Microsoft is stepping up its push to bring artificial intelligence directly to personal computers, unveiling a new coding model that can operate locally and security technology designed to prevent autonomous AI agents from accessing data or performing tasks without authorisation.
The announcements, made Wednesday at an event in San Francisco, form part of Microsoft's broader strategy to transform Windows into a platform where AI agents can carry out increasingly complex tasks, from writing software to managing demanding business projects.
The company also introduced a high-performance Surface Laptop Ultra powered by Nvidia's RTX Spark chips, targeting users who require substantial computing power to run AI applications locally.
Microsoft challenges cloud-first AI model
Microsoft's move reflects a bet that some AI workloads currently processed in its Azure cloud data centres can instead be handled by powerful Windows desktops and laptops.
That shift could allow businesses and consumers to perform certain AI tasks locally, potentially reducing cloud computing costs while giving users greater control over sensitive data.
It also plays to Microsoft's long-standing strength in the personal-computer market, where customers rather than Microsoft typically bear the cost of purchasing the underlying hardware.
Apple is pursuing a similar opportunity with its latest Mac computers as both technology companies seek to make personal devices capable of handling more sophisticated AI workloads.
For Nvidia, meanwhile, a stronger AI-capable Windows PC market could provide an opportunity to expand further into a segment of the computing industry that has historically been dominated by Intel and Advanced Micro Devices.
Microsoft introduces safeguards for AI agents
As AI agents become increasingly capable of operating independently, Microsoft is also attempting to address concerns over security and unauthorised access.
The company announced Microsoft Execution Containers, or MXC, a technology designed to restrict AI agents from accessing data or carrying out unauthorised operations when running on desktop and laptop computers.
Pavan Davuluri, Microsoft's executive vice president for Windows and devices, said Anthropic, OpenAI and Nvidia would use the technology.
“We needed to make the desktop the most secure place for agents to execute,” Microsoft CEO Satya Nadella said.
Davuluri said the new tools would allow corporate IT departments to establish rules governing what AI agents can do, with Windows enforcing those restrictions on individual employees' machines.
Nvidia CEO Jensen Huang, who appeared on stage alongside Nadella, described the technology as a significant development for the deployment of autonomous AI systems.
“is going to revolutionize how agents are built and deployed,” Huang said.
“Without it, (agents are) a complete nonstarter,” he added.
The technology comes as the industry grapples with the security implications of giving AI agents greater access to computers. Nvidia is seeking to avoid a repeat of the hack involving AI platform Hugging Face, while Apple is also working to tighten controls over how AI agents can access files and other resources on Mac computers.
AI models move from cloud to Windows
Microsoft also unveiled AI models capable of running locally on high-powered Windows machines, reducing the need for constant cloud connectivity.
Among the models highlighted was Nvidia's open-source Nemotron model.
Davuluri said a version of Chinese AI company DeepSeek's V4 model could run on machines with at least 60 gigabytes of memory and outperform OpenAI's GPT-5 on some coding and reasoning tasks.
Microsoft is taking a hybrid approach with its own Copilot assistant, allowing workloads to be divided between cloud-based systems and local models.
“Copilot will still use the cloud for the hardest tasks, but for times when cost or privacy matter more, it can delegate down to local models that run directly on your computer,” said Jacob Andreou, Microsoft's Copilot chief.
Microsoft also said Meta's Muse personal assistant would come to Windows devices as a native application, using some of the company's new security technology.
The open-source AI agent system OpenClaw can also work with Microsoft's security tools, Davuluri said.
High hardware costs threaten local AI push
The biggest obstacle to Microsoft's vision may be the cost of the hardware required to run increasingly sophisticated AI models locally.
The new Surface Laptop Ultra will start at $2,599 and rise to $5,899 for a configuration featuring a 20-core processor, 128 gigabytes of memory and one terabyte of storage.
The price puts Microsoft's most powerful configuration firmly in the premium computing category.
For comparison, Apple's MacBook Pro with a 40-core graphics chip, 128GB of memory and 2TB of storage costs $6,700, although the specifications and configurations are not directly comparable.
Microsoft's entry-level Ultra is also considerably more expensive than Apple's $1,999 base MacBook Pro, although it comes with 24GB of memory compared with 16GB on Apple's entry-level model.
The cost challenge has become more pronounced as a shortage of memory chips pushes up prices across the technology industry.
Nvidia has increased the price of its DGX Spark AI desktop machine by about 75% since its launch, bringing the price to $6,950 amid rising memory costs.
The economics represent a sharp change from Microsoft's initial vision for AI-enabled PCs two years ago, when many of the targeted laptops were priced below $2,000.
‘Only people who have the budget’ can afford local AI
The combination of increasingly capable software and expensive hardware could limit the adoption of local AI, particularly among consumers and smaller businesses.
Anshel Sag, an analyst at Moor Insights & Strategy, said the technology had reached a point where both the software and hardware were capable of supporting local AI, but the cost of obtaining that hardware could prove prohibitive.
“the software wasn't ready, but the hardware was. Now the software is ready and the hardware is too expensive to actually run it locally,” Sag said.
“So it's becoming this thing where only the people who have the budget can really afford to run AI locally.”
Microsoft's strategy nevertheless signals a potentially important shift in the AI market: from systems that rely overwhelmingly on massive cloud data centres to a hybrid model in which powerful personal computers perform an increasing share of AI workloads.
The success of that transition could ultimately depend not only on the quality and security of local AI software, but also on whether hardware prices fall enough to put the technology within reach of a much broader market.