At the Internationale Funkausstellung in Berlin on Friday, AMD unveiled a workstation it describes as a 'personal supercomputer,' designed for what the company calls a 'completely new era of computing' driven by artificial intelligence. The Threadripper Halo Station, presented by AMD Senior Vice President Jack Huynh, is an ultra-high-performance machine that brings data-center-level AI capabilities to a desktop format.
The workstation is built around up to 96 cores of Threadripper Pro processors, two MI350P data centre accelerators, and a staggering memory configuration: up to two terabytes of system memory and 576 gigabytes of high-bandwidth memory (HBM3E). All of this is kept cool by a liquid cooling system, a necessity given the thermal demands of such hardware.
Huynh called the Threadripper Halo Station the 'most powerful workstation in the world,' capable of running AI models with more than a trillion parameters. To demonstrate its potential, he generated an entire 3D world and a flight simulator from a single prompt, a glimpse of the kind of workloads that were previously confined to cloud data centers.
Bringing AI home
AMD's pitch is straightforward: as demand for AI grows across industries, organisations increasingly want to run large language models and agentic AI systems on their own infrastructure rather than paying recurring cloud fees. The Threadripper Halo Station is designed to make that possible, offering a platform that can handle massive models locally.
The device is slated for release early next year, though AMD has not yet disclosed pricing. Industry expectations put it above €100,000, which means it is aimed at corporate and institutional budgets rather than individual consumers, despite the 'personal' label. For European research labs, engineering firms, and financial institutions, the cost may be justified by the savings on cloud AI subscriptions, which can accumulate quickly.
The direct competitor is Nvidia's DGX Station, which also brings data-center AI to a desktop form. The two take different approaches: AMD's machine dominates on memory capacity and bandwidth, offering more than three times the memory of Nvidia's, enabling it to run trillion-parameter models locally. Nvidia, however, offers more seamless integration when moving code from a local workstation to the cloud, and crucially, its product is already available at around €100,000.
Industry analysts see AMD's announcement as a potential catalyst for reshaping the competitive landscape of the local AI computing market. With cloud AI costs adding up for many institutions and businesses, demand for computers like the Threadripper Halo Station to run large models on their own infrastructure is expected to grow.
For European companies, the shift toward on-premise AI could have broader implications, from data sovereignty to energy consumption. Running AI locally means sensitive data does not have to leave the building, a key consideration under the EU's General Data Protection Regulation. It also means more control over uptime and performance, but it requires significant upfront investment and technical expertise.
As the AI race intensifies, the choice between cloud and on-premise will become a strategic one for many European organisations. AMD's entry into this space gives them another option, and the competition with Nvidia is likely to drive innovation and price adjustments. The Threadripper Halo Station may not be for everyone, but it signals a new era where the most powerful AI tools are no longer confined to the cloud.
