on top also shortens the thermal path for cooling.
Compared to the A64FX, Monaka narrows its vector units from 512-bit SVE to dual 256-bit SVE2 engines, but adds new data types like FP8 and INT8MM aimed at AI inferencing, plus support for confidential computing and per-core voltage/frequency control. The platform ships in two SKUs: a 350W air-cooled version running at 2.1 GHz and a 500W liquid-cooled version at 2.9 GHz, both offering the same 144-core count and scaling to 288 cores in a two-socket node. I/O includes 96 PCIe 6.0/CXL 3.0 lanes and 12-channel DDR5-8000+ memory support.
Fujitsu is already sampling Monaka to evaluation partners, with volume production slated for 2027. The company has also outlined a follow-up, Monaka-X, targeting a 2029 release with SME2 support and NVIDIA NVLink Fusion integration for the next-generation FugakuNext supercomputer.
What We Know
| Spec | Detail |
|---|---|
| Cores | 144 per chip (288 per 2-socket node) |
| Architecture | Arm v9.3-A with dual 256-bit SVE2 units |
| Compute die process | TSMC N2P (2nm) |
| Cache/IO die process | TSMC N5 (5nm), 3D hybrid-bonded beneath compute die |
| High-Performance SKU | 2.9 GHz base, 500W TDP, liquid cooling |
| High-Efficiency SKU | 2.1 GHz base, 350W TDP, air cooling |
| I/O and memory | 96 PCIe 6.0/CXL 3.0 lanes, 12-channel DDR5-8000+ MT/s |
| Timeline | Evaluation samples available now, volume production in 2027 |
Frequently Asked Questions
How many cores does the Fujitsu Monaka CPU have?
Monaka has 144 cores per chip, and in a two-socket platform configuration it scales to 288 cores total.
What manufacturing process does Monaka use?
Monaka uses a 3D chiplet design: the compute die is built on TSMC's 2nm N2P process, while a separate cache/IO die uses TSMC's 5nm N5 process, hybrid-bonded together.
When will Fujitsu Monaka be available?
Evaluation samples are available now, with volume production planned for 2027, according to Fujitsu's Hot Chips 2026 presentation.
What is Monaka designed for?
Monaka is a server CPU aimed at AI and HPC data center workloads, emphasizing power efficiency for what Fujitsu calls green AI data centers, and succeeds the A64FX chip used in Japan's Fugaku supercomputer.
