But separating YouTube and GCP into separate companies would presumably cause GCP to expose pricing / scheduling modes that would enable more compute to take ad

But separating YouTube and GCP into separate companies would presumably cause GCP to expose pricing / scheduling modes that would enable more compute to take advantage of regional downtimes. I'm sure there's plenty of ML training jobs that are just aching for regionally idle compute time.