1. The "GPUs" in these datacenters won't be worth much. (Because they can't easily be used for anything else, not even gaming.)
2. Employers might like a cyber-employee that (versus a human) is 0.5x as good at 0.1x the price, but they won't accept 0.5x at 1.2x the price.
3. They attracted customers with ~1% costs and everyone was super-eager, but recently going towards ~5% everybody recoiled and started rationing.
My only critique is that if we're asking "what rises after the crash and bankruptcy" sense, then #3 shouldn't be using inference+R&D costs, but closer to inference alone because the original company and its R&D debts might be gone.