Sapphire's Strix Halo Mini-PCs Aim to Revolutionize Scalable AI Clustering

Sapphire is pushing forward with its Strix Halo mini-PCs, designed for scalable AI clustering. At Embedded World 2026, the company demonstrated two linked units processing large AI models. The goal is to create a system where multiple mini-PCs can work together seamlessly.

The Strix Halo mini-PCs each pack 128GB of LPDDR5X memory and are built to connect with other identical systems. A live demo at the event showed two units handling a 235-billion-parameter AI model. However, the current USB-C port only allows linking two systems at once.

Sapphire plans to overcome this limitation by switching to Ethernet for inter-system communication. The company envisions a daisy-chain setup with no upper limit on the number of linked units. While the Strix Halo targets clustering like Nvidia's DGX Spark, no official integration with Nvidia's systems has been announced yet.

The solution is still in development, with Sapphire aiming for a finalised version by Computex 2026. A more realistic launch is expected in the second half of the year. AMD, meanwhile, has not pursued a similar approach for freely interconnectable AI systems.

Sapphire's Strix Halo remains a work in progress, with Ethernet set to replace USB-C for broader linking capabilities. The company's demo highlights its potential, but full market availability is still months away. For now, the focus stays on refining the technology before wider adoption.