ASRock Industrial and Phison redefine Edge AI deployment with advanced memory expansion
ASRock Industrial and Phison Electronics have announced a strategic collaboration to accelerate large-scale local AI deployment through next-generation AI memory expansion powered by Phison aiDAPTIV technology. Together, ASRock Industrial AI BOX-A395 and Phison aiDAPTIV enable 120B-parameter large language models to run locally with only 64GB of system memory and 85GB of aiDAPTIV cache memory, reducing traditional memory requirements by up to 50% and significantly lowering the barrier for large-scale edge AI deployment. At Computex 2026, both companies further demonstrated this solution, enabling large-scale LLMs to run stably on-device and delivering a highly efficient and seamless edge AI experience. This collaboration delivers next-generation edge AI with reduced total cost of ownership (TCO), faster inference performance, and support for larger AI workloads on compact edge devices. This breakthrough enables scalable local LLM deployment across ASRock Industrial’s latest Edge AI platforms, providing greater flexibility for enterprise and industrial AI applications. ASRock Industrial will also feature validated Phison aiDAPTIV configurations through its AI-Pathfinder tool, helping customers identify optimised AI systems based on workload requirements and deployment scenarios.
The AI BOX-A395 with AMD Ryzen AI Max+ 395 reduces system requirements for large-scale AI workloads, while the NUC Ultra 300 BOX Series with Intel Core Ultra Series 3 CPU enables on-device inference of large language models in a compact form factor. Leveraging aiDAPTIV’s MoE (Mixture of Experts) inference architecture, AI BOX-A395 can run 120B-parameter LLMs with only 64GB system memory, while 128GB configurations intelligently offload model data to SSD cache, freeing DRAM/VRAM for operating systems, edge agents, and additional AI workloads. This architecture improves overall system utilisation and enables a single edge device to handle inference and multi-task AI workloads simultaneously. By transforming SSD into a high-speed memory layer optimised for AI inference, aiDAPTIV removes traditional memory constraints and enables larger models to run more efficiently without relying on oversized infrastructure, significantly enhancing flexibility and deployment scalability for real-world edge AI applications.
As AI workloads continue shifting toward local and on-premise deployment, ASRock Industrial and Phison are advancing a new generation of scalable AI infrastructure designed to bring larger models, greater efficiency, and more flexible deployment closer to where intelligence is needed most. By combining high-performance edge AI platforms with next-generation memory expansion architecture, the collaboration represents a significant step toward making large-scale local AI more accessible, practical, and ready for real-world deployment across enterprise and industrial environments.
