Because you don't generate new knowledge by simulation and following past experience. AI is trained on past experience, just like your undergraduate college students. When you hire a new grad you don't expect them to be able to invent new things right away.But what are they computing? 5 times nothing is still nothing. beyond virtual things, and communications I see no improvement in day to day tech, than 30 years ago.....really, it has been down hill, with the exception to some things....battery tech being one.
If AI is all super duper powerful, how come they do not direct the AI to make ultra efficient solar panels ending the worlds power "issue".
Some say "Because we have not figured it out yet".......well with the help of super computers that can make billion of decisions per minute, and access to all of the worlds data......coming up with something as simple as a 100% efficient solar panel should not be that far away. Or cure cancer. Or, make a car that gets 100 miles per gallon?
Maybe. Such a building would be a very easy target for a near peer to destroy. To bank most capabilities on such a building would be foolish IMO.
More likely the data to be used against us, than a military adversary
Also AI can't do anything without a physical experiment in a lab to back things up. A lot of companies are already doing that though, they have things called "digital twins" and use simulation from AI to tell research what to test in real world then repeat again.
Most data centers users are on redundant setup and have disaster recovery plan. You probably have to blow up more than 2/3 of the worlds' data center before you can wipe out the large providers instead of slowing them down. Now when you have space data center then you have to wipe out the orbits too.
Most of the problems in the world is due to economic feasibility rather than technological feasibility.