According to Zhitong Finance APP, Bernstein recently released a research report on the data center supply chain of US hyperscale cloud providers.
The firm conducted channel interviews with data center architecture and procurement heads at five leading cloud providers, concluding that regulatory and local pushback stemming from the US midterm elections could become a new constraint on AI infrastructure; projects already under construction are largely unaffected, but a new wave of greenfield projects faces enormous uncertainty regarding site selection and pacing, with cooling and power equipment orders potentially experiencing a temporary gap from Q4 2026 to Q1 2027—though this represents order deferral rather than a complete disappearance of demand.
The firm noted that following the US midterm elections, risks of permitting pauses and rising local opposition to data centers have prompted the market to discuss whether political factors will become a new bottleneck for US data center construction.
Based on interview feedback, projects that have already obtained approvals, secured funding, and are under construction are basically guaranteed, with capacity plans expected to come online between 2027 and early 2028 essentially locked in, as hyperscale cloud providers will not abandon committed projects already under construction.
The firm estimates that over the next 12 to 18 months, earnings per share for related equipment companies will remain largely unaffected, with midterm election uncertainty mainly impacting subsequent brand-new greenfield projects rather than existing construction.
The real variable is concentrated in future new projects: hyperscale cloud providers have generally entered a wait-and-see cycle, awaiting policy clarity from the new administration, with particular focus on Texas policy direction, while Georgia, Ohio, and Pennsylvania are also included on the risk watch list.
Interviewed companies are simultaneously evaluating alternative construction regions, with North Dakota, New Mexico, Arizona, and Colorado becoming key candidates under consideration.
Bernstein emphasized that overall industry computing demand has not disappeared; companies are simply prioritizing regions that simultaneously satisfy policy acceptance, land availability, and power resources for new capacity, with the core contradiction shifting from "whether to build" to "where to build and when to start."
Driven by uncertainty in site selection and timing, orders for data center cooling and power supporting equipment may face a temporary gap.
The firm noted that before relevant legislation is finalized, equipment procurement intensity could weaken from Q4 2026 to Q1 2027.
Interviewees expressed divergent views: some executives said they would prudently place equipment orders without sufficient certainty to avoid inventory buildup where equipment is procured but cannot be deployed for construction; others would not significantly change their ordering pace but would seek more flexible contract terms from equipment suppliers, such as lowering cancellation penalties on orders.
Bernstein defined this phenomenon as demand deferral rather than demand destruction, and once the policy approval framework is clear and new capacity sites are finalized, deferred orders are expected to be released, with an order rebound anticipated around the second half of 2027.
To mitigate local resistance, "being a good community neighbor" has evolved from a bonus option to a standard operating playbook for data center development.
Hyperscale cloud providers are increasing local operational investment, including spending on local schools, workforce training, and public infrastructure, with some projects also considering bearing a portion of utility costs, while proactively disclosing water and electricity usage to communities.
These measures were already initiated before the elections and will become more routine going forward.
The firm believes such investments will not fundamentally undermine project economic return models, but companies still need to continuously assess how much supporting cost is actually required.
Beyond self-building, the importance of colocation data centers and neocloud providers has risen significantly.
Interview feedback shows that the computing share of third-party colocation and neocloud providers has increased from 30-40% two to three years ago to over 50%.
In an environment where greenfield project approvals are becoming more difficult, the value of ready-made computing assets that have already secured power supply and obtained construction permits has further increased.
Many cloud providers are partnering with service providers like Crusoe to supplement self-built capacity gaps through mature third-party computing power; the industry widely anticipates that ready colocation and neocloud resources will be rapidly rented out to hedge against future capacity risks.
Summarizing all five expert interviews, Bernstein concluded that existing under-construction projects have a sufficient safety cushion, but uncertainty surrounding next-generation greenfield expansion plans continues to accumulate.
Short-term equipment orders will face pressure, but the overall direction of AI computing infrastructure buildout has not reversed; the industry is mostly adjusting deployment locations and timing.
On the investment side, the firm differentiated strategies for investors with different holding periods.
For medium- to long-term investors who can withstand volatility of two to three quarters or more, the institution maintains its existing preference for the data center sector, with current valuations offering attractive earnings per share growth; while investors with shorter holding periods should exercise moderate caution, as the market has not yet fully priced in the potential impact of policy and approval delays.
The firm specifically noted that caution does not equal bearishness, and policy evolution will bring two-way stock price volatility risks.
Comments