U.S.-China Relations
NextFin News - Wall Street's AI infrastructure trade met a political wall on Wednesday, as a Democratic sweep of the 2026 midterm elections triggered a premarket selloff in data center stocks and forced investors to price in a new risk: that the fastest infrastructure buildout in American history is about to run into a much less permissive regulatory landscape. Shares of CoreWeave fell as much as 3.2% in premarket trading, while Applied Digital dropped 3% and Nebius slid 1.3%. The move came as Democrats captured majorities in both the Senate and the House and flipped traditionally Republican states including Texas and Kansas, setting up potential legislative friction for the remainder of President Trump's term.
The selloff was concentrated in the companies most directly exposed to building and operating AI data centers. CoreWeave, the specialist cloud provider whose valuation rests on continued rapid capacity expansion, led the decline. The weakness spread to Applied Digital and Nebius, two other neocloud operators, and to SpaceX, which sells data center capacity and is developing AI systems. Oracle also declined in premarket activity. The losses came after a bruising stretch for the sector: on Friday, the S&P 500 and the Nasdaq both posted their worst days since mid-June after Intel and Amazon, two key suppliers to AI companies, reported mixed earnings results, and on Monday the Nasdaq 100 entered a correction, a 10% drop from its most recent high, where it remained early Wednesday.
The political shift injected a federal regulatory dimension on top of structural headwinds that were already building. In the second quarter of 2026 alone, at least 45 data center projects worth $68 billion were disrupted by local opposition, according to research group Data Center Watch. That followed an even more disruptive first quarter, when 75 projects worth roughly $130 billion were blocked or delayed. Communities imposed development moratoriums preemptively, often before developers had filed permit applications, and rules governing data center location and resource consumption were implemented in 30 state legislatures. Data Center Watch counts 843 opposition groups across all U.S. states except Hawaii, and in July, 142 AI data center protests were staged across 42 states in a single weekend.
Tuesday's results gave that grassroots opposition a partisan vehicle. Democrats won gubernatorial races in Texas, Ohio and Georgia, states that together account for roughly a third of all U.S. AI data center spending and projects. Texas and Georgia alone hold a significant share of the national project pipeline, and Texas environmental regulators had already halted data center permits before the election. On September 21, Governor Greg Abbott directed the Texas Commission on Environmental Quality to pause the issuance of all permits related to data center projects until the Electric Reliability Council of Texas completes an audit of grid and water impacts, with a report expected by mid-December. The Texas pause also closed off on-site generation as an escape route for developers looking to avoid interconnection delays.
The Transmission Mechanism: From Ballot Box to Balance Sheet
The chain from election result to equity price runs through three channels: permitting, power and the cost of capital. Each channel operates on a different clock, which is why the market impact is more likely to unfold over quarters than days.
First, slower permitting. Democratic governors and state legislators in pipeline-heavy states have signaled support for stricter AI regulation and more searching local review. Every month a project sits in review is a month of deferred revenue on capital that has already been raised. For developers that have drawn down construction loans against a completion date, a six-month permit delay is not a scheduling nuisance; it is an interest-cost event that compounds. The neocloud operators sit at the sharp end of this dynamic because their growth stories are written in megawatts added per quarter; miss one quarter of additions and the entire capacity trajectory shifts.
Second, power. Data centers are, above all, power projects. A less permissive environment for transmission lines, substations and on-site generation raises the cost of every megawatt-hour delivered to a rack. That cost does not vanish; it gets absorbed somewhere along the chain, either in developer margins or in the prices charged to hyperscaler tenants. The political salience of the issue is precisely about who pays: a July poll found 77% of respondents concerned about AI making electricity costs higher, and a Gallup survey found that only one-fifth of data center opponents cited cost implications such as higher utilities as their reason for opposing local projects. The gap between those two figures — 77% worried about cost, only 20% of opponents mentioning cost — is where the political risk lives. Politicians can campaign against data centers on environmental grounds; governing means deciding whether ratepayers or developers fund the grid upgrades that AI requires.
Third, the discount rate. Many of the most exposed names — CoreWeave, Applied Digital, Nebius — carry heavy debt loads and trade on growth expectations that assume uninterrupted expansion. Treasury yields declined slightly Tuesday night as traders guessed that a slowdown in the AI sector could reduce corporate bond issuance, but the relief is double-edged. Yields falling on growth fears is not the same as yields falling on disinflation; if the market begins to read slower data center buildout as a leading indicator of weaker AI-driven earnings, lower yields will not cushion the equity multiple. The timing is awkward for leveraged developers: the neocloud sector had only just begun to recover from a financing scare in late October, when concerns about CoreWeave's debt load dragged Nebius down 10% and Applied Digital and IREN down 9% each in a single session.
