A buddy who works at a data center warned me about AI server cooling costs last week
He told me a single rack of GPUs can pull 40 kilowatts and needs liquid cooling just to stay on, and I had zero idea it was that extreme. It made me rethink all those articles about AI cheapness, since nobody talks about the electric bill for keeping those chips alive. Has anyone else run into real world numbers on what this stuff costs to run at scale?