Your Windows 11 PC can now natively execute AI tasks, even if it doesn't have the Copilot+ designation.

      For almost a year, Microsoft has been indicating that the future of AI on Windows revolves around Copilot+ PCs. To access the most advanced local AI features offered by Microsoft, users needed a computer equipped with a dedicated Neural Processing Unit (NPU). That was the understanding. However, it seems Microsoft is changing the game.

      Newly released information reveals that the Language Model APIs in Windows 11 can now function on non-Copilot+ PCs, as long as they have an Nvidia GeForce RTX 30-series GPU (or a newer version) with a minimum of 6GB of VRAM. On the surface, this appears to be an update geared towards developers. Yet, it may signify one of the most important changes in Microsoft’s AI strategy for PCs since the launch of Copilot+ last year. More significantly, it prompts a long-standing question since the rise of AI PCs: Did we actually need NPUs at all?

      The era of CoPilot+ exclusivity has always felt somewhat uncomfortable

      When Copilot+ PCs were introduced in June 2024, Microsoft marketed them as the gateway to local AI experiences on Windows. To qualify as such, a device needed 16GB of RAM, SSD storage, and an NPU capable of achieving at least 40 TOPS of AI performance. The messaging implied that these specialized chips were vital for running AI tasks locally. While this was true concerning efficiency, it didn't convey the entire picture.

      Anyone knowledgeable about AI hardware understood that GPUs could effectively handle these tasks. In fact, modern graphics cards often outperform NPUs when it comes to running language models and generative AI applications. This is the reason many enthusiasts experimenting with local AI tools, ranging from small language models to image generators, have relied on GPUs for years. Yet, native AI experiences in Windows remained confined to devices with the Copilot+ label.

      This resulted in a peculiar situation. A gaming PC with an RTX 4070 had ample power to run AI models locally, but lacked access to Microsoft's native AI framework due to the absence of an NPU. Conversely, a lighter laptop with a qualifying NPU could use it. The recent change doesn’t eliminate that gap completely, but it certainly narrows it significantly.

      Microsoft seems to be preparing for AI without NPUs

      With the newly available Language Model APIs, developers can now utilize local AI capabilities on compatible Nvidia hardware. Microsoft announced that these APIs can operate on non-Copilot+ systems featuring RTX 30-series GPUs or newer, as long as they have a minimum of 6GB of VRAM. These APIs rely on Phi Silica, Microsoft’s compact on-device language model. Applications can employ it for activities like summarizing text, rewriting content, converting text into tables, formatting information, and generating responses from inputs.

      Consider it a lightweight, local iteration of the AI features typically associated with services like ChatGPT. The key difference is that everything is processed directly on the device instead of being sent to the cloud. This is essential for two reasons. First, privacy — if AI processing occurs on your PC, sensitive files, notes, emails, and drafts do not need to leave the machine. Second, performance — local AI features can function instantly without reliance on cloud servers, subscriptions, or an internet connection.

      An intriguing aspect is Microsoft’s approach to distributing these capabilities. If an application requires Phi Silica, Windows can retrieve the necessary model via Windows Update and run it locally on compatible hardware. Thus, the operating system is starting to regard AI models as another component of Windows, rather than as a premium feature limited to a certain category of PCs. This represents a noteworthy philosophical change.

      Is this the beginning of the end for CoPilot+ exclusivity?

      However, don’t get too carried away; this does not imply that every AI feature will suddenly be accessible on older Windows machines. Features such as Recall, Click to Do, and certain AI-driven creative tools still seem to be tied to systems equipped with NPUs. Currently, the expanded support is limited to Language Model APIs, which mainly focus on text-based AI experiences.

      Yet, history indicates that these barriers seldom remain in place indefinitely. Once Microsoft proves that local AI can operate effectively on common RTX hardware, it becomes increasingly challenging to defend why specific AI experiences should remain exclusive to NPUs. Developers are unlikely to be concerned whether the AI workload is running on an NPU or a GPU, as long as the experience is satisfactory. Consumers will certainly not care either. This is why this update feels more impactful than the change in documentation might imply.

      For now, it’s just one API, but it also marks Microsoft’s first significant move toward recognizing something many PC enthusiasts have pointed out all along: capable GPUs were never the obstacle. And if local AI can operate seamlessly on millions of existing RTX-powered PCs, the difference between a “Copilot+ PC” and an ordinary Windows PC may begin to matter much less than Microsoft originally intended.