About 40%, because if you're comparing relative to the early 2000s, mobile devices are now filling the role that only desktops filled them. Drawing a distinction makes about as much sense as distinguishing desktops and laptops.

Plus, given that when folks like Darl McBride said "He can't have written an entire OS" they were clearly not thinking about just the Linux kernel, I think there's a solid argument for counting macOS and iOS too, both of which depend on the assumption that open-source code is of comparable quality to proprietary, commercially-developed code. Everything from their core kernel to the shell to the database for storing contacts (SQLite) is code developed in ways that the McBrides of the world said was untenable. That bumps it to almost 60%.

(Don't try me, I still have my Slashdot-honed debate skills.)

Wow, that comment was misunderstood... I use Linux, don't use Windows, and still think it's obvious that there is a certain feature set, i. e. the classic consumer desktop, that the community tried and failed to compete for. Of course Linux is a perfectly fine OS, but it again took a commercial effort (Android) to reach mainstream adoption on mobile.

Please read this in the context of the analogy: the OS is the science, the UI is typesetting/distribution/etc.

So, yes, GNOME and KDE didn't put a dent in Windows. But that wasn't the argument the anti-Linux folks were making. They were saying that as a process of producing software, open-source development was missing some magical property of the development practice - testing, professionalism, sustainability, security, etc. - that only closed-source proprietary development could provide. The reason Linux desktop failed to displace Windows was none of that: there's no (serious) argument that customers stayed with Windows because it had better stability or security. They stayed with Windows largely for compatibility and because it was free. The only missing feature was 100% compatibility with existing Windows software, which wasn't part of the claim.

As it happens I think that's actually directly analogous to the open access fight now. Universities stick with pricy journals not because those journals provide better typesetting or distribution (or science) but because those journals are the only way to get to existing papers. If Windows had never existed until today, nobody would switch to it; similarly, if Elsevier had never existed until today, nobody would start subscribing to them.

Or you could ask what does the top 500 supercomputers all run... Linux. See: https://itsfoss.com/linux-runs-top-supercomputers/

Or servers, or mobile devices in the form of Android phones. Desktop usage is only one many use cases for Linux.

To be fair, running on anything in the top 500 is a pain in the butt. Cray's are a major pain. I hear the NSCCs computers (TaihuLight and Tianhe) are essentially linpack machines, in reference to their difficulty. But I've never used them. HPC is just overall difficult and frustrating. But it is getting better.

But if we look at servers, there's a ton of Linux servers out there. Might be a better metric than computers that a small subset of programmers use. Linux accounts for >30% of this share and Unix like is >60%. Those are really important.

Also a lot of embedded systems run Linux.

Either way I think we pretty much agree. Linux is everywhere. Just not big on desktops. But huge everywhere else (which has a larger share of the computing resources in the world).

Virtually all TVs (especially ones with any type of “smart” functionality) sold today run Linux too.