Microsoft on Wednesday unveiled a broader strategy called "Windows for hybrid intelligence", aimed at making Windows PCs capable of using AI models both locally on the device and in the cloud. The announcement was made by Pavan Davuluri, Microsoft's Executive Vice President for Windows+ Devices, in an official Windows Experience blog post which was then shared by Microsoft CEO Satya Nadella in a post on X.
ALSO READ: The New PC Era: Microsoft, Nvidia CEOs Unveil Next-Gen AI Laptop In San Francisco Event
Nadella also highlighted the strategy at the company's event in San Francisco, where he appeared alongside NVIDIA CEO Jensen Huang. Microsoft is positioning the move as a way to bring more AI processing directly to personal computers rather than relying entirely on cloud infrastructure, Reuters reported.
The central idea is that AI tasks can be performed locally when appropriate and shifted to the cloud when more computing power is required. Microsoft says that this move could improve privacy, responsiveness and the economics of running AI workloads.
MAI Code 1.1 Flash
One of the major announcements is the the move of Microsoft's MAI Code 1.1 Flash coding model to local Windows PCs. Microsoft states that MAI Code 1.1 Flash has 137 billion total parameters, and was designed for real-world coding workloads. It states that it is using 3 bit precision to reduce the model's size by nearly 80 percent, while maintaining coding quality and allowing it to operate with a 256K token context window locally.
Microsoft stated that the local version uses around 53GB of memory and retains much of the capability of its cloud variant, while optimisation and speculative decoding are used to improve responsiveness.
The company also said that Windows will support additional locally running models, including an upcoming NVIDIA Nemotron model with more than 70 billion parameters and DeepSeek V4 Flash, a 284 billion parameter model, on PCs based on NVIDIA RTX Spark.
Microsoft has additionally announced llama.cpp support in Windows ML, giving developers access to more open-source AI models and allowing them to experiment with newer models as they become available.
GitHub Copilot Gets Local AI
Microsoft is also bringing hybrid intelligence to GitHub Copilot, allowing the coding assistant to decide whether a task should be handled by an on-device model or a cloud-based model. Microsoft said the approach is intended to balance speed, performance and cost while giving developers more control over where AI workloads are processed.
The company stated that the capability is expected to arrive by the end of October and will initially target NVIDIA RTX Spark Windows PCs, including the Surface Laptop Ultra. It also said that the system will allow Copilot to coordinate local and cloud interface automatically, reducing the need for developers to make infrastructure decisions themselves.
Copilot To Get More Control Over Windows
Microsoft said that Copilot+ PCs will receive three major local capabilities - local context, local actions and local models - as part of the hybrid intelligence strategy. With user permission, Copilot will be able to understand relevant files and recent activity on a PC.
The company said that Copilot will also be able to perform actions across Windows, including organising files, assessing device diagnostics, troubleshooting problems, coding and completing workflows.
Microsoft stated that Copilot will be able to combine local processing with cloud intelligence depending on the requirements of a particular task. It demonstrated the approach through its Autopilot experience, which can work with local files and perform multiple steps on a user's behalf. The company stated that hybrid intelligence features for Copilot are expected to begin rolling out on Copilot+ PCs in the coming months.
Microsoft Execution Containers For AI Security
Alongside AI capabilities, Microsoft announced the Microsoft Execution Containers (MXC), now generally available, to restrict what AI agents can access and do on Windows. According to Microsoft, developers and IT administrators can define resources such as files and network destinations that an agent is allowed to use, with those restrictions enforced at runtime.
Microsoft stated that MXC is built around three principles - containment, identity and manageability - allowing organisations to restrict agent access, distinguish agent activity from user activity and manage agent deployments.
Reuters reported that Microsoft is positioning the technology as a safeguard against increasingly autonomous AI agents that can access data, execute code and perform tasks on computers. Nadella said that the company wants to make the desktop a secure environment for agents to operate.
Surface Laptop Ultra Unveiled
Microsoft has also unveiled the Surface Laptop Ultra, powered by NVIDIA's RTX Spark technology, as part of its push towards high-performance local AI computing. Microsoft said that the machine can be configured with upto 128GB of unified memory and can run AI models exceeding 120 billion parameters locally.
Reuters reported that the Surface Laptop Ultra starts at $2,599, with higher-end configurations reaching $5,899. The report also noted that the high cost of memory and AI hardware could limit local AI adoption among consumers.
Microsoft's broader strategy is to divide AI workloads between PCs and cloud data centres, potentially allowing tasks where privacy, latency or cost are important to be handled locally while more demanding workloads continue to use cloud-scale models. The move represents Microsoft's attempt to make Windows a platform for AI agents while shifting some workloads away from its Azure infrastructure, according to Reuters report.
Microsoft stated that the new generation of Windows PCs will also include systems from ASUS, Dell, HP, Lenovo and MSI, expanding the hardware ecosystem for its hybrid AI strategy.
ALSO READ: OpenAI Takes On Microsoft With ChatGPT's New Office Suite-Style Features
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.