Prediction markets called the midterms — and the data center moratorium wave — almost perfectly

Prediction Markets Called the Midterms—and the Data Center Wave—Almost Perfectly

Prediction markets delivered on their promise. Call Sheet hit 95% accuracy when polls opened and maintained 81% accuracy three months before election day, compared to Vote Hub's 77% average. Early on, when odds sat closer to 60-40 than 80-20, that predictive edge feels significant. The real test now is whether markets will continue to inform high-stakes decisions or, like recent Oscar broadcasts, simply spoil the outcome before the announcement.

One major surprise slipped through the net: Texas governor. No serious observer called for a Democratic win. When prediction markets flipped in the final hours, that late reversal didn't help anyone positioned months in advance. But the broader story was not Texas. It was the 43 statewide data center ballot initiatives that failed.

The data center moratorium sweep is real

Forty-three to one. That is how decisively voters rejected data center expansion across the country. North Dakota was the sole exception. This is not a close call or mixed signal—it is a near-total political rebuke of AI infrastructure buildout in the continental U.S.

The consequences ripple through multiple layers. Neo Cloud operators face immediate capital constraints. Many locked in long-term compute contracts at fixed rates and cannot capitalize on spot market premiums. Now those same companies cannot secure permits to build new capacity where they might sell compute at higher rates on the spot market. Talented teams exist. Financing does not. Political permission definitely does not.

Hyperscalers emerge as structural winners

Amazon, Google, and Microsoft have already-permitted facilities coming online. As smaller players face permitting freezes, these companies gain relative market share in AI compute. Investors may view this as a net positive even for the large cap stocks themselves. Political gridlock becomes an excuse to moderate spending and return capital instead of tapping debt markets for uncertain buildout timelines. Dividend upside over deployment risk.

Nvidia, by contrast, faces a mixed picture. Fewer data centers mean fewer chips shipped in the near term. But in a world where power becomes the binding constraint—where the total available electricity to facilities is fixed and growing capacity is impossible—there is incentive to swap older chips for the latest generation to maximize tokens per watt. That efficiency play could offset near-term volume losses over time.

The geographic and geopolitical reset

Permitting did not disappear entirely. It relocated. States without moratorium legislation on the ballot—those that simply never took the anti-data-center position—remain open for business. Infrastructure and energy access vary wildly, but receptive jurisdictions exist outside California and other high-resistance areas. Shopify's Harley Finkelstein was direct: Canadian cold, clean hydropower, open land, and union labor frameworks make Canada a "generational" opportunity. The capital that would have gone to restricted U.S. markets will look north.

Elon Musk's quip about space-based data centers landed partly as joke, partly as signal. Panthalassa's ocean-based tidal power approach and SpaceX's Starship program operate on longer timelines, but neither faces the near-term political headwinds of terrestrial permitting. If Starship catches itself on the next launch (expected to attempt a booster catch with "Mechazilla," the chopstick mechanism), space-based compute goes from science fiction to engineering timeline by 2027-2028.

The China angle sharpens

The inverse scenario now materializes abroad. China has abundant power, no permitting constraints, and a decade of edge-chip investment underway. The recurring anxiety in AI policy—that safety-motivated slowdowns hand dominance to Beijing—was theoretical when U.S. buildout seemed unlimited. It stops being theoretical if 43 states erect barriers and compute migration accelerates eastward. How policymakers react to this dynamic, and whether U.S. buildout actually slows or simply relocates, will define the next phase of the AI race.

For now, prediction markets nailed the political outcome. The harder question is whether markets can price what comes next: a fragmented North American compute strategy, accelerated internationalization, and a new scarcity premium on pre-moratorium-era permitted capacity.