Fujitsu Monaka

Specifications

Fujitsu Monaka — A64FX Successor for AI Data Centers

Monaka is Fujitsu's next-generation Arm server CPU, unveiled in depth at Hot Chips 2026. It succeeds the A64FX (best known for powering the Fugaku supercomputer) and moves Fujitsu squarely into the commercial data center market. The 144-core design uses the Armv9.3-A architecture and a rare 256-bit SVE2 SIMD — twice the 128-bit width of most Arm server cores, with dual SVE2 units and dual 256-bit load/store units per core.

Monaka is engineered around energy efficiency. Fujitsu runs the chip at roughly 30% lower voltages than comparable designs, and uses chiplets across process nodes: the core die on TSMC N2P (2nm, GAAFET) for density and performance, and the SRAM/IO dies on TSMC N5 (5nm), joined with die-to-die hybrid bonding over a silicon interposer. Fujitsu notes N2P accounts for under 30% of total silicon area, accelerating time-to-market.

Power Optimization & Memory

Power-focused techniques include a floating-point register cache (ideal for GEMM-heavy AI workloads), predicate-register bypass, and masking of narrower vectors to cut SIMD power. Monaka offers 1, 4, or 8 NUMA node configurations (144, 36, or 18 cores per node) with up to 63 MPAM partitions, and integrates Arm Confidential Compute Architecture for confidential-computing deployments.

SKUs & Roadmap

Monaka ships in two 144-core SKUs: a high-performance variant at 500 W (2.9 GHz base) and a high-efficiency variant at 350 W (2.1 GHz base), with the former delivering roughly 50% more performance. Monaka arrives in 2027; Fujitsu is already developing Monaka-X on a 1.4nm process with NVLink Fusion support for tighter coupling to NVIDIA and other NVLink accelerators.

QS Compute supplies Fujitsu Monaka processors and Monaka-based Arm server platforms for energy-constrained AI data centers and sovereign-AI deployments.

Related Products

• Arm AGI — Arm's First In-House CPU
• Intel Diamond Rapids — Xeon 7
• NVIDIA Vera CPU
• AMD EPYC 9006 Venice