"Assuming a more restrictive data center development environment, this could create incrementally more negative sentiment for AI infrastructure exposed names," Evercore ISI analysts led by Amit Daryanani said in a note.
The asymmetry matters. Hyperscalers such as Microsoft, Amazon and Google can absorb delays and higher costs across diversified businesses with deep balance sheets. Combined 2026 AI capital expenditure guidance from those four companies alone is tracking toward roughly $700 billion to $725 billion, a figure that gives them leverage with regulators that smaller developers simply do not have. The neoclouds cannot absorb delays. Their valuations embed a straight-line extrapolation of 2025-2026 capacity additions; any kink in that line re-prices the equity.
How Much of the Pipeline Is Actually at Risk
Bernstein's pre-election analysis put numbers on the exposure, and the scale is large enough to move markets. Of approximately 430 gigawatts of data center projects in the U.S. pipeline, roughly 250 gigawatts — about 60% — sit in states where the gubernatorial front-runner favored greater restrictions on data center construction. Counting Public Utility Commissioner seats alongside the 36 gubernatorial races, 305 gigawatts, or 70% of the pipeline, faced election-related political risk. Bernstein ultimately identified 244 gigawatts, around 56% of the 433-gigawatt pipeline, as exposed once election probabilities were applied.
Yet Bernstein argued the risk was overstated, recommending purchases of Quanta Services, Eaton, Vertiv, Equinix, Digital Realty, Legence, nVent Electric, Johnson Controls and Trane Technologies. The firm's view is that the debate will focus more on making developers pay a greater share of electricity and grid infrastructure costs than on stopping projects altogether, and that the economic incentive to keep building remains strong. That split — 244 gigawatts exposed versus projects continuing anyway — is the crux of the investment debate.
The bullish case rests on a simple observation about incentives. A data center project that has broken ground represents construction jobs, future property tax revenue and, in many rural counties, the largest private investment in a generation. Governors who campaigned on restraint inherit those commitments the day they are sworn in. Bernstein's wager is that the gap between campaign rhetoric and governing reality narrows quickly once a delayed project means delayed jobs. The firm's preferred names reflect that judgment: equipment suppliers and colocation operators with existing, permitted assets rather than pure-play developers with greenfield exposure.
The Counter-Thesis: Opposition Is Broad but Shallow
The strongest argument against a durable slowdown is that opposition to data centers is broad but economically shallow. A Gallup poll conducted March 2-18, 2026, of 1,000 U.S. adults found 71% oppose building AI data centers in their local area, including 48% strongly opposed, with only about one-quarter in favor. Opposition to local data centers ran higher than opposition to local nuclear plants, 71% to 53%. Forty-six percent of respondents said they worry a great deal about the environmental impact of AI data centers, and another 24% worry a fair amount. The margin of sampling error for the poll is plus or minus 4 percentage points.
But that same poll reveals the weakness of the opposition's economic grounding: only one-fifth of respondents mentioned cost implications such as higher utilities. When the trade-off becomes jobs and tax revenue versus abstract environmental concerns, elected officials have a history of recalibrating. Bernstein's bet is exactly this: once governors confront the capital and employment tied to a delayed project, the rhetoric softens. History offers some support for that view. Nuclear power, once the most politically toxic form of generation in the United States, has seen a bipartisan rehabilitation as grid planners confront the physics of firm, carbon-free electricity; data centers could follow a similar path from pariah to necessary infrastructure if the economic stakes become unavoidable.
JPMorgan Chase strategists led by Dubravko Lakos-Bujas struck a sterner tone, warning that a Democratic Congress could bring "possible new data-center restrictions, and higher taxes on corporations and top earners," which would be "all potential headwinds for tech." The firm also flagged investigation risk for CEOs with close ties to President Trump and noted that healthcare and clean energy could emerge as relative beneficiaries of a Democratic-controlled Congress. JPMorgan has estimated that AI data center-related stocks now make up 45% of the S&P 500, which is why a sector-specific headwind transmits so quickly to the index. A Goldman Sachs view circulating before the election held that a Democratic Congress could take aim at data center tax breaks, adding a fiscal channel to the regulatory one.
