Microsoft unveiled new technologies at an event in San Francisco that are designed to transform the Windows operating system into a major platform for autonomous AI agents. The company aims to move some of the computing from its Azure cloud data centers directly to users' personal computers, putting it in direct competition with Apple.
Microsoft unveiled new technologies at an event in San Francisco that are designed to transform the Windows operating system into a major platform for autonomous AI agents. The company aims to move some of the computing from its Azure cloud data centers directly to users' personal computers, putting it in direct competition with Apple.
Among the main announcements are the powerful Surface Laptop Ultra laptop based on Nvidia RTX Spark chips and a large-scale security system for autonomous AI models, Reuters reports .
To prevent data leakage and unauthorized actions of autonomous AI agents on local devices, Microsoft introduced the Microsoft Execution Containers (MXC) tool.
The technology isolates the actions of AI agents, preventing them from accessing confidential information or performing unauthorized tasks.
System administrators will be able to centrally set AI rules that Windows will enforce on employee PCs.
Anthropic, OpenAI, and Nvidia have already announced their intention to use MXC. Nvidia CEO Jensen Huang said that MXC “revolutionizes the process of creating and deploying agents.”
“We had to make the desktop the safest place for AI agents to perform tasks,” emphasized Microsoft CEO Satya Nadella.
In addition, Microsoft announced that Meta's Muse personal assistant will be released on Windows as a native app, and the open-source AI agent system OpenClaw will also support new MXC security tools.
Microsoft is moving to a hybrid model for the Copilot assistant: complex tasks will be handled in the cloud, and those prioritized for privacy or cost will be handled directly on the device.
On a Windows PC, you can now run locally:
Open-source Nemotron model from Nvidia;
a version of the DeepSeek V4 model, which on computers with at least 60 GB of RAM is capable of outperforming GPT-5 in programming and logical reasoning tasks.
To run heavy AI models locally, Microsoft, together with Nvidia, introduced the Surface Laptop Ultra.
Base model: Starts at $2,599 (includes 24GB of RAM, which is more than the base 16GB in Apple's $1,999 MacBook Pro).
Top configuration: The model with a 20-core processor, 128 GB of RAM, and a 1 TB SSD is priced at $5,899.
The main barrier to mass adoption of local AI remains the high cost of hardware, caused by the rise in the price of memory chips. Moor Insights & Strategy analyst Anshel Sag notes that while previously the software was not ready for local AI, now that the software has matured, the hardware has become too expensive for the average user. Currently, local AI remains a segment for corporations and users with high budgets.
