A laptop displays an AI file-management interface connected to a processor and cloud servers.
Copilot+ PCs have spent much of their short life looking like sports cars kept in the garage. Each one has a neural processing unit (NPU), but after the early features, Microsoft gave that hardware little new work. Microsoft now says that will change with "hybrid intelligence." It's a new set of Copilot features that can read local context, take actions on the PC and run local AI models, and it can still send work to the cloud when needed. Windows Central covered it this week under the headline that the features "will breathe new life into Copilot+ PCs."

There is one important caveat. None of this has shipped yet. Microsoft only says rollout will start in the coming months.

A laptop displays an AI file-management interface connected to a processor and cloud servers. What Microsoft actually announced​

Microsoft first described the features in an October 7, 2026 Windows Experience Blog post by Pavan Davuluri, executive vice president of Windows + Devices. Windows Central's October 19 article came 12 days later. Unite.AI also reported that the announcement extends the new Copilot Microsoft introduced on September 25, 2026, which is built around three experiences: Home, Code, and Autopilot.

For Copilot on Copilot+ PCs, Microsoft lists three local capabilities:

  • Local context: If you allow it, Copilot can understand relevant content on your PC, such as files and recent activity, and use it in the task you're working on.
  • Local actions: Copilot can do things for you across Windows. Microsoft's examples are organizing files, checking device diagnostics, troubleshooting problems, coding and running workflows directly on the device.
  • Local models: Copilot can use AI models that run on the PC for suitable tasks and combine that local compute with cloud models when needed.

Microsoft's summary is that Copilot "understands your PC, takes action with your permission, and uses local models when it makes sense." It adds that the user stays in control "at every step."

In short: Copilot can see your work, act on it and choose where the processing happens. It only does this on Copilot+ PCs, and so far it's a promise.

How it plugs into Home, Code and Autopilot​

Microsoft linked each capability to one of the three parts of the new Copilot:

ExperienceWhat hybrid intelligence adds (per Microsoft)
HomeUses local context to pull in files you've been working on and help turn them into collaboration-ready documents
CodeCreates native Windows apps from a single prompt, runs code more securely with Microsoft Execution Containers (MXC) support, and uses local models where appropriate to help control token costs
AutopilotAdds local context, local models and PC actions to what Microsoft calls a persistent, proactive, personal agent that keeps working while you do other things

Look at how often tokens come up. Microsoft's own pitch says customers' AI needs are "outpacing what their cloud budgets can support." So hybrid intelligence is a cost story as much as a privacy or speed story. Every task a local model handles is one Microsoft (or you) doesn't pay to run in a data center.

The fine print: "hybrid" doesn't mean "private"​

The name is easy to misread. Hybrid does not mean everything stays on your machine. WindowsForum's earlier coverage pointed out a gap between how some outlets described the feature and what Microsoft actually said. PCMag's summary describes local models that don't send your queries to Microsoft's cloud. Microsoft, however, describes a hybrid system that sends work to the cloud when needed. Local models will handle some tasks on the device, but Microsoft has not said that all Copilot prompts or data will stay on your PC.

Microsoft hasn't yet said:

  • What data goes to the cloud for a given request
  • Whether you'll see an indicator showing local or cloud processing
  • How Copilot decides where a task runs, or which local model it uses
  • What happens when you're offline
  • What the permission prompts and settings will look like

Microsoft's footnotes also hedge on timing. Per Unite.AI's reading, availability and timing will vary by device, market, and silicon platform. The blog also lists Windows Update as a requirement. Don't expect every Copilot+ PC to get every feature on the same day.

In short: read "hybrid" as "local when it makes sense." It's not a privacy guarantee.

The NPU question​

Windows Central argues that requiring a Copilot+ PC "naturally implies" the features will run on the NPU. That's a fair guess, since a powerful NPU has been part of the Copilot+ hardware spec since the category launched in 2024. But Microsoft's announcement doesn't say which chip handles which task. In the same post, Microsoft describes Windows ML, its model runtime, as running models across GPU, NPU and CPU. The NPU may do much of the work, but Microsoft hasn't confirmed that.

