Sapphire's Strix Halo Unveiled: Scalable AI Clustering for Large-Scale Projects

Sapphire has unveiled its Strix Halo solutions at Embedded World 2026, targeting scalable AI clustering. The system allows multiple mini-PCs to link together for handling large-scale projects. A live demonstration showed two units processing a 235-billion-parameter AI model in tandem.

The Strix Halo mini-PCs resemble high-performance APU setups, each equipped with 128GB of LPDDR5X memory. Sapphire's goal is to create a daisy-chain configuration that can expand to an unlimited number of connected systems. Currently, USB-C supports linking only two units, but the company plans to introduce Ethernet as the primary method for inter-system communication.

The technology is designed to work with Nvidia's DGX Spark systems, enabling larger-scale AI deployments. While Sapphire aims to finalise the product by Computex 2026, industry observers suggest a more realistic launch window may be the second half of the year. AMD has not yet adopted a similar scalable clustering approach.

At the event, Sapphire demonstrated how two linked mini-PCs could collaborate on complex AI workloads. The company envisions this flexibility as a key advantage, allowing users to scale systems freely without vendor lock-in.

Sapphire's Strix Halo offers a modular approach to AI processing, with plans to expand connectivity options beyond USB-C. If successful, the system could provide an alternative to existing scalable solutions like Nutanix's hyperconverged clusters. The final product's release, however, may still face delays beyond the initial Computex target.