What if the market’s greatest comfort is also its least noticed weakness? The US data-centre boom has been priced as if electricity, land and capital will always meet the appetite of artificial intelligence. Voters just reminded investors that the real scarce resource is not compute, but permission. That is why the November 2026 US midterms matter more than a routine partisan swing. They exposed a simple truth that markets often avoid until the last minute: when a system grows fast enough, it attracts resistance from the people asked to host it.
The result is not a death blow to the buildout. It is something more interesting and, for disciplined investors, more useful: a delay. Democrats won control of both congressional chambers and flipped governorships in data-centre-heavy states. Goldman Sachs economists said those state and local wins would likely lead to short-term data centre moratoria, potentially affecting about 45% of planned US data centre capacity expansion. In a market built on speed, a pause is not a small thing. It is the handbrake on a machine that has been assuming a clear road.
The political lesson is as old as history. Rome learned that roads can bind an empire, but they also channel invasion. Industrial systems always look strongest just before they meet their first serious constraint. Railways needed right-of-way. Oil needed pipelines. Now AI needs substations, permits, water and local consent. Investors often talk about bottlenecks as if they are temporary inconveniences. In practice, bottlenecks are where leverage is revealed. They show which assumptions were built on engineering and which were built on hope.
Aidan Yao, senior investment strategist at Amundi Investment Institute, said: “In our view, the significance of these midterm elections is more like a referendum on AI than a referendum on (Donald) Trump.” That framing is more than political theater. It captures a deeper shift in public tolerance. AI has been sold as a clean abstraction, a cloud phenomenon that exists far from neighborhoods. But data centres are not clouds. They are boxes, transformers, cooling systems and transmission demands planted in real places, where people vote and notice the noise.
The exits and local contests made that visible. NBC News exit polling found 76% of voters would oppose a new data centre in their community, and 56% said AI and data centres were a crucial or major factor in their congressional vote. Redbud Consulting counted 47 local data-centre votes, with at least 38 favoring restrictions. Those are not just numbers. They are a warning that the social license for infrastructure can vanish faster than the capex plans that depend on it. Markets are full of models for demand. They remain worse at modeling consent.
Texas shows why. The state accounts for roughly 25% of planned US data centre capacity. Democrat Gina Hinojosa defeated incumbent Greg Abbott for governor after campaigning on an immediate pause. The policy change will not arrive until she takes office in January 2027, but markets do not wait for the swearing-in ceremony to reprice risk. They discount the future the moment it becomes legible. That is why the ballot matters well before the law changes. It changes the odds embedded in every project timeline, every financing assumption and every equipment order.
The useful paradox is that a slower buildout may be healthier than an unconstrained one. The FT’s core argument is that delays are a silver lining because they reduce the risk of trillions invested in soon-obsolete equipment, and deferred capex is cheaper on a discounted basis. That sounds counterintuitive only if one assumes every dollar of spending is equally wise. It is not. In a technological arms race, building too early can be as damaging as building too late. Capital misallocated into the wrong generation of hardware becomes a monument to enthusiasm, not productivity.
This is where investor psychology tends to fail. In boom phases, people confuse urgency with necessity. They fear being left behind more than they fear being wrong. Game theory explains the trap. If every participant believes competitors will keep building, each one must build too. But when the payoff curve depends on future efficiency gains, the first movers can end up owning obsolete capacity before the economic case has matured. That is not a moat. It is a race to a depreciating asset base.
Anthropic’s nearly $900bn in cumulative compute spending commitments, disclosed in an IPO filing that week, gives the story added weight. The number is vast enough to sound abstract, which is usually when investors should become careful. Large commitments are comforting because they imply conviction and scale. Yet they also reveal how far ahead expectations have run of present utility. If the market is financing a future that is still politically contested, the discount rate should rise, not fall. Deferred spending is not failure. It is the market rediscovering that time has a price.
That is why the election hit the stocks tied to the buildout. Core Scientific fell 11.6% on Wednesday after the election. Iren fell 12.5%. Applied Digital, Hut 8, Cipher Digital, GE Vernova, Constellation Energy and Vistra all slumped. These moves do not prove a permanent reversal. They do show how exposed the trade is to regulatory friction and local backlash. When a theme depends on endless acreage, endless power and endless tolerance, even a modest chill can expose the fragility hidden beneath the narrative.
The market should not be surprised. Industrial history is full of examples where the public embraced the final product but rejected the machinery required to produce it. People like the light switch, not the power plant. They enjoy the internet, not the substation. They want lower latency, but not a warehouse humming at the edge of town. The asymmetry matters. Benefits are diffuse; costs are concentrated. That is a classic recipe for political resistance. The error is assuming the public will quietly subsidize private optimism forever.
The deeper lesson is that a pause can improve capital discipline. If nearly half of planned US data centre capacity expansion is at risk of short-term moratoria, then at least some speculative projects will never be built, and that is not automatically a loss. It may prevent the kind of overhang that haunts every boom after the story changes. In engineering, a safety valve is not a defect. It is the feature that keeps pressure from splitting the vessel. Markets should treat local resistance the same way: as a mechanism that forces better sequencing.
There is also a second-order effect. If investors can no longer assume uninterrupted buildout, then the quality of projects matters more than the quantity. Sites with better power access, better economics and stronger local support should command a premium. Others should be repriced toward the reality that not every promised megawatt becomes an operating asset. That is a healthier market structure than one in which every plan is treated as destiny. The strongest systems are not those that never face resistance. They are those that absorb it without breaking.
Elon Musk’s response after the election captured the mood in a single line: “Data centres in space looking pretty appealing right now.” The joke works because it points to an uncomfortable truth. The more infrastructure collides with local politics, the more fantastical the alternatives begin to look. But the answer is not to flee to space. It is to admit that terrestrial expansion was never frictionless. Markets love to extrapolate smooth curves. Reality prefers gradients, obstacles and reversals.
That is why the smartest reading of this election is not bullish or bearish in the usual sense. It is corrective. It reminds investors that the AI boom is still subject to old laws: scarcity, backlash, obsolescence and the arithmetic of time. The ballot box has not killed the data-centre trade. It has made it more honest. And in markets, honesty is often expensive in the short run but cheaper than denial in the long run.
