Nextorage Announces aiDAPTIV Station to Run Large AI Models on 32GB RAM PCs
Nextorage, a subsidiary of Phison, has announced the "aiDAPTIV Station," an external hardware device designed to run large language models like OpenAI's GPT-OSS-120B on PCs with only 32GB of RAM. The device connects via USB-C and uses Phison's aiDAPTIV technology to dynamically pool local DRAM and high-endurance NAND flash as virtual memory. This hardware could democratize the local execution of massive open-weight models on consumer-grade laptops and PCs without requiring expensive enterprise GPUs with massive VRAM. However, the practical performance and latency of running inference over a USB-C interface using NAND flash remain key challenges. The aiDAPTIV Station supports Windows and Ubuntu Linux, and Nextorage plans to launch it in Japan by the end of the year alongside pre-built software packages. It leverages Phison's aiDAPTIV technology to dynamically allocate resources between system DRAM and the external high-endurance NAND storage.
## BACKGROUND
Running large language models (LLMs) locally typically requires massive amounts of high-speed GPU memory (VRAM) or system RAM, which is costly for consumer hardware. Phison's aiDAPTIV technology addresses this by using high-endurance SSDs/NAND flash as an extension of system memory, allowing larger models to load by swapping data between DRAM and flash. GPT-OSS-120B is an open-weight Mixture-of-Experts (MoE) model released by OpenAI in late 2025.