> The liquid-cooled AI racks being used today for training and inference workloads almost certainly have far higher power output than that. 75kW is a sane "def

> The liquid-cooled AI racks being used today for training and inference workloads almost certainly have far higher power output than that. 75kW is a sane "default baseline" and you can find plenty of deployments at 130kW. There's talk of pushing to 240kW and beyond...