Data centre capacity: The first desktop supercomputer for local use of artificial intelligence is presented

AMD unveils world’s most powerful desktop workstation for AI computing At the International Funkausstellung (IFA) trade show in Berlin, American chipmaker Advanced Micro Devices (AMD) introduced the Threadripper Halo Station, a new high-performance desktop workstation designed to bring data center-level artificial intelligence (AI) capabilities to individual users. The system, described by AMD as an “personal supercomputer,” is aimed at a new era of computing driven by AI. According to the company, the workstation is capable of running AI models with over a billion parameters. The Threadripper Halo Station is equipped with up to two terabytes of system memory and up to 576 gigabytes of high-bandwidth HBM3E memory. Inside, it features up to 96 cores of the Threadripper Pro processor, alongside two MI350P data center accelerators for AI computation. Due to the system’s demanding performance requirements, it is also equipped with liquid cooling. During the IFA event, AMD demonstrated the workstation’s capabilities by generating an entire 3D world and flight simulator with a single command. This demonstration highlights AMD’s vision for local AI computing: enabling complex AI models and applications to run directly on user infrastructure, reducing reliance on distant data centers. AMD aims to meet growing demand for AI computational power by shifting data center-like capabilities into a desktop form factor. The company says the system would allow businesses and institutions to locally run large language models and AI agent systems instead of relying on increasingly expensive cloud services. However, the term “personal supercomputer” might be misleading, as the device is not intended for typical home users but rather for companies, research institutions, and other organizations with substantial budgets. The price of the Threadripper Halo Station has not yet been announced, but industry analysts expect it to exceed €100,000. This positions the workstation well above traditional personal computers and places it primarily within the professional workstation and AI infrastructure segments. Nvidia already offers a similar solution in the form of its DGX system, which enables the execution of large AI models in a desktop environment. While both companies aim to provide comparable performance, AMD emphasizes massive memory capacity and high bandwidth, claiming the Threadripper Halo offers more than three times the memory of competing solutions. Nvidia, however, holds a significant advantage in terms of software ecosystem, allowing for smoother code transfers between local workstations and the cloud. Additionally, Nvidia’s DGX is currently available for sale at around €100,000. Industry analysts suggest that AMD’s announcement could significantly impact competition in the rapidly growing market for localized AI computing. For businesses and institutions, the cost of using AI in the cloud can be a major consideration. As the usage of large AI models continues to expand, localized infrastructure may become an attractive alternative. AMD plans to release the Threadripper Halo Station early next year. The exact pricing will be revealed closer to the launch date.