Sapphire's Strix Halo Redefines AI Scaling with Ethernet-Based Clustering

Sapphire is tackling a key challenge in AI hardware: seamless communication between multiple systems. At Embedded World 2026, the company revealed its Strix Halo solutions, designed to enable scalable AI clustering. These systems aim to connect Nvidia DGX Spark units in ways that current technology cannot match.

The Strix Halo platform focuses on overcoming the limits of existing connections. USB-C, for instance, only supports linking two systems at once due to port constraints. Sapphire's approach instead uses Ethernet to create a freely interconnectable network, allowing users to daisy-chain an unlimited number of devices.

During the event, a live demonstration showcased two mini-PCs running a 235-billion-parameter AI model. The setup highlighted the potential of Sapphire's clustering solution, which no other manufacturer has paired with Nvidia DGX Spark systems as of March 2026. AMD, meanwhile, has not adopted a similar interconnectable design. The Strix Halo mini-PC is expected to launch as early as this summer, though Sapphire aims to finalise the product by Computex 2026. A more realistic release window falls in the second half of the year, according to the company's roadmap.

Sapphire's Strix Halo could change how AI systems scale by removing current hardware limitations. The Ethernet-based design allows for flexible, large-scale clustering without the restrictions of USB-C. If successful, the solution may set a new standard for interconnecting high-performance AI units.