‘Capital alone no longer clears a site’: Morgan Stanley says data centers’ big money era is over

Morgan Stanley warns “capital alone no longer clears a site” for U.S. AI data centers, citing rising local political/community resistance as material development risk. It estimates a 38GW U.S. data-center power shortfall between 2026-2028, with some regions facing 5-7 year grid interconnection waits, while hyperscalers are forecast to spend ~$785B on capex this year (rising toward ~$1T by 2027) and have >$1.2T in data-center lease commitments. The report flags Texas audits of ~474GW pending interconnection requests (about 90% tied to data centers) and Virginia’s “cost causer pays” shift, arguing the largest projects may be hardest to approve due to non-linear opposition scaling with size.

Analysis

The immediate market impact is not that AI demand disappears; it is that the conversion rate from announced capacity to operating capacity gets worse. That matters most for the names whose valuation depends on a smooth ramp in AI infrastructure spend: every quarter of delay pushes out depreciation, lease-up, and revenue recognition while capex and financing costs keep flowing, which is a FCF and multiple problem for the most aggressive builders. By contrast, existing-site operators with smaller footprints and embedded power access should capture a larger share of constrained demand, so the relative winner is less "more AI" than "more already-powered real estate."

Second-order effects favor infrastructure that reduces time-to-power: modular generation, gas turbines, switchgear, and interconnect services should see a longer order book, while greenfield campus developers face a higher rejection rate and more expensive permitting optionality. For hyperscalers, the real risk is not demand destruction but a forced redesign of expansion plans toward phased builds, retrofit capacity, and distributed inference, which lowers scale efficiency and can compress the moat around the largest campuses. Dominion is a sleeper beneficiary if regulators push cost-causation through to large-load customers, because that reduces ratepayer backlash and makes new load a more bankable utility growth vector.

The contrarian view is that the selloff risk in AI infrastructure could be front-loaded while the structural implication is actually a geography shift, not a capacity collapse. Over 1-3 months, local hearings and state-level interconnection audits are the catalysts that can delay projects and create negative earnings revisions for the most capex-heavy names; over 6-18 months, the likely outcome is a slower but more distributed buildout rather than a broken demand curve. The thesis is falsified if federal or state permitting reform accelerates enough to compress time-to-power, or if hyperscalers keep capex guidance intact without an increase in project cancellations or deferred starts.

Key Decisions for Investors

• Long EQIX / short a basket of capex-heavy AI infrastructure names (AMZN, MSFT, GOOGL, META, ORCL) over 1-3 months. Rationale: relative scarcity value rises for powered, existing capacity; risk/reward improves if project delays keep widening the gap between announced spend and usable capacity. Falsify if hyperscaler build timelines remain on schedule and interconnect wait times begin to shorten materially.
• Buy CRWV puts or put spreads into strength for a 1-2 quarter view. CRWV is the cleanest expression of "time-to-power" risk because its equity story is most exposed to rapid capacity rollout; downside is meaningful if permitting friction forces deferrals. Cover if management shows pre-leased, already-powered capacity can absorb demand without new greenfield builds.
• Add D on pullbacks as a utility/regulated-infrastructure hedge. The cost-causer-pays framework can improve project economics by reducing socialized grid costs, but only if regulators keep passing incremental transmission expense to large loads. Exit if Virginia or Texas reverses course and shifts costs back onto the broader rate base.
• No-trade alert on MS and MCO for now. The article is more about physical bottlenecks than financing availability; any benefit to advisory/ratings revenue is too indirect to underwrite a directional position without evidence of spread widening, covenant stress, or project cancellations.