At IFA 2026, GEEKOM unveiled a local inference cluster built from four A9 Mega mini PCs interconnected via USB4, eliminating the need for dedicated rack or proprietary networking hardware. The system, demonstrated at the company booth, creates a compact supercomputer ideal for keeping AI workloads and data entirely local. Each node is powered by AMD's Ryzen AI Max+ 395 with 16 Zen 5 CPU cores, integrated Radeon 8060 S graphics, and up to 128 GB unified LPDDR5X memory.

The 171 x 171.1 x 70.1 mm systems include dual 2.5 Gb Ethernet, Wi-Fi 7, Bluetooth 5.4, USB4, and HDMI 2.1, with a 140 W TDP and NPU rated at up to 50 TOPS. The entire cluster draws less than 1,000 W at peak load. Targeted at businesses and workstation enthusiasts handling sensitive information, the cluster runs Deep Seek V4 Flash across all nodes with Ubuntu, AMD's ROCm stack, and GEEKOM's Dwarf Star distribution.

It processes up to 250,000 token context windows at approximately 14.61 tokens per second with single concurrency, featuring an Open AI-compatible API for seamless integration with coding and automation agents without cloud data transfer.