The AI frontier is currently limited by physical power requirements.

The ability of operators to bring new capacity online to service compute is bound by a variety of regulatory and physical/market constraints.

Do folks not understand this?