Data Center Capex – The Constraint Trade, from Carder Dimitroff
The US data center buildout looks increasingly supply-constrained, not because of capital, but because of infrastructure. Announced projects far exceed what can realistically be delivered on current timelines. New capacity requires a rare intersection of:
Proximity to internet hubs.
Reliable, large-scale power (plus generators/transformers with multi-year lead times).
Viable cooling and water.
State/local support in communities that are already pushing back on load growth and land use.
Access to critical, long-lead-time equipment currently has multi-year queues.
Finding sites that clear all five hurdles is getting harder in the US and across most Western markets. This is solvable, but on a 5–10 year grid and permitting timelines, not on 12–18 month tech timelines.
Near-term, not many operational changes, but the narrative probably does. When the market internalizes that a meaningful slice of “announced” capacity is delayed or won’t get built, I’d expect:
Incumbent capacity with grid access to gain pricing power.
Late-stage greenfield and some capex-heavy stories to be repriced.
Utilities, grid upgrade plays, and on-site power to screen better than marginal new-build data center names.
High-level takeaway: AI demand is real; the binding constraint is infrastructure. The gap between slide-deck capacity and physically deliverable capacity is likely where the mispricing lies.