AMD Unveils Threadripper Halo Station, a Desktop-Class AI Supercomputer

At the Internationale Funkausstellung (IFA) technology show in 2026, AMD introduced the Threadripper Halo Station, a workstation built to run frontier AI models locally. Senior Vice President and General Manager of AMD’s Computing and Graphics Group, Jack Huynh, presented the system as the most powerful workstation currently available, capable of handling AI models with more than a trillion parameters.

Euronews‘This is about as close as you can get to a personal supercomputer’: AMD has Nvidia in its crosshairs with the Threadripper Halo AI workstation

The Halo Station integrates a 96-core Threadripper Pro 9995WX CPU, two MI350P data-center accelerators (expandable to four GPUs), up to 2 TB of system memory, and 576 GB of high-bandwidth HBM3E memory. System-level bandwidth reaches roughly 16.4 TB/s, and total memory capacity is about 3.4 times that of Nvidia’s DGX Station, AMD’s direct competitor. While Nvidia’s DGX Station is already on the market at an approximate price of €100,000, AMD has not disclosed pricing for the Halo Station but expects it to exceed that level. The workstation is slated for release early next year.

Euronews

Industry analysts view the announcement as a potential catalyst for the desktop AI computing market. Rising subscription costs for cloud-based AI services have driven enterprises to seek on-premise alternatives. Research from Gartner projects that token consumption costs for AI agents could surpass developer salaries by 2028, while Signal65 reports that agentic workloads may consume up to fifteen times more tokens than traditional chatbots. IDC identifies a new “sidetop” category of high-density, near-user AI systems, suggesting that devices like the Halo Station could help organizations curb token-related expenses.

‘This is about as close as you can get to a personal supercomputer’: AMD has Nvidia in its crosshairs with the Threadripper Halo AI workstation Euronews

Jack Huynh described the Halo Station as a platform that lets individuals and small teams develop and run large-language models without relying on costly cloud infrastructure. AMD frames the product as a means to “redefine personal AI” and to provide data-center-class performance using standardized components, thereby lowering the barrier to entry for serious AI work.

Euronews

If adopted, the Halo Station could reduce reliance on cloud APIs, which are increasingly expensive as token usage climbs. By enabling local inference for large models, enterprises may achieve cost savings comparable to those reported for Nvidia-based workstations in recent benchmark tests. The system’s high memory capacity also positions it to support future AI workloads that exceed current cloud-based token limits.

‘This is about as close as you can get to a personal supercomputer’: AMD has Nvidia in its crosshairs with the Threadripper Halo AI workstation