Sapphire's Strix Halo Redefines AI Scaling with Modular Mini-PCs

Sapphire is showcasing its new Strix Halo solutions at Embedded World 2026. The system aims to revolutionise scalable AI clustering by allowing multiple mini-PCs to link together. A recent demo proved two units could jointly handle a 235-billion-parameter AI model.

The Strix Halo is a compact mini-PC equipped with 128GB of LPDDR5X memory. Its key feature is the ability to connect with other identical systems, forming a larger AI processing network. Currently, USB-C limits connections to just two units, and Ethernet support is still under development.

At the event, Sapphire demonstrated a live setup where two linked mini-PCs processed a massive AI workload. The company envisions a future where an unlimited number of systems can daisy-chain together. This would allow for far greater scalability in AI projects.

Sapphire plans to refine the Strix Halo PC ahead of Computex 2026, though a wider release is expected later in the year. Ethernet is the preferred method for future inter-system communication, addressing current limitations. While Nvidia's DGX Spark systems exist, no official combinations with Strix Halo have been documented yet. AMD has also not adopted a similar interconnectable approach.

The Strix Halo solution targets large-scale AI applications by enabling flexible, linked setups. Sapphire's roadmap includes finalising Ethernet support and expanding system compatibility. If successful, this could change how AI clusters are built and scaled.