[zephen]

Speculation:

1) This could actually be an attempt to gain more revenue from big customers that have users who use the free version to test that code can synthesize and run unit tests (by pretending to use smaller parts), and then only use the paid version for the final integration into the actual larger parts.

2) This could give them more customer data more easily. They make no secret of the fact that the free tiers share data with the mothership for product improvement reasons. Maybe they only want to maintain the infrastructure to do this on Windows, or maybe it's harder for customers to subvert on Windows.

3) There will be people running the Windows version on Linux, and explaining how to do it, in 3... 2... 1...

[consp]

Are the prices so insane this is actually a viable option instead of just buying a bigger license? I always tell juniors to just submit a request for a licence since it will likely cost far less than retraining them or making them work harder and more to compensate for splitting things up (not FPGA related though).

[zephen]

The problem with FPGA builds is that the search space is ginormous, and repeated optimizations are tried in order to, e.g. make it meet the circuit timing requirements.

So, depending on exactly what you are doing it might take many hours to do a full build. And that might soak up all the capacity of a computer. And your Xilinx licenses are either node-locked (so only on that computer) or floating (so, only for one process/user on one computer at a time). You could conceivably have a big computer and timeshare multiple jobs on it, but (a) then you have to have the node-locked license, and (b) no matter what, you'll be slowing down your long job somewhat, by reducing the number of cores and amount of RAM available to it.

So it's definitely worthwhile to have multiple builds of different things going, preferably on different computers.

Organizations that use these sorts of tools typically have a lot of different tools that cost huge bucks compared to Xilinx software (think $100K/seat vs $4K), so this means that (a) they have entire organizations devoted to license management and working hard to ensure that all licenses are reasonably utilized; and (b) the relative cost to them to counteract this move by Xilinx (AMD) and just buy a few more damned Vivado licenses will not really be that high.

Now, do I think this is short-sighted? Yes, probably.

But do I also think that it could be revenue-positive for AMD in the short term? Yes, probably.