Yeah, this is just fluff ignoring prior art. ML and all sorts of other fancy modeling is nothing new for power grids.

The optimization for their datacenters used a 3 layer neural network with 6 neurons. I wonder if this is much more complicated than that...

You can still do cutting edge work with 3 small layers, for example most of the advances in robotic control (using neural networks) use networks with similar sizes.

Based on the datacenter cooling work, you'd do better to set up a Google Patents alert instead.