AI jitters erupt after Democrats threaten data centre build-out
Investors are worried that Democratic gains in the US midterms could slow or stop data-centre construction. Democratic candidates who pledged to crack down on data-centre building won in Texas, Georgia and Ohio, and more than $3tn was wiped off US stocks last Thursday amid AI-market jitters. The build-out is huge: Statista expects Amazon, Microsoft, Alphabet and Meta to spend $760bn on capex in 2026, up from $413bn in 2025, and McKinsey puts global data-centre construction spending at $7tn in 2026. The 50 largest US electrical contractors made $80.4bn of revenue in 2025 (+40%). Examples: Crusoe is finishing eight buildings at the 1.2GW Stargate site in Abilene, Texas (backed by OpenAI, Oracle and SoftBank), where Rosendin Electric has a $3.4bn contract; Meta plans a 5GW, ~$250bn campus in Richland Parish, Louisiana; CoreWeave runs >1.5GW and targets 8GW by end-2030. Oracle had spent $77.8bn on equipment against $25.5bn on buildings by end-August 2026; equipment lasts 1–6 years, buildings about 40. Counterweights: power, permits and labour were already the main bottlenecks (UBS), CBRE says ready-to-use space hit a 20-year record low at end-2025, and most suppliers still report very strong growth. Caveats: the Democratic plans are pledges, not enacted moratoria or rules; no lawmaker or bill is named. The 2030 US capacity estimates (BloombergNEF 118GW, S&P Global 183GW) differ and measure different things.
Why it matters
This is the political risk to the AI build-out, now after the election rather than before it. Hyperscaler capex of $760bn this year flows to developers, neoclouds, contractors and power suppliers, many of them companies Dealroom tracks. If state-level crackdowns follow through, the bottleneck shifts from chips and power to permits and politics. That would favour players with sites already built or permitted.
Executive takeaways
1) After the midterms, the market is pricing political risk to data-centre construction: >$3tn came off US stocks last Thursday. 2) Demand still looks strong: Big Four capex is forecast at $760bn in 2026 vs $413bn in 2025, and ready-to-use capacity is at a 20-year low. 3) The threat is pledges from newly elected Democrats in Texas, Georgia and Ohio, not enacted law, so the near-term effect is on sentiment and permitting speed. 4) Suppliers are exposed: the top-50 US electrical contractors grew revenue 40% to $80.4bn. 5) Equipment is the bigger, shorter-lived cost (Oracle: $77.8bn equipment vs $25.5bn buildings; 1–6 years vs 40), so delays strand fast-depreciating capital.
What Financial Times may be missing
The piece names no lawmaker, bill or state measure, so we cannot tell how concrete the threat is or when it could bite. It does not separate the stock-market fall from other causes such as AI valuations, rates and earnings. It does not say which projects (Abilene, Richland Parish, the 651MW Shadowbox site in Georgia) sit in affected states or are already permitted and therefore safer. It leaves out how hyperscalers could respond (moving sites to friendlier states or abroad, or building their own power) and what this means for Europe as an alternative location. It also does not reconcile the very different 2030 capacity forecasts.