Don’t forget their absurd power requirements - their datacenter costs must be astronomical with GPT5 using 8x the compute of GPT4, check gearscouts.com to see what efficient power delivery actually looks like vs the inefficient monstrosity they’ve built.