Perhaps their internal capacity is already tied up in other tasks, so while they have 1000 cores internally, they can't all be monopolized for 80 hours for a single task like the AWS machines can.