October 13, 2026
Last updated: October 2026.
The AI backlash that matters for portfolios is unlikely to be a single regulatory "off switch." The more plausible form is a stack of frictions: on power, permits, labor politics, cyber risk and geopolitics. Each one raises the cost and slows the pace of the build-out, and they reinforce each other because they converge on the same two chokepoints.
The physical chokepoint is grid capacity and local permission to build.
The political chokepoint is who captures the gains and mitigates the risks.
That said, in this Scenario Lab, we’re excluding the framing deliberately to focus instead on what an economically and politically legible backlash looks like. This is the version that’s already observable and investable, and which should most concern portfolio and asset managers and other institutional investors.
Our view is that the backlash will surface first in bricks and megawatts rather than in rulebooks. Local opposition blocked or delayed $130 billion of planned US data-center construction across 75 projects in the first quarter of 2026 alone, the same amount as the whole of 2025. A further 45 projects worth $68 billion were blocked or delayed in the second quarter, according to Data Center Watch data reported by Bloomberg. In July, New York became the first US state to impose a statewide moratorium on new hyperscale data centers. The executive order pauses 39 pending applications for up to a year.
• Power bills. By May 2026, more than half of US registered voters blamed new data centers "a lot" for rising power costs, up from 28% in August 2025, and seven in ten said they would oppose one near their home, up from four in ten. The macro impact of the Iran War on energy costs more generally is further stoking this anger.
• Jobs. In mid-2026, Gallup found 79% of Americans expected AI to reduce jobs over the next decade, up from 73%. Pew followed a month later and said that, among 18-to-29-year-olds, 73% expected fewer jobs, up from 61% in 2024.
The counterintuitive part is that the conventional jurisdictional map is close to inverted at the physical layer. Convention wisdom has Europe regulating and the US and China running a kind of "wild west." This is incorrect. Each of these superstates, in its own way, practices a form of central planning when it comes to AI.
• In 2026, the EU actually loosened its hardest rules. The Digital Omnibus on AI entered into force in July and pushed the AI Act's high-risk obligations from August 2026 to 2 December 2027 for standalone systems and 2 August 2028 for embedded ones.
• The US is permissive at the federal level. It is also where the most capital has actually been stopped: through zoning boards, state utility commissions and governors' executive orders.
• China is not permissive at all. It regulates content tightly while directing mass adoption. With spare generating capacity, it is the jurisdiction least exposed to an energy-driven backlash.
• Spending. In September 2026, Gartner forecast worldwide AI spending of $2.67 trillion in 2026, up 49.5% year-on-year, with infrastructure taking 56% of it.
• Electricity. The IEA's baseline has data-center electricity use rising from about 415 TWh in 2024 to about 945 TWh by 2030. The US accounted for 45% of 2024 consumption and China 25%.
• A backlash that shaves even a few percentage points off the pace of that build-out, orshifts a meaningful share of its cost onto developers, reprices a very large and very concentrated pool of equity value generating capacity, it's the country least exposed to an energy-driven backlash.
Data-center projects are being blocked. In the first quarter of 2026 alone, local opposition blocked or delayed about $130 billion of planned US data-center construction, as much as in all of 2025. In July 2026 New York became the first US state to impose a statewide moratorium on new hyperscale data centers.
Power prices and grid rules are shifting. The largest US grid operator's capacity auctions hit their price cap and still fell short of reliability targets, with the jump in forecast demand blamed almost entirely on data centers. Regulators, starting with Virginia, have moved to make data centers pay directly for the grid upgrades they trigger. In Europe, Spain tied grid access to strict renewable-energy conditions, Denmark paused new connections, and Ireland allowed new data centers only if they bring their own power.
AI is being named in layoffs. Through much of 2026, AI was the most-cited single reason for announced US job cuts, and a string of well-known companies framed cuts around AI. Total layoffs, however, were well down on the year before.
The US–China fence goes up from both sides. Washington keeps its most advanced chips off-limits to China and has added major Chinese tech firms to its military-company list. Beijing has blocked imports of US AI chips and ordered state-funded data centers to use domestic ones. Trade truces between the two have been extended only in short, rolling increments.
Hostile states are using AI tools. Russian, Chinese, Iranian and North Korean groups have used AI for espionage, fraud and influence campaigns. Security researchers say that so far AI mostly makes existing attackers more productive rather than creating dramatic new capabilities.
The first-job squeeze. National labor data doesn't show an AI jobs shock. The narrower picture does. By mid-2026, workers aged 22–25 in the jobs most exposed to AI had employment about 19% below peers in less-exposed roles, and that gap was widening. It shows up as fewer entry-level hires, not layoffs, so it never appears in the headline jobs number. It is exactly the kind of grievance that builds political pressure.
Public mood turns into policy ideas. Around eight in ten Americans expect AI to reduce jobs, and young