[okdood64]

They used ML to get a ~40% cooling efficiency on their datacenters. It's mentioned in a talk by Cassie Kozyrkov.