Microsoft Reveals Its “Unmetered Intelligence” Vision, Wants Your Windows PC to Run AI Locally

Microsoft is pushing Windows into a much bigger AI era. The company wants PCs to handle more intelligence locally, instead of relying entirely on cloud models. The vision is termed as “unmetered intelligence”, meaning powerful PCs can run AI without paying cloud costs for every token.

Microsoft wants your PC doing more AI work

Microsoft’s Corporate VP of Windows & Device Sales, Mark Linton, says that local and cloud computing should work together. PCs can handle everyday AI tasks while the cloud tackles heavier workloads. Linton further pointed to models reaching 120 billion parameters on desktop hardware. That could give developers much more room to experiment locally.

Microsoft has already outlined this direction through Windows ML and Windows AI Foundry. These tools let developers run models across CPUs, GPUs, and NPUs. The company previously described Windows as an opportunity to deliver “unmetered intelligence” through on-device computing.

Windows is also getting built around AI agents

The other major piece is security. Microsoft says businesses do not want autonomous agents running freely across corporate systems. Its Microsoft Execution Containers technology can isolate agents and apply security policies. Administrators can also monitor, audit, and log what those agents do.

Microsoft is also preparing a more developer-focused Windows experience. Recent work includes Project Zenith, which targets developer PCs with 64GB or more unified memory. The company says these systems can run 30-billion-plus parameter models locally without metered cloud tokens. That makes local AI increasingly practical for coding and agent workloads.

Microsoft wants Windows PCs itself to become part of the AI infrastructure, using local compute whenever it makes sense. That could mean lower cloud costs, faster responses, and more control over sensitive workloads. The cloud still remains important when local hardware cannot handle the task. Moreover, Linton says more developer and platform capabilities are coming in the months ahead.

In other AI news, Microsoft’s new MAI-Image-2.6-Flash is now available, offering faster and cheaper image generation. Meanwhile, OpenAI has brought GPT-6 Astra to ChatGPT Plus, although availability is still limited. Early reports also suggest that GPT-6 Astra can spawn large numbers of agents, which could put additional pressure on CPU resources.