Microsoft is relaxing its strict policy on local artificial intelligence workloads, which it previously kept only for Copilot+ computers. The company announced that Windows 11 devices with appropriate hardware can now run native language model APIs.

With this update, systems with an Nvidia RTX 30 series or newer graphics card and at least 6GB VRAM capacity will be able to benefit from local artificial intelligence capabilities. This attack shows that Microsoft is reconsidering its strategy of ending on-device artificial intelligence features only with the Copilot + brand.

Expanding artificial intelligence capabilities with GPU support

When Copilot+ computers were released on June 18, 2024, special NPU hardware was required for local artificial intelligence processes. However, modern GPUs offer a very convenient structure to manage such workloads thanks to their high parallel processing capacities.

Microsoft has confirmed that developers can now run the language model APIs on computers that do not have an NPU but have supported GPUs. This feature is currently available at the developer layer rather than being offered directly to end users and works through applications using the Windows AI framework.

At the center of the system is a small local model called Phi Silica, which can be downloaded via Windows Update when needed. This model takes advantage of GPU power to perform processes such as text summarization, text editing and structured information creation on the device.

Advantages and current limitations of local processing

The fact that all processes run on local hardware instead of the cloud protects data privacy and also improves reaction times. Especially for enterprise users and developers, this makes a valuable difference in the adoption of AI tools.

Despite this, some prominent Copilot+ features, such as Windows Recall and Click to Do, still remain exclusive to NPU-equipped systems. Current GPU support is limited to only the language model API layer for now, rather than a broad artificial intelligence integration.

This step by Microsoft expands the compatible hardware pool by removing NPUs from being the only option for local artificial intelligence.

What do you think about this development, which somewhat reduces the special position of Copilot+ computers?