Microsoft is adding GPU support to Windows 11's on-device AI stack, OC3D reported. The change routes local inference through graphics processors rather than relying solely on the NPUs Microsoft promoted when the Copilot+ PC category launched.

OC3D's headline posed the question "Is Microsoft making NPUs useless?" — a rhetorical frame for what the publication describes as a significant redirection of Microsoft's on-device AI strategy.

What does the change cover?

OC3D's report does not enumerate the specific features affected. Microsoft's local AI stack refers to the set of capabilities that execute models directly on the user's hardware rather than calling out to cloud servers. The GPU path appears in unreleased or recently updated Windows 11 builds, according to OC3D.

Which GPUs gain support?

OC3D does not specify supported vendors, driver versions, or capability thresholds. The publication leaves open which discrete or integrated graphics products qualify. Modern GPUs from Nvidia, AMD, and Intel all expose general-purpose compute APIs capable of running transformer-class models, though Microsoft has not published a compatibility list.

Does this make NPUs useless?

OC3D poses the question; it does not answer it. NPUs are specialised processors designed for matrix-heavy inference at low power. GPUs can execute the same model classes through general-purpose compute paths, typically with higher absolute throughput but greater power draw. The two paths can run side by side rather than as a substitution. NPUs also retain an efficiency advantage for sustained, always-on workloads where quiescent power matters.

What has Microsoft confirmed?

Nothing publicly, at the time of writing. OC3D's report is the cited source. Microsoft has not issued a press statement, blog post, or release note acknowledging GPU support for Windows 11 local AI.

What to watch

Microsoft's Windows Insider release notes for the Dev and Beta channels typically document new AI compute paths as they ship. Confirmation in those notes — along with the supported model list and driver prerequisites — would convert OC3D's report into a documented feature. Concurrent driver updates from Nvidia, AMD, and Intel would corroborate the shift.

Why it matters strategically

Microsoft and its OEM partners have spent the past year marketing Copilot+ PCs on the strength of their neural processors. The certification, which requires an NPU meeting Microsoft's stated TOPS threshold, became a key buying incentive for new hardware. If the same on-device AI features run on any reasonably modern GPU, the certification's exclusivity narrows substantially.

The installed-base arithmetic shifts accordingly. Pre-Copilot+ systems equipped with discrete or modern integrated graphics outnumber Copilot+ units by a wide margin. Routing local AI through the GPU would put those features within reach of a far larger population of existing PCs.

What it means for buyers

For consumers weighing a Copilot+ PC purchase today, the practical step is to wait for Microsoft to confirm GPU support and publish its compatibility list before deciding whether the NPU premium is worth paying. For existing owners of systems with discrete graphics, the change would unlock AI features that currently require a Copilot+ certification — provided the GPU meets whatever minimum bar Microsoft eventually sets.