Microsoft’s long-awaited Windows Copilot Runtime is finally beginning to take shape with the release of a developer preview for the Arm build of its DirectML AI tool. Though it’s still far from production-ready, this update is an important step forward. Developers now have the opportunity to experiment with local AI applications using Copilot+ PCs powered by Qualcomm’s Hexagon neural processing units (NPUs). This marks a significant leap toward integrating AI into the Windows ecosystem, but there are still challenges ahead before the platform is fully optimized for everyday use.
To make AI work on the edge, hardware accelerators are essential. The Copilot+ PCs, which have been heavily marketed for their AI capabilities, come equipped with a powerful 45 TOPs AI accelerator. This accelerator is paired with a suite of AI models and development tools designed to help developers create their own applications. Despite some delays in the release of key components like the Win App SDK and PyTorch tools, this initial batch of resources offers a solid foundation for developers eager to explore AI on these new devices. The goal is to empower users to build local AI applications without the reliance on cloud processing.
At the core of the DirectML Arm release is a new driver for the Copilot+ PCs’ NPU, which is integrated with a newly updated Open Neural Network Exchange (ONNX) runtime. This update adds support for several open-source models from Hugging Face, allowing developers to access a range of machine learning tools and experiment with existing models. Additionally, Microsoft has curated a repository of DirectML-ready models optimized by Qualcomm, making them immediately available to developers working with the new runtime. These optimizations aim to boost performance, but the preview release comes with some limitations.
Despite the excitement, this early release is far from perfect. Serious bugs and a limited selection of supported models mean that developers may encounter frustrations when trying to build more complex AI applications. However, the integration of DirectML and NPU drivers represents an essential milestone for the Copilot Runtime and the broader Windows ecosystem. DirectML is a key API within the DirectX family that offers access to hardware accelerators like GPUs and NPUs, which is crucial for developing high-performance AI applications. For now, developers must manually install the Qualcomm NPU drivers by downloading them from the Qualcomm developer portal. This process, while somewhat cumbersome, will eventually enable smoother integration of AI tools across different platforms.