Windows 11 has no shared local AI model store, RTX Spark testing finds

Thurrott.com's testing on RTX Spark points to a mismatch between how Windows treats files and how local AI apps treat models. On Windows, a downloaded DOCX file is simply a file in the file system, available to the system and all apps and services, not sandboxed or isolated. Local AI models are also just files sitting in the file system, and Microsoft claims there are over 40 local AI models on every Copilot+ PC, but they are hidden somewhere and updated by Windows Update or the Microsoft Store. Whatever apps Microsoft makes available know how to find those models, but there is no formal storage location or system service acting as a middleman. As a result, a model downloaded in one app—Hermes, LM Studio, Clairvoyance, Windows Llama, or another—is not visible to a different AI app, so the same model is downloaded twice. That duplication can become problematic as model sizes balloon, especially on RTX Spark computers that can have up to 128 GB of RAM and allocate most of it to the GPU that interacts with those models. The author notes that experimenting with different AI apps, runtimes, and models means moving back and forth and re-downloading, and the digital detritus of this activity can start taking over the PC's storage. Cleaning house would be easier if the models' location were known, and it is possible (likely?) that uninstalling an app does not remove the models it downloaded.