Second-Order Effects: Rationing Supply Without Killing Demand
The market's first-order read is simple: fewer permits, slower buildout, lower revenues. The second-order effect is more interesting, and it explains why the selloff did not touch the entire infrastructure complex evenly. Slower U.S. capacity additions do not reduce demand for AI compute; they ration its supply. That rationing benefits whoever already owns operating capacity and the equipment that keeps it running.
Power and thermal-management names such as Vertiv, Eaton and Quanta Services sit on multi-year order backlogs that are largely insulated from next year's permitting cycle; their revenue is recognized as equipment ships and is installed, not as new sites are approved. Colocation operators with existing campuses, such as Equinix and Digital Realty, could see the value of their permitted, powered sites rise if new supply is constrained, because scarcity attaches to the asset that is already connected to the grid. The pain concentrates in developers whose assets exist primarily on paper and in the balance sheets of companies that must refinance construction before a single server is racked.
There is also a geographic and technological arbitrage. Capital denied in Texas or Georgia does not disappear; it migrates to friendlier jurisdictions or to alternative architectures. Elon Musk, whose SpaceX sells data center capacity, captured the logic on X: "Data centers in space looking pretty appealing right now." The orbital idea faces significant engineering and cost hurdles, but the instinct behind it is real: when one jurisdiction closes, capital looks for the next open door. States that keep their permitting regimes predictable — or that actively court displaced projects — stand to capture investment that their neighbors turn away.
The broader market context amplified Wednesday's move. The data center stocks were already sitting on a fragile base: the Nasdaq 100 correction that began Monday followed a week in which mixed earnings from Intel and Amazon reminded investors that the AI supply chain is not a straight line up. Wednesday's election-driven selloff therefore layered a policy risk on top of an earnings risk, and the two reinforce each other. If permitting delays push capacity additions into later quarters, the revenue recognition that chip and equipment suppliers were already struggling to forecast becomes that much harder to model.
Cyclical Shock or Structural Shift?
This is a structural shift wearing a cyclical mask. The election-day selloff is cyclical — a sentiment repricing that could fade if governors moderate or if hyperscaler capex guidance holds. But the underlying constraint is structural and predates Tuesday's vote.
The evidence for a regime change is plain. Local opposition is organized and durable: 843 groups across 49 states do not disband after one election. State legislatures have already codified new rules in 30 states. Hyperscalers have spent more than $1 trillion on data infrastructure since 2023, and the backlash arrived precisely because that spending became visible in local electricity bills, water withdrawals and transmission corridors. A constraint born of visible local cost does not mean-revert on its own; it gets priced into every future project. This is not a cycle in the traditional sense — there is no inventory glut to work off, no margin compression that automatically triggers a supply response. The constraint is political and physical at once, and neither dimension self-corrects quickly.
What is cyclical is the pace of the policy shock. The first 12 to 18 months after an election are when new administrations test their mandates; that is the window of maximum uncertainty. If moratoriums and new permitting rules materialize in Texas, Ohio and Georgia within the next two quarters, the structural read is confirmed. If they do not, the market will have over-discounted the risk. The distinction matters because it determines whether Wednesday's move is a buying opportunity in quality infrastructure names or the first leg of a longer de-rating in the most politically exposed parts of the trade.
What to Watch
The falsifying signal is specific: if Democratic governors in Texas, Ohio and Georgia do not announce new data center permitting restrictions, moratoriums, or cost-allocation rules within 180 days of taking office, the slowdown thesis loses its foundation. A second signal is hyperscaler capex guidance over the coming earnings season — if Microsoft, Amazon, Google and Meta reaffirm their buildout plans unchanged, the demand side remains intact regardless of permitting friction. A third is the ERCOT audit in Texas, due by mid-December: a report that clears data centers to proceed would undercut the permitting-delay narrative in the single most important state for the pipeline.
For investors, the exposure splits cleanly by time horizon. In the short term, sentiment favors caution in neocloud names whose valuations assume uninterrupted expansion, particularly those carrying heavy debt into a higher-cost-of-capital environment. Over the medium term, the beneficiaries are owners of permitted, powered capacity and the equipment suppliers with funded backlogs — the names Bernstein highlighted, whose revenues are tied to what gets built rather than whether it gets permitted. Over the long term, the question is whether the U.S. retains its share of AI infrastructure investment or cedes ground to jurisdictions with clearer rules of the road.
The 2026 midterms did not create the data center backlash; they gave it a governing majority. The AI boom was always going to collide with the communities hosting it. The only question now is whether the collision slows the boom, redirects it, or does both.
Explore more exclusive insights at nextfin.ai.
