Geekom has brought something unusual to IFA 2026: four A9 Mega mini PCs wired together over USB4, running large AI models as a single distributed system. It is a demonstration rather than a product, but the idea behind it is worth understanding.
What it is doing
The A9 Mega uses AMD’s Ryzen AI Max+ 395 with up to 128GB of unified memory, of which as much as 96GB can be assigned to the GPU. That is what makes a single machine capable of running large language models locally — the memory pool, more than the raw compute.
But there is a ceiling. A model needs to fit in memory, and once you exceed what one machine holds, you are stuck. Geekom’s demonstration links four machines over USB4 so their combined resources can tackle models no single unit could handle, without a server rack, without specialist networking, and without the noise and power draw of datacentre hardware.
Why photographers should care, eventually
Not today, to be clear. This is an engineering demonstration, and nobody is buying four mini PCs to edit photographs.
The reason it is interesting is what it says about direction. Local AI is becoming genuinely useful for creative work — culling, keywording, denoising, transcription, generative editing — and the constraint has always been that a bigger model needs a bigger machine. If capability can instead scale by adding another small box, that changes the economics for a studio.
It also sidesteps the reason many photographers are wary of cloud AI. Everything stays on your own hardware, which matters for client confidentiality, for NDA work, and for anyone who would rather their images were not used to train someone else’s model.
The honest caveat is that USB4 is not a fast interconnect by datacentre standards, and distributing a model across machines introduces latency that a single unit does not have. Whether the approach is genuinely efficient or merely possible is exactly what independent testing would need to establish.
Still, the A9 Mega has been one of the more capable compact machines we have looked at, and seeing what four of them can do together is a reasonable glimpse of where this is heading.
Geekom distributed AI setup: what we know
Configuration: four Geekom A9 Mega mini PCs connected over USB4, operating as a distributed local-AI system. Per machine: AMD Ryzen AI Max+ 395, Radeon 8060S graphics, up to 128GB unified memory with up to 96GB assignable to the GPU. Purpose: running large AI models locally without a traditional server rack. Status: demonstration shown at IFA 2026 — not a commercial product, with no pricing, packaging or availability announced.
