AI in action · Markets
Barclays: The AI Build-Out Is Now a 'War of Attrition'
Local opposition will make the AI data-centre build-out pricier rather than stop it, Barclays' head of US credit strategy said after the midterm sell-off.
A day after the Nasdaq's worst session since April 2025, the question on trading desks on Friday, November 6, was whether the midterm elections had changed the economics of building data centres. Dominique Toublan, head of US credit strategy at Barclays, gave CNBC a precise answer: "AI pushback is bipartisan but local, making the buildout pricier rather than stalling it." His conclusion was about who wins: "The AI race shifts from a prisoner's dilemma to a war of attrition: scarce power favors the biggest balance sheets."
The local resistance is measurable. Between January 2024 and May 2026, at least 46 AI data-centre projects in 20 US states, representing more than $170 billion of announced investment, were blocked, withdrawn or stalled after community opposition, according to the International Data Center Authority, a think tank cited by CNBC. Dozens of local ballot measures in 10 states that would restrict data-centre construction were on the ballot on Election Day.
Thursday's market showed how investors priced that risk. Nvidia fell 9.3%, Broadcom 10.4%, AMD 17.3% and Micron 17.5%, while the neocloud operators that rent out AI computing fell hardest among infrastructure names: Nebius lost 15%, IREN 17% and CoreWeave 10%. Kyle Rodda of Capital.com said the political risk arrived when the AI trade was "already on shakier foundations" after an earnings season he called an overall disappointment.
Not everyone sees a slowdown in the near term. Goldman Sachs estimates AI-related capital spending could total about $7.6 trillion from 2026 to 2031 across computing, data centres and power, and much of the capacity due online in 2027 is already under construction. Dan Ives of Yorkville Ives & Co., one of the most bullish voices on the trade, now expects Big Tech's capex growth to peak in 2027 rather than in 2028 or 2029, and asked whether the elections would slow "the planned 800 to 1,000 datacenters across the US".
Toublan's framing has a clear implication for credit and equity investors alike. If every new site costs more and takes longer, the companies that can fund delays from their own cash flow, the largest hyperscalers, gain ground on those that depend on borrowing, such as the neoclouds whose bonds already yield in double digits. The build-out continues; the dispersion between its winners and its weaker participants is what the midterms widened.
Sources: CNBC — AI stocks hammered by midterm shock as data center buildout becomes a 'war of attrition' (Nov 6, 2026)