I haven't looked at the XP codebase but I bet it is massive enough that AI will choke on several Intel specific parts. If you are saying in a roundabout way that you have already done it successfully, tell us more.
Well, I meant that I have already done similar porting, but not on such a large scale, since the free models have a very limited context and usage limit, XP would probably require a paid and the most powerful model available, and it would also require porting many programs without which the OS itself would be useless.
Also, as a reference, one could use the source code of Windows 4.0 leaked into the network, which was already on the NT kernel, but at the same time had support for the PowerPC architecture, and just as I heard that there were projects to launch it on the PowerMac G5, which also means that the required drivers are available for this hardware, which means these drivers can be ported to Windows XP, and part of the PowerPC ASM code can also be taken from Windows 4.0 src.
I did porting only small projects - decompiling C# programs like MCCToolChestPE and games (Angry Birds, Fruit Ninja, PvZ, Chess) with Windows Phone 7 (Silverlight Runtime / XNA C#) through ILSpy, then processing through AI to refine the pseudocode into a working source code, while the AI coped even with heavily obfuscated code, which is already quite a lot. After that, I also used AI to port the existing code to Windows 10 ARM32 15035 (a networked ARM 32-bit build) and Windows 10/11 ARM64 (and x86, x64), which had never had native ports before.
Thus, the AI helped me transfer one Win x86 program and several WP7 games to previously never-supported platforms. Obviously, these projects are not comparable in complexity to Windows XP, but this is at least the equivalent of a very successful porting to another architecture using AI.