AWS SA here, but speculating, so take it with a grain of salt.

A lot of customers -- especially the larger ones -- actually share their architecture and have in depth conversations about what they are doing and pain points, etc, with their account teams.

My guess is that they knew because their customers told them (or complained?) that they had fleets of EC2 instances just waiting

I would guess host metrics show they're idle, and talking to say 10 such customers showed obvious patterns. Is it harder than that?

Wouldn't AWS have metrics about physical and virtual CPU utilization from the hypervisor?