At the IFA 2026 trade show in Berlin on September 4th, AMD unveiled the Threadripper Halo Station: a desktop workstation built to run AI models with more than a trillion parameters entirely on-premise, with no dependency on the cloud.
The system, announced by Jack Huynh, senior vice president and general manager of AMD's Computing and Graphics group, is built on the Threadripper PRO 9995WX processor with Zen 5 architecture, featuring 96 cores and 192 threads, up to 2TB of eight-channel DDR5 memory, and 128 PCIe Gen5 lanes. On top of that sit up to four AMD Instinct MI350P accelerators, each with 144GB of HBM3E memory at 4TB/s of bandwidth — in the base two-card configuration, the system already delivers 288GB of GPU memory, expandable to 576GB with all four cards installed.
Both the processor and each MI350P accelerator carry independent liquid cooling, a departure from the air-cooled norm for workstations that reflects the heat AMD expects this configuration to generate under sustained AI workloads. Each MI350P card draws up to 600W.
AMD did not announce a price or final sales configuration. The system will become available sometime through 2027, and given its components, the cost is expected to land in the hundreds-of-thousands-of-dollars range — clearly a product for research labs and large AI developers, not the end user.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel