Microsoft broadens on-device AI support in Windows

Microsoft is expanding its Windows AI APIs to reach more Windows 11 PCs, as part of a wider push to make Windows a stronger platform for local AI and agent-driven software. The company said the updated APIs will cover more CPU and GPU configurations, allowing developers to build features that run on device instead of relying entirely on cloud services.

The announcement came during Microsoft Build 2026, where the company outlined a broad set of updates aimed at developers working on modern apps, automation tools and AI agents. Microsoft said its goal is to reduce friction for developers while giving them more flexibility to run workloads locally, in the cloud or across both environments.

Among the key changes, Microsoft said its speech-to-text recognition API will now be available on neural processing units and CPUs. The company also said its on-device small language model support will expand to capable discrete GPUs, opening the door to local text intelligence features on more devices. In addition, Video Super Resolution will be available on CPUs, which Microsoft said will let developers deliver richer experiences without sending data to the cloud.

The company positioned these updates as part of a broader effort to support what it described as unmetered intelligence on Windows. That includes new on-device small language models, such as Aion 1.0 Instruct and Aion 1.0 Plan, which Microsoft said are designed to support faster local inference and reasoning tasks in the months ahead.

Microsoft also framed the changes around agent workflows, a growing focus across the Windows platform. The company said it wants developers to be able to build secure, capable agents that can operate locally when that makes the most sense, or coordinate with cloud services when needed. That approach is meant to avoid trade-offs between performance, security and flexibility.

The Windows developer updates announced at Build extend beyond AI APIs. Microsoft also introduced tools and infrastructure intended to help developers set up faster and work more efficiently on Windows 11. These include Coreutils for Windows, which is now generally available, WSL containers coming to public preview, and Windows Developer Configurations, which are also generally available. The company described those offerings as part of an effort to make it easier to move between Linux, Windows and cloud-based environments.

Microsoft also highlighted security features for agentic software, including its Microsoft Execution Containers SDK, which is now in early preview. The SDK lets developers define what an agent can access, such as files or the network, while enforcing runtime containment boundaries. Microsoft said that enterprise protections from products including Defender, Entra, Intune and Purview will help secure local agents as they become more common inside organizations.

The broader theme of the Build presentation was Windows as a trusted development platform for AI-era software. Microsoft said it wants Windows 11 to be more secure, more reliable and easier for developers to use, while also supporting new workloads such as autonomous agents and local AI processing.

For developers, the expanding AI APIs may be the most practical signal yet that Microsoft is leaning harder into local-first AI on Windows. By making speech, text and video features available across more hardware, Microsoft is betting that more agent workflows will be built to run directly on PCs, rather than treating the cloud as the default destination for every task.