Microsoft Tests Windows AI on Discrete GPUs to Expand Copilot+

What happened?

Microsoft has taken a significant step in democratizing local artificial intelligence on Windows. According to Tom's Hardware, the company is testing a new capability in the Windows App SDK that allows running Copilot+ AI features on discrete GPUs, instead of requiring exclusively a Neural Processing Unit (NPU). The feature is available in an experimental build of the Windows Insider channel with developer mode enabled. This suggests that Microsoft is exploring ways to expand the reach of Copilot+ beyond devices equipped with NPUs, such as the Snapdragon X Elite or upcoming Intel Lunar Lake and AMD Ryzen AI chips.

Historically, Microsoft has been cautious about hardware requirements for its AI features. Since the announcement of Copilot+ in May 2024, the company has insisted that the NPU is essential to ensure efficient performance and low power consumption, similar to how Apple has integrated the Neural Engine into its M1 and later chips. However, this stance has limited adoption to a fraction of the market: according to IDC data, only 10% of PCs shipped in 2024 include a dedicated NPU. With this test, Microsoft tacitly acknowledges that the NPU is not the only viable path, and that discrete GPUs, present in millions of computers, can take on these tasks.

Why is it important?

Until now, Windows 11's local AI features, such as Windows Studio Effects, Recall, or Cocreator, required a dedicated NPU to ensure efficient performance and low power consumption. This limited access to users with recent hardware. If Microsoft allows these features to run on discrete GPUs (such as those from NVIDIA, AMD, or Intel), millions of existing PCs with powerful graphics cards could benefit from local AI without needing to upgrade their equipment. This not only expands the user base of Copilot+, but also accelerates the adoption of AI in the Windows ecosystem, directly competing with Apple Intelligence on Mac.

The market impact is potentially massive. According to Statista, there were over 1.4 billion Windows PCs in use in 2023, a significant proportion of which have discrete GPUs, especially in the gaming and workstation segments. If Microsoft makes this capability official, gamers and content creators will be able to access features like enhanced background filter or image generation without relying on the cloud, reducing latency and improving privacy. For businesses, this means they can deploy local AI assistants without renewing their entire IT infrastructure, a key selling point against cloud-based solutions like ChatGPT Enterprise.

Consequences and context

This decision has strategic implications. On one hand, Microsoft reduces its dependence on chip manufacturers to drive Copilot+, which could pressure Qualcomm, Intel, and AMD to offer more competitive solutions. Qualcomm, in particular, has bet heavily on the NPU in its Snapdragon X Elite chips, and this move could erode its initial advantage. On the other hand, using discrete GPUs could imply higher energy consumption, which could affect battery life in laptops. However, for desktop computers or workstations, the performance of a discrete GPU far exceeds that of a current NPU. For example, an NVIDIA GeForce RTX 4090 offers up to 1,321 TOPS (trillions of operations per second) in AI, compared to 45 TOPS from a typical NPU like the one in the Snapdragon X Elite. This opens the door to more complex AI applications, such as local large language models or real-time video generation.

Furthermore, this move could incentivize developers to optimize their AI applications for GPU, leveraging technologies like DirectML. Microsoft has already worked with NVIDIA to integrate DirectML into its drivers, and AMD and Intel are expected to follow suit. However, it is not yet clear whether this capability will reach the stable version of Windows 11 or remain limited to development environments. Microsoft has not made official comments on this, but its inclusion in the Insider channel suggests the company is seriously evaluating the technical feasibility and impact on user experience.

"Microsoft's move to test AI on discrete GPUs is a game changer: it breaks the NPU entry barrier and puts local AI within reach of a much larger installed base." — TheVortiq

What should readers know?

• The feature is in an experimental phase and requires enabling developer mode in an Insider build. This means it may still have bugs or limitations.
• Not all Copilot+ features may be supported; some, like Recall, which requires constant memory access, might still need NPU for energy efficiency.
• Performance and power consumption will depend on the GPU used; high-end GPUs will offer a better experience, but low-end ones may not meet real-time performance requirements.
• This feature could arrive in Windows 11 in a future update, but there is no confirmed date. Interested users can try it now through the Windows Insider program, but with caution.

Impact analysis

If Microsoft makes this capability official, the PC market with local AI will expand significantly. Gamers and content creators with dedicated GPUs will be able to enjoy features like enhanced background filter or image generation without relying on the cloud. For businesses, this means they can deploy local AI assistants without renewing their entire IT infrastructure, reducing costs and accelerating adoption. However, there remains the question of how it will affect security and privacy, as local processing on GPU could expose sensitive data if not managed properly. Microsoft has already implemented process isolation in Windows to protect AI data, but developers will need to follow best practices.

Compared to past events, such as the introduction of TPM 2.0 requirements for Windows 11, which sparked controversy by leaving out older hardware, this move is a turn toward inclusion. Microsoft seems to have learned from that experience and now seeks to maximize the user base of Copilot+. However, the company will need to balance performance, power consumption, and security to avoid fragmenting the experience. In summary, Microsoft is paving the way for mass adoption of AI on Windows, and this test is a crucial first step that could redefine the landscape of personal computing.