I’m currently a little under the weather, and was about to physically shut down for the evening, when I caught a glimpse of Microsoft’s latest announcement in relation to Hybrid Intelligence. The article struck a chord, because in recent weeks and months I’ve been in and around discussions that have mentioned “the cost of AI” and probably, soon-to-be-in-the-dictionary-if-not-already, term dubbed “tokenomics”. More recently I was on the receiving end of an Apple presentation, in which the benefits of Apple hardware, with it being strict on privacy, and with the capability to run AI calculations locally on device was being sold as a big win, not only from a security and privacy point of view, but also that of a financial point of view. And I totally get it, meaty AI usage can cost hundreds if not thousands of pounds. If that usage can run locally on a more expensive device, then at somepoint, the device will have paid for itself – in theory.

This Hybrid Intelligence article, took me flooding back to those “I’m a PC” and “I’m a Mac” debates/adverts. This is clearly, Microsoft’s way of challenging, where Local AI and a new generation of hardware underpin Microsoft’s next chapter for Windows.

Microsoft has outlined its vision for hybrid intelligence on Windows: AI agents that can work on your behalf, run models locally and reach the cloud ONLY when a task needs more computing power. Announced today (October 7), well, seemingly an hour ago UK time, the strategy brings together new security controls, updates to Copilot, intelligent model routing and PCs designed for demanding AI workloads.

The idea is practical. As people ask AI to handle more work, cloud usage can become expensive. Running suitable tasks LOCALLY on a PC that’s capable could help reduce that consumption while preserving access to cloud models for other requests. Microsoft’s approach combines local models, software that decides where work should run and hardware capable of supporting it. (Much like Apple…)

It was this bit particularly, that peaked my interest, like I say, it coincides with what Apple was saying, and until now, something I thought Apple pretty much had a monopoly on. Sure, Copilot PC’s did (and do exist), but really, they don’t seem to have taken off. I think this new angle/push could help with that.

The changes this new angle brings to Copilot could be more visible to everyday users. On Copilot+ PCs, Microsoft plans to introduce three capabilities: access to relevant local context with permission, actions across Windows and the use of local AI models alongside cloud intelligence. That could mean helping organise files, troubleshoot a device or turning material already on a PC into something ready to share. Microsoft says these capabilities will extend across Copilot’s Home, Code and Autopilot experiences, with rollout expected to begin in the coming months. Availability will vary by device, market and silicon platform.

Naturally, this leads into discussions about new, capable, hardware… Within the published article, Microsoft discusses a new Surface Laptop Ultra, which offers up to 128GB of unified memory and, according to Microsoft, can run models exceeding 120 billion parameters locally. Pre-orders are now open (starting at £2599, or £5899 if you want 128GB) with availability beginning October 16. Surface RTX Spark Dev Boxes also became available to pre-order, with US shipments planned for November. At the higher end, DGX Station for Windows systems are expected later this year.

All in, the announcement suggests a broader role for PC hardware. A place where AI can perform ongoing work, with local computing helping support that activity. The practical test will be how reliably these experiences handle real tasks… unfortunately, I haven’t had the pleasure of putting a Copilot PC or Mac Device through its paces on that front.

Actually, that’s not quite true. I did have XCode create an app using it’s Claude interface a few months ago… but I didn’t have it in mind to take much notice of how well it performed, or handled things. I certainly wasn’t thinking about how a Windows device might fair in a similar situation. Maybe a shootout is one for down the line. Assuming PC hardware can actually be procured, that is……………