"Which operating system do I need for AI?" The question comes up a lot. And the most honest answer is anticlimactically simple: usually none in particular.
So before you start overhauling your system — take a breath. The answer depends entirely on what you actually plan to do with AI.
When AI happens in your browser
This applies to almost everyone. You type your question into ChatGPT, Perplexity, Claude or Gemini and get an answer back. Done.
And here the truth is simple: the operating system doesn't matter at all. A browser is a browser. Windows, Mac, Linux, that old laptop, the kids' Chromebook — it all works the same. Anyone telling you to switch systems here is selling you a problem you don't have.
So: if AI means browsing for you, stop reading and just carry on. Honestly.
When you want to self-host
Now it gets interesting. You want to run AI models locally — on your own machine, without your input ending up in some stranger's cloud. Ollama for the model, OpenWebUI as the interface, maybe Qdrant for a memory. Almost all of it runs in Docker — a technology that packages programs cleanly so they start the same everywhere.
And here I'm clear: Linux is the shortcut. Docker runs there natively, no detours. No WSL2, no Hyper-V, no "why won't this container start" afternoon. It just works.
The best part: it costs you nothing. Truly nothing.
- No new hardware. Your current machine is enough. Linux often runs more smoothly on old hardware than the Windows currently on it.
- No purchase. Linux is free. No license key, no subscription, no activation.
- No risk. You don't have to commit. A dual-boot lets Windows and Linux live side by side — you pick at startup. And if Linux wins you over, you can wipe Windows later.
In short: trying it out costs an afternoon and a USB stick. That's it.
And the Mac?
Also fine. Works too. Docker runs on macOS as well, and the OpenClaw-style tools of this world feel most at home on Linux and macOS anyway. So if you already have a Mac: keep it, all good. I just wouldn't buy one specifically for this.
One more word on local models on the Mac, since that's often sold as the big advantage: yes, Apple's chips can technically do this really well. But honestly, I haven't yet found a use case for local models on the Mac that blows me away. If you like it — go for it. So far I get along fine without.
The catch that genuinely annoys me
So much for the relaxed conclusion: for serious AI tinkering, Linux is the obvious choice. And then something like this happens.
The Claude Desktop app — the official client from Anthropic, an AI company no less — has been available for Windows and Mac for what feels like forever. For Linux? Came last. As a beta. It only showed up at the end of June, and even then only for Ubuntu and Debian. Fedora and others: still waiting.
And because a beta wouldn't be beta enough, it's promptly missing features that Mac and Windows have long had — like letting Claude control programs directly. On Linux: coming later. At some point.
Of all things, the very platform on which half the AI world runs its stuff gets served last and half-finished. That, with respect, is enough to make you tear your hair out. But let's stay fair: there were always ways around it. If you wanted to give Claude file access on Linux, you reached for the command line or the Claude Code plugin for VS Code or Codium — which, incidentally, is exactly what builds this blog. That only taps into part of the ecosystem, but it works. That was the workaround for a long time. And honestly: the beta runs stable for me by now.
So not earth-shattering — you can fall back on Claude in the browser, and that runs flawlessly. But a bit of appreciation for the people keeping all this AI stuff running would have been nice all the same.
What you can do now
- Using AI in the browser? Stay where you are. No switch needed.
- Want to experiment locally? Grab a USB stick, set up a dual-boot with a beginner-friendly Linux (Ubuntu or Linux Mint) and try it. Windows stays put for now — the way back is always open.
- Got a Mac? Perfect, dive right in. Install Docker, done.
There is no "right" operating system. There's only the right one for you — and in nine out of ten cases that's exactly the one you already have.