The clock is the constraint
Three hyperscalers changed the metric they lead with. Companies do that when the old number stops describing what is actually stopping them.
A metric change is a disclosure
Capex answered who could afford to build. It says nothing about who can get a building energised, and that is now the question that decides who deploys. When the leading number changes, the binding constraint has already changed and the reporting is catching up.
Announced spending that cannot be energised is not capacity. It is a commitment.
Which means anyone who has been reading capex as a proxy for compute has been reading a number one or two steps removed from the thing they wanted to measure.
What time-to-energy actually contains
Interconnection queues. Substation lead times. Transformer availability. Grid capacity planned decades ago against different assumptions. None of it responds to money on the timescale a quarterly call cares about, which is precisely why it makes a good constraint and a bad slide.
The buildings tell you the same thing
A four-building campus in Wharton County and a five-gigawatt target in northeast Louisiana are sited for power, not for proximity to anyone.
That is an inversion. The previous generation of data centres went where the users and the fibre were, because latency mattered. Training does not care about milliseconds; it cares whether the building can draw continuously. So the map got redrawn around generation.
Five gigawatts is not a tenant on a grid. It is a reason to build one.
What this asks of you
If you are planning compute more than a year out, price delivery schedules and power, not accelerators. Silicon is the fastest-responding part of this supply chain. Everything around it responds to permits, and permits do not accelerate for anyone.
And watch who ends up holding equity in generation. A supplier with stakes in the electricity its customers need occupies a position no regulator has yet decided is interesting.
Data Center Knowledge — New Data Center Developments: August 2026 → · 24/7 Wall St. — 3 AI Infrastructure Stocks Powering the Data Center Boom in August →