Bernstein: New U.S. AI Data Center Orders May Hit a Gap in Q4 2026
nashnova research
Bernstein's channel checks with five hyperscaler procurement heads flag a rising risk: new-build data center orders could soften in Q4 2026 through Q1 2027 as permitting stalls post-midterms — but this is demand deferral, not demand destruction.
How did politics become a bottleneck for data centers?
Bernstein interviewed data center architecture and procurement leads at five major hyperscalers.
Core finding: after the US midterm elections, local permitting-pause risk has risen and community opposition has intensified. Political factors are now a new constraint on new-build projects.
This means → the bottleneck for data center expansion is no longer just power and chips — the approval process itself is now the chokepoint.
Are projects already underway affected?
Projects already in progress are largely unaffected. Capacity planned for commissioning between 2027 and early 2028 is locked in.
The real uncertainty sits with the next wave of greenfield projects — entirely new builds on undeveloped land.
In plain terms = buildings already going up won't stop, but the next plot hasn't been secured — everyone is waiting for the new administration's policy signals.
When does the order gap hit?
Procurement of cooling and power-infrastructure equipment could weaken in Q4 2026 through Q1 2027.
Interviewees split two ways: some executives will hold off on equipment orders until they have enough certainty, to avoid stockpiling gear with nowhere to install it; others will keep ordering but push suppliers for looser contract terms — such as lower cancellation penalties.
Bernstein expects no material EPS impact on related equipment companies over the next twelve to eighteen months.
What separates "demand deferral" from "demand destruction"?
Bernstein explicitly defines the current dynamic as "demand deferral," not "demand destruction."
This means → orders are not disappearing — they are being postponed. Once the permitting framework clears and sites are confirmed, deferred orders should release in a wave.
Bernstein expects an order rebound around H2 2027.
Where to build — the new core question?
High-risk states: Texas, Georgia, Ohio, Pennsylvania. Alternative regions: North Dakota, New Mexico, Arizona, Colorado.
The industry's core tension has shifted from "whether to build" to "where to build and when to break ground."
Third-party colocation and neocloud providers now account for over 50% of compute capacity, up from 30%-40% two to three years ago. This reflects a rapid repricing: facilities that already have power locked in and permits in hand are becoming significantly more valuable as greenfield approvals get harder.
What should investors do now?
Longer-horizon investors (comfortable with two to three quarters of volatility): Bernstein maintains its existing preference for the data center sector.
Shorter-horizon investors: exercise caution — the market has not yet fully priced in the potential impact of policy and permitting delays.
Put simply = if you can hold through the chop, hold; if you can't, don't add — whether the H2 2027 order rebound materializes on schedule is the key test for this call.
市场有风险,内容仅供研究参考,不构成投资建议。
