The main problem is a lack of revenue; most AI Data Centers have a negative ROI. This is like the dot-coms in 1999-2002. They couldn't generate income, and their loans were due.

As I see it, the various players will be able to self help via AI and analyze/find/fix these problems at AI speed, much like AI is is digging out bugs in code (bug reports are huge, but are being fixed/tested by AI at light speed = bug reports ebb in a few weeks) and AI fix/test will show this ebb as the code online is run through the wringer. While it is true that small malware coders/spreaders have easy entry and dedication, they are fewer in numbers and resurces than giants like Google/Microsoft/Oracle/ et al and quite quickly their walware will be rooted out and extinguished. There are code efficiencies rising that use tokens more efficiently to make libraries whereby wheels are not invented twice and simply parsed from the flow and that code re-used. AI spontaneously?? invented this process??. Also a number of AI data centers have been cancelled as these efficiencies were made known = lower cost per token + coupled with lower useage should soften this crash into a plateau or at worst a slow decline. After all, these players know all this and can be expected to correct promptly.