Microsoft is experimenting with running its Phi Silica model on Nvidia GPUs for Windows AI features, a sign that the company may be broadening how on-device AI works across PCs. The testing appears to be aimed at making the model available on hardware that does not rely solely on dedicated neural processing units, or NPUs.
Phi Silica is part of Microsoft's small model lineup designed for local AI tasks in Windows. The model has been associated with features that can run on-device rather than in the cloud, which can help reduce latency and improve privacy for some workloads. By bringing Nvidia GPUs into the mix, Microsoft could make those capabilities available on a wider range of systems.
The move is notable because many of Microsoft's recent Windows AI efforts have focused on Copilot+ PCs, a category built around NPUs for handling AI tasks locally. Support for Nvidia GPUs would suggest Microsoft is looking beyond that hardware path and exploring additional ways to deliver AI features on Windows machines that already include discrete graphics processors.
Microsoft has not said that the testing means a broad rollout is imminent. At this stage, the work appears to be exploratory, with the company evaluating whether Phi Silica can operate effectively on Nvidia-based systems. There is no indication from the source material that a public release date has been set.
If Microsoft moves forward, the change could matter for users with gaming PCs and creator laptops, many of which use Nvidia graphics chips but do not necessarily have the same NPU capabilities as newer AI-focused devices. Expanding support to that hardware class would align with a broader push to make Windows AI features more accessible without requiring buyers to purchase a new PC built specifically for on-device AI.
The testing also reflects a practical challenge for Microsoft and other software developers. AI features can be accelerated in different ways depending on the device, and support across CPUs, NPUs, and GPUs can make deployment more complex. Nvidia GPUs, however, remain one of the most common sources of AI acceleration in personal computers, which makes them an attractive target for broader compatibility testing.
Microsoft has been steadily adding AI functionality across Windows and its broader product lineup, often pairing those efforts with Copilot branding and dedicated hardware support. Phi Silica's potential role on Nvidia systems suggests the company continues to refine how its models are delivered, optimized, and distributed across consumer PCs.
For now, the development is best understood as a test rather than a finished product change. Still, it points to a possible future in which Windows AI is less dependent on a single class of processor and more adaptable to the hardware already in users' machines.