Construction is in fact done at some point. You cannot build an unlimited amount of anything, let alone data centers.
Do you honestly believe there will be enough constant construction on data centers to replace millions of jobs lost to AI? Even if that was somehow true, there are tons of people that are not capable for one reason or another (age, physical fitness, etc) of doing construction work. Construction work also pays significantly less than many of the jobs being lost as is, so it's not a good replacement in that regard. And what happens when construction gets further streamlined and/or automated and requires fewer people? Or none at all, in the future?
Even your most optimistic hypothetical scenario results in millions of people doing hard manual labor for poverty wages. If anything, this just strengthens the case against AI and AI companies.
Construction is in fact never done. If it's not data centers then it will be something else: high-rise residences, bridges, spaceports, oil refineries, whatever. People always want more stuff and we'll never hit a limit in our lifetimes. Skilled construction work is tough and occasionally dangerous but it pays pretty well.
https://mikeroweworks.org/
Speculation about science fiction automation scenarios is not a sound basis for public policy. Lots of things could happen in the future. We can't reasonably plan ahead for all possible contingencies.