Microsoft Lets RTX 30+ GPUs Run Local Windows 11 Language Models

Overview

• Microsoft updated the Windows App SDK documentation to mark an experimental feature that lets Language Model APIs run on non‑Copilot+ PCs using supported GPUs.
• Eligible hardware is Nvidia GeForce RTX 30 series or newer with at least 6 GB of VRAM, which the SDK lists as the minimum threshold for GPU-based local models.
• The local models use a small on‑device model called Phi Silica that apps can download via Windows Update and run locally on the GPU for text tasks.
• The capability is developer-facing and limited to text-focused APIs such as summarize, rewrite, text-to-table, and prompt generation, and it does not yet enable Copilot+‑only features like Windows Recall.
• The change widens which machines can run on‑device AI and weakens Copilot+’s hardware exclusivity, a shift that could alter OEM marketing, developer targeting, and how users balance privacy and cloud AI use since Copilot+ debuted in 2024.