Microsoft unveils Windows hybrid intelligence, Surface Laptop Ultra, and a more capable Copilot

Microsoft wants your Windows PC to do more of the AI work

Microsoft is making a major push to turn Windows into a platform where AI agents can work across your PC and the cloud. In an October 7 announcement, Windows and Devices chief Pavan Davuluri outlined the company’s plans for Windows hybrid intelligence, alongside new Surface hardware, expanded Copilot capabilities, and actions built directly into Windows Search.

The idea is straightforward: not every AI task should need a trip to a remote server. Windows will increasingly use local models when they make sense, while keeping cloud-based intelligence available for work that needs it. Microsoft says that combination can help users get more out of their AI budgets without giving up access to more capable models.

There is plenty of new hardwarsee behind that plan, too. Surface Laptop Ultra and Surface RTX Spark Dev Box are opening for preorder, joining a broader lineup of NVIDIA RTX Spark-powered Windows PCs. However, several of the software experiences announced alongside them will arrive through previews or phased rollouts rather than becoming available to everyone immediately.

Copilot will use local files, take actions, and run models on your PC

One of the biggest changes is coming to Copilot on Copilot+ PCs. Microsoft says Windows hybrid intelligence will add three capabilities to its recently introduced Home, Code, and Autopilot experiences: local context, local actions, and local models.

Here is what that means:

• Local context: With your permission, Copilot will be able to use relevant files and recent activity on your PC to better understand what you are working on.
• Local actions: Copilot will be able to help organize files, assess device diagnostics, troubleshoot problems, write code, and carry out workflows directly on the device.
• Local models: Copilot will be able to use models running on your PC for appropriate tasks, alongside cloud intelligence when needed.

For example, Microsoft says Copilot Home could bring together files you have been working on and help turn them into something ready to share with colleagues. In Code, the company envisions creating native Windows apps from a prompt, with sandboxed code execution and local models helping manage cloud token costs. Autopilot will bring those capabilities into a persistent agent that can continue working while you focus elsewhere.

These features are expected to begin rolling out to Copilot+ PCs in the coming months. Microsoft notes that availability and timing will vary by device, market, and silicon platform, with Windows updates required for relevant experiences.

Microsoft Execution Containers adds guardrails for AI agents

Giving an agent access to files and the ability to take actions raises an obvious question: what stops it from doing more than it should?

Microsoft’s answer includes Microsoft Execution Containers, or MXC, which is now generally available on Windows 11. MXC lets organizations define which files and networks an agent can access, with those policies enforced while it runs.

The broader security approach has three parts: containment to limit access, identity to show which agent performed an action, and manageability to help organizations govern activity through services such as Microsoft Agent 365 and Intune. Governance at scale may require additional services.

Microsoft lists OpenAI Codex, GitHub Copilot, OpenClaw, Replit, LM Studio, NVIDIA OpenShell, and Unsloth AI among the tools already supporting MXC. Additional support is planned from providers including Anthropic Claude Code, Perplexity, and Raycast. Meta’s Muse personal AI agent is also coming to Windows as a native app with MXC integration.

These are security controls, not a promise that agents cannot make mistakes. The important shift is that Microsoft is building a way to restrict and track agent activity into the Windows platform rather than leaving every tool to handle it on its own.

Surface Laptop Ultra leads the new RTX Spark PC lineup

Microsoft’s new Surface Laptop Ultra is built around NVIDIA RTX Spark and offers up to 128GB of unified memory. The company says it can run AI models exceeding 120 billion parameters locally, positioning it for developers, creators, and people working with demanding AI workloads.

The laptop includes a 15-inch touchscreen and a new thermal design intended to support sustained performance. Alongside it, the Surface RTX Spark Dev Box brings local AI compute to a desktop form factor for developers and AI engineers experimenting with models.

Both devices are available for preorder starting October 7. Surface Laptop Ultra becomes available October 16, while the Surface RTX Spark Dev Box is scheduled to ship to U.S. customers in November.

Microsoft’s hardware partners are also opening preorders for RTX Spark-powered laptops:

• ASUS ProArt P16 and ProArt P14
• Dell XPS 16 Creator Edition
• HP OmniBook Ultra 16
• Lenovo Yoga 9n 2-in-1
• MSI Prestige N16 Flip AI+
• Microsoft Surface Laptop Ultra.

These laptops begin shipping October 16, with RTX Spark-based dev boxes and mini desktops coming later this year.

Microsoft also shared performance claims of up to 2.1 times faster time to first token, 4.3 times faster AI image generation, and 6.2 times faster AI video generation compared with a 16-inch MacBook Pro with M5 Pro. Those figures come from specific Microsoft-commissioned or NVIDIA tests using preproduction Windows hardware and selected models—not an across-the-board comparison of everyday laptop performance.

Local AI is getting more models and smarter routing

The hardware is only part of the story. Microsoft is bringing MAI Code 1.1 Flash to devices using 3-bit precision, which it says reduces the model’s size by nearly 80% while preserving coding quality and supporting a local 256K context window. The model has 137 billion total parameters, with 6.8 billion active parameters.

Microsoft also highlighted an upcoming NVIDIA Nemotron model and DeepSeek V4 Flash among the models being optimized for local use on RTX Spark. These capabilities target appropriately equipped hardware; they are not a promise that every existing Windows laptop will run the same workloads.

To decide where work should happen, Microsoft is extending GitHub’s HydraFusion model-routing technology to Windows. It will be able to route tasks to models running locally as well as models in the cloud. An experimental preview is planned for later in October in the GitHub Copilot app, GitHub Copilot CLI, and Visual Studio Code.

Microsoft is also adding llama.cpp support to Windows ML, giving developers more options for bringing open-source models into Windows applications.

Mini PCs and deskside supercomputers get a place in the plan

Microsoft’s hybrid intelligence strategy stretches beyond laptops.

At the smaller end, the company is introducing a simpler out-of-box setup for popular agents on mini desktop PCs. For OpenClaw, that includes a native Windows gateway and MXC integration, aimed at reducing the terminal setup and configuration work involved in running an always-on agent. Microsoft intends to expand these experiences to more Windows devices over time.

At the other end is NVIDIA DGX Station for Windows, scheduled to arrive later this year. Powered by the GB300 Grace Blackwell Ultra Desktop Superchip, these systems are designed for demanding local AI workloads, including models with up to one trillion parameters.

Microsoft says a DGX Station can also act as a shared AI resource for teams, supporting 32 or more simultaneous agents. Announced systems include Dell Pro Precision with GB300 and HP ZGX Fury AI Station. The pitch is greater control over sensitive data and less dependence on cloud inference for suitable workloads.

Windows Search will let you do more than find things

For users who are not buying an AI workstation, the Windows Search changes may be the most immediately recognizable part of the announcement.

Microsoft is introducing taskbar Search actions that let you make changes or complete tasks without first navigating through another app. Examples include:

• “Switch to dark mode”
• “Turn on Do Not Disturb”
• “Enable Airplane mode”
• “Dim my screen”
• “Minimize all my windows”
• “Text Elisa I’ll be there soon.”

Microsoft says the new actions begin rolling out to Windows Insiders in the experimental channel on October 7. It also plans an optional Copilot integration that can provide quick answers inside Search, with a path into the full Copilot app for longer conversations. That opt-in experience will start rolling out in select markets later this year.

Windows on Arm gaming gets another push

Microsoft is also using the RTX Spark announcement to highlight its continuing work on Windows on Arm gaming, including Prism compatibility improvements, anti-cheat support, and XBOX PC app compatibility.

Gears of War: E-Day is being used to showcase RTX Spark gaming capabilities, including DirectX raytracing, variable rate shading, and DirectStorage. Microsoft says it will also be the first game to offer expanded Advanced Shader Delivery support, cutting first-launch shader compilation from minutes to seconds and helping reduce in-game stutter.

The company highlighted Fortnite and Alan Wake 2 among the growing Windows on Arm catalog, with additional titles including League of Legends, Valorant, PUBG: Battlegrounds, and Genshin Impact coming soon. Microsoft also announced that Call of Duty is coming to RTX Spark in 2027.

What Windows users should watch next

Microsoft’s announcement ties together several different releases, so the rollout dates matter. MXC is generally available now, new RTX Spark laptops are entering preorder, Windows Search actions are starting with Insiders, and the more capable Copilot experiences are expected over the coming months.

The bigger change is how Microsoft wants Windows to handle AI: your PC supplies local context and compute, the cloud remains available when needed, and agents operate within defined permissions. That could make Windows more useful for everything from coding to everyday file management—but the real test will be how reliably those experiences work once they reach users.