The two OS kernels and API are super close (outsider perspective). I used win2k for like 10 years mostly on the back of applications and games supporting XP for so long. I can't recall the big API differences. Maybe XP has UAC and there were APIs to check for it? Anyways, I still have fond memory of manually patching out API calls hard coded into EXEs to bypass XP only parts which were almost always superfluous.

That's not really quite accurate. Windows NT began life on the i860, and then was ported to i386 and MIPS, and very shortly after the initial release, ported to Alpha. A year after that, PowerPC.

Through the life of Windows NT 4.0's life span, MIPS and PowerPC died early, and Alpha support was axed just before the very end. In Windows 2000's development, i386 and Alpha were the only ports left, and Alpha was axed before it could make it to final release.

x86 and Alpha lived simultaneously for most of its development. It wasn't done "first" on Alpha (quite the opposite).

True, but the flexibility that two attempts at porting to botched architecture introduced also resulted in 2003 becoming Microsoft's first x86-64 operating system. The very next day after the 2003 release, they released Windows XP Professional x64 Edition based on the 2003 codebase, which a user could run on their brand new x64 computer without much of an issue -- that is, if they had even heard of it... it was under-marketed because they needed to save the hype for Vista. So the struggle and failure may have been worth it in the grand scheme of things.

If Microsoft had just iterated on Windows 2000 forever instead of continuing with XP/2003 and beyond it's quite possible I wouldn't be a Linux desktop user today.

The NT 3.x series might have had stricter userspace separation than 4.x did where they moved things like GDI into kernel space, but it still fell over hard at even the faintest whiff of fuzz-testing the native API. It wasn't until 5.x (Win2K) that they started taking that sort of thing seriously.