Phi Silica

Windows 11 is starting to loosen one of its most exclusive AI rules. Certain local AI functions can now run on more PCs, including systems that do not meet the Copilot+ NPU requirement, as long as they use supported Nvidia GeForce RTX 30 series GPUs or newer with at least 6GB of VRAM. The change matters because many gaming and creator PCs already carry strong AI-capable graphics hardware. Until now, those machines were often left outside Microsoft’s local AI path even though their GPU performance could handle demanding machine learning workloads. A narrower door, but an important one Microsoft first made the NPU central when Copilot+ PCs were introduced on 18 June 2024. That category also came with baseline hardware expectations such as 16GB of RAM and an SSD, reinforcing the idea that dedicated AI acceleration would be the key requirement for local AI features. The new direction does not remove that framework altogether. Instead, it gives GPU-based systems a limited entry point, signaling that Windows 11 AI is no longer tied only to devices built around an NPU. Developers get access first The rollout is not a broad consumer feature yet. Microsoft has placed the change in the developer layer through its built-in local language model APIs, with documentation and GitHub posts indicating that the API can now run on non-Copilot+ PCs with supported GPUs. That status is still experimental, so the immediate goal is to let developers build and test applications that can use on-device AI. Over time, this could become the basis for wider Windows features if Microsoft expands the support further. Phi Silica powers the new path The model behind this approach is Phi Silica, a compact AI model that does not need to be preinstalled on every device. It can be downloaded through Windows Update when an application requests it. Once available on the system, Phi Silica runs locally and uses compatible GPU resources instead of relying on cloud processing. This design keeps the feature available only when needed, rather than loading every PC with an AI component it may never use. Text tasks are the first focus At this stage, GPU support is limited to Windows.AI.Text APIs. That means the new capability is focused on language tasks rather than the visual AI features associated with Copilot+ PCs. Apps can use the APIs for summarizing long content, rewriting text, turning text into structured formats, and generating smarter prompts. For writing-heavy workflows, that gives local AI a more practical role without forcing every request through the cloud. Because the processing stays on the device, Microsoft can point to two main advantages: faster responses and better privacy. Users do not need to send data to external servers for these text operations, which makes the feature especially relevant for notes, documents, and drafting work. Copilot+ exclusives remain in place Despite the new GPU support, Microsoft has not opened the full Copilot+ feature set to non-Copilot+ hardware. Windows Recall and Click to Do still remain tied to systems with an NPU. So while the boundary has been softened, it has not disappeared. The new policy only opens the local language model layer, not the broader premium AI experience that still distinguishes Copilot+ PCs from standard Windows machines. Even so, the shift is meaningful for the Windows ecosystem. More RTX-equipped PCs can now participate in the local AI development path, and that may gradually broaden how Windows 11 handles on-device intelligence in the future. Source: inet.detik.com

RTX GPUs Can Now Replace the NPU Requirement in Windows 11, But Only for Text AI