What Microsoft actually requires
Microsoft's own guidance describes a Copilot+ PC as needing an NPU of 40 or more TOPS, 16GB of RAM, and a 256GB or larger SSD. It also lists Windows 11 version 24H2 or newer as part of the minimum. That 16GB figure is a device-category threshold. It tells you whether a machine qualifies as Copilot+. It does not tell you that 16GB is ideal for every workload, or that you need 32GB to unlock Copilot+ features.
Microsoft's AI PC buying guide uses broader bands:
- 16GB is "good" and covers browsing, email, streaming and light multitasking.
- 24–32GB is "better", keeping apps responsive while AI-powered tools run in the background.
- 64GB is "best" for advanced AI experiences, creative projects and heavy multitasking.
This is retail buying guidance from a vendor that sells hardware. It is not a compatibility requirement and not a benchmark.
The leaked roadmap: interesting, unconfirmed
TechSpot reported on November 4 about a slide that appears to come from Microsoft and was posted by the leaker HXL. Memory sorts PCs into AI tiers in this slide. TechSpot said the Copilot+ tier splits into three bands: quick local questions at 16GB, delegated tasks at 32GB and long-running jobs at 64GB. The slide reportedly also places a plain 8GB "Windows PC" tier below those, and a 128GB "Builder PC" tier above them.
Treat this carefully. TechSpot itself says none of it is confirmed. A separate WindowsForum discussion of the same leak makes the same point: Microsoft hasn't published this as a spec sheet. The tiers and memory thresholds are a leak, not a purchasing rule. They do hint at where Microsoft wants Windows to go, with Copilot reading local files, taking actions and using on-device models. That would raise memory demands over time.
Windows 11 is getting leaner on 8GB
Microsoft's November 3 Windows Experience Blog post targets 8GB PCs specifically. Microsoft claims Windows is now up to 20% more responsive on those machines, and multitasking is up to 40% faster in its lab testing. Those numbers come with heavy caveats:
- Responsiveness: based on telemetry for Search, Start, right-click, Settings, Snipping Tool and File Explorer, comparing build 26300.9550 with 26200.7628.
- Multitasking: a single lab scenario on an HP OmniBook 3 with 8GB and a Snapdragon X Plus. It opened 20 Edge tabs, a 4K video, large Word, PowerPoint and Excel files, Outlook and File Explorer. Microsoft says this is not a benchmark.
- Memory tuning: compression changes let Windows keep up to 2GB more active data in memory. Widgets load only when opened on PCs with 8GB or less, saving roughly 200MB. Edge now sleeps inactive tabs after 15 minutes by default, down from two hours.
Microsoft's own wording is balanced. It says a PC with 16GB or 32GB can take on more, and that demanding work such as 3D modeling, advanced video editing or large-scale data processing benefits from 16GB or more. These gains do not make 8GB equal to 16GB or 32GB. They do mean the OS tax on memory is shrinking, which helps every configuration a little.
To get the improvements, Microsoft says to update to Windows 11 version 26H2, build 26300.9818 or later:
- Open Settings, then Windows Update.
- Turn on "Get the latest updates as soon as they're available."
- Select Check for updates.
- Keep Edge, Microsoft 365 and Teams updated, since some changes ship through app updates or the Microsoft Store.
Cloud AI versus local AI
This is the distinction that decides most purchases. A cloud assistant does its heavy lifting on someone else's servers, so your laptop just needs to run a browser or app comfortably. One buying guide puts it bluntly: if you only use cloud AI in a browser, an ordinary 16GB laptop is enough. That is an independent retailer-side view, but it matches how cloud services work.
Local models load into your machine's memory. Microsoft's Foundry Local documentation says it enables local execution of large language models on your Windows device, using compatible CPU, GPU or NPU variants. It lists "enough memory and disk space for the model you select" as a prerequisite. It also says a dedicated GPU, NPU or Copilot+ PC isn't required when the model has a compatible CPU variant. In its troubleshooting section, it suggests choosing a smaller model if a model misbehaves and memory is short.
Microsoft sets no universal RAM threshold for local AI. The right number depends on the model you choose.
One more wrinkle for local AI: on laptops with a discrete Nvidia GPU, one buying guide says graphics memory, not system RAM, decides which models run. That claim comes from a third-party site rather than Microsoft, so use it as a prompt to check your GPU's VRAM, not as gospel.
Which one should you buy?
| Your use | 16GB | 32GB |
|---|---|---|
| Browsing, streaming, schoolwork, Office, Teams | Fits Microsoft's "basics" band | Probably unneeded |
| Cloud AI assistants | Fine | Little extra benefit |
| Heavy multitasking with AI tools running | Can feel tight | Matches Microsoft's "better" band |
| Large video projects, memory-hungry games alongside other apps | Depends on the project | Gives more headroom |
| Small local models | Possible | More comfortable |
| Larger local models, VMs, development | Likely limiting | Better, with 64GB for the heaviest jobs |
HP's guidance lands in a similar place: 16GB with a 40+ TOPS NPU for everyday Windows AI features, and 32GB or more with a dedicated GPU for creative AI tools and running models on your own laptop. That is vendor guidance, but it agrees with Microsoft's bands.
Practical advice
- Check upgradeability first. Many thin laptops use soldered memory, and what you buy is what you keep. Check the manufacturer's spec sheet for your exact model. Some machines are upgradeable and many aren't.
- Price the premium. TechSpot notes that memory has only gotten pricier this year as AI data centers soak up supply, so the 32GB upcharge is not trivial. Compare the actual configurations you can buy.
- Test your real workload. Open your most demanding apps together and watch memory use in Task Manager before spending. IT teams buying in bulk should pilot representative workloads, not rely on spec-sheet tiers.
- Look at the whole PC. Microsoft's own commissioned testing, from Principled Technologies, found a 2026 8GB Dell XPS 13 matched a 2021 16GB model in several tests. Microsoft says memory still matters for heavy multitasking. That is company-commissioned and specific to one device, so don't generalize it.
The bottom line
For students, office workers and anyone leaning on cloud AI, 16GB is a sensible default and meets the Copilot+ minimum. Pay for 32GB if you multitask heavily, edit large media, game alongside other apps, develop, or plan to run local models. The leaked roadmap suggests Microsoft's AI ambitions will want more memory over time, but it is unconfirmed. Buy for the workload you have now, with a realistic eye on the next few years.