My theory is that most of their outages are compute and scale related. IE. A few GPU racks blows out and some customers see errors. They don't have any redundant compute as backup because supply is constrained right now. They're willing to lower reliability to maximize revenue.