There's an odd result too. The biggest local models in the announcement are aimed at a different class of machine. Those include MAI Code 1.1 Flash, which Microsoft is shrinking to 3-bit precision, and large NVIDIA and DeepSeek models built for RTX Spark-based "Builder PCs." The earlier WindowsForum report covered this under the headline that Microsoft restricts Copilot hybrid intelligence to Copilot+ PCs and excludes powerful GPUs. Microsoft says Copilot's local context, PC actions and on-device models will roll out to Copilot+ PCs in the coming months, but timing varies and other PCs aren't confirmed. If you bought a desktop with a large GPU, that's frustrating. Your machine may have more AI muscle than a thin-and-light laptop, but it isn't on the list so far.

Is the Copilot+ brand dead?​

Windows Central says Microsoft retired Copilot+ PC as a marketing brand this year. It says the name survives as a feature set, and that "Hybrid intelligence PCs" looks like the new marketing push. The same article says Microsoft announced Aion Instruct, an on-device small language model that replaces Phi Silica and can run on the NPU, GPU or CPU.

Those claims belong to Windows Central. Microsoft's October 7 hybrid-intelligence post doesn't mention Aion Instruct. It also uses "Copilot+ PCs" all the way through, including a section headed "Copilot+ PCs, with hybrid intelligence" and a link to shop for them. So the brand looks quieter on device boxes than in Microsoft's own blog.

Microsoft did give two numbers meant to show that local AI is already running at scale:

  • More than 2 trillion inferences run locally each month across Copilot+ PCs, covering search, imaging and video calls
  • More than 40% of laptops being built for business are Copilot+ PCs

These are Microsoft's figures and haven't been independently checked. The blog doesn't explain how they were measured. They describe existing features, not the new Copilot ones.

What IT admins should watch​

Local actions are where things get serious for IT. An assistant that reorganizes files and runs diagnostics on its own is a new kind of actor on a managed device. Microsoft's answer is the security model it announced the same day. MXC lets organizations define which files and networks an agent may access, with policies enforced at runtime. It pairs containment with agent identity and management through Agent 365 and Intune, allowing IT to separate an agent's actions from those of the device user.

MXC is now generally available on Windows 11, and Microsoft names it in the Copilot Code scenario. But Microsoft hasn't said that every Copilot local action will run inside MXC. Its footnote also admits that "governance at scale may require additional services."

A sensible plan before rollout:

  1. List your Copilot+ PCs. Microsoft says more than 40% of new business laptops qualify, so you may have more than you think.
  2. Check how Copilot is configured in your environment and how you manage updates, because Microsoft lists Windows Update as a requirement for these features.
  3. Wait for the permissions documentation before turning on file-aware features on shared or sensitive machines. Tech Journal gives the same advice: review Copilot permissions before enabling file-aware features, especially on work or shared computers where local data may be sensitive.
  4. Look at MXC policies now if you already run third-party agents. Microsoft says OpenAI's Codex, GitHub Copilot and others already support it.
  5. Watch Microsoft Ignite in November. Microsoft has said it will share more there about using hybrid intelligence in organizations.

Don't mix this up with GitHub's HydraFusion preview​

A separate developer feature was announced the same day and is easy to confuse with this one. GitHub's HydraFusion, which routes each task to the right model in the cloud, is being extended to Windows so it can also tap into models running locally on device. Hybrid intelligence on Windows powered by HydraFusion is coming to the GitHub Copilot app, GitHub Copilot CLI, and Visual Studio Code in experimental preview later in October 2026. That's a developer-tool preview with its own schedule. It's not the Copilot app feature for Copilot+ PCs.

The verdict​

Windows Central is right that Copilot+ PCs finally have a reason to exist beyond Click to Do and semantic search. An assistant that knows your files and can fix your settings is more useful than one that only chats. But the announcement leaves big gaps. There are no permission screens, no list of what goes to the cloud, no confirmed hardware mapping and no date. Microsoft has promised a lot before. Until hybrid intelligence actually reaches a Copilot+ PC, treat it as a plan and not a feature you can use.

 

References

  1. Windows 11's Hybrid Intelligence features will breathe new life into Copilot+ PCs - Windows Central Windows Central 2026-10-19T14:05:22+00:00
  2. Microsoft to Give Copilot Local File Context and OS-Wide Actions unite.ai
  3. Building Windows for hybrid intelligence | Windows Experience Blog blogs.windows.com