Geekom has launched a distributed AI cluster using four A9 Mega mini PCs powered by AMD Ryzen AI Max+ 395 chips, running DeepSeek V4 Flash. The cluster provides 512GB of RAM and connects via USB4, enabling enterprise AI workloads to operate locally without routing sensitive data through public clouds.
- Four A9 Mega mini PCs cluster via USB4 with 512GB RAM total
- Runs DeepSeek V4 Flash locally with OpenAI-compatible API
- Targets enterprise AI without cloud exposure, costs $16,000
What happened
Geekom has released a new AI computing cluster by linking four A9 Mega mini PCs, each equipped with AMD Ryzen AI Max+ 395 processors that combine powerful CPU cores, Radeon graphics, and unified memory into a small form factor. These devices are connected through USB4, forming a cluster with a combined 512GB of RAM running the DeepSeek V4 Flash large language model optimized for enterprise workloads.
Instead of using traditional data center servers or cloud infrastructure, the cluster runs the AI model fully locally on Ubuntu with ROCm and DwarfStar software that distribute the workload across the nodes. The setup includes an OpenAI-compatible API to connect enterprise applications and AI agents directly to the cluster while enabling organizations to keep all processing and sensitive data on-site.
Why it matters
This mini PC cluster architecture addresses key enterprise concerns around data privacy and cloud reliance by enabling powerful AI computation inside local infrastructure. Organizations can build private assistants that analyze contracts, manuals, and internal reports without exposing confidential information externally. The cluster’s ability to handle long-context prompts up to 250,000 tokens enhances its suitability for complex document-heavy tasks.
Performance testing from Geekom shows the cluster achieves approximately 14.61 tokens per second with a single concurrent process and a low latency for the first token generated. The modular design also offers flexibility—companies can start with one or two A9 Mega units and scale up to four as computing demand grows, balancing cost and performance.
What to watch next
Watch for further independent benchmarks and real-world testing to verify the cluster’s sustained performance and reliability in enterprise environments, as current figures are vendor-provided. Observing customer adoption will also reveal how organizations value local AI deployments compared to cloud-based options, especially for sensitive workloads.
Additionally, expansions of the ecosystem via software improvements or integration with other AI tools will be critical. Geekom’s approach to creating scalable, on-premise AI clusters could set a trend for smaller, more energy-efficient AI infrastructure outside traditional data centers.