Fujitsu Opens Global Sales of Monaka AI Chip in November

TOKYO - Fujitsu Ltd. said Monday it will begin global sales in November of FUJITSU-MONAKA, a Japan-designed central processing unit aimed at artificial-intelligence inference, and will sell made-in-Japan Monaka servers first in Japan and Europe as governments and enterprises look for lower-power alternatives to GPU-heavy racks.

Monaka is a 3D-stacked Arm-based CPU that pairs a 2-nanometer process for the compute cores with a 5-nanometer process for cache and I/O, Fujitsu said in a Kawasaki press release. The company claims a peak clock of 3.8GHz, memory transfer speeds of 8,800MT/s and roughly twice the AI inference throughput of competing CPUs, with hardware acceleration for matrix operations alongside SVE2 vector instructions. Fujitsu also lists confidential computing based on Arm CCA so data in memory can stay encrypted even under multi-tenant cloud software.

Alongside the standalone chip - which Fujitsu plans to sell to cloud operators, data-center providers and server vendors, with broader U.S. and Asia-Pacific availability targeted for the fourth quarter of fiscal 2026 - the company is launching Fujitsu MONAKA Server units built at its Kasashima plant. From November it will offer 1U and 2U rack-mount servers for Japan and Europe, pitched at AI agents and inference workloads, with academia and high-performance computing getting multi-node 2U systems at the same time. Broader server shipments are slated to ramp from April 2027; select financial, telecom and manufacturing customers may get earlier access in the second half of fiscal 2026.

Kyodo News framed the move against intensifying U.S. and Chinese competition over AI hardware and rising European demand for Japanese-made kit on security grounds. Fujitsu argues Monaka can run inference with less data and power than GPU fleets that dominate training and many serving stacks today, and that air-cooled 1U designs - operable up to about 40 degrees Celsius ambient, according to the company - can expand AI capacity where water cooling is scarce. The release did not disclose pricing. Whether Monaka can chip away at Nvidia’s grip on AI accelerators will depend on real-world throughput, software tooling and how quickly sovereign buyers convert press-kit claims into purchase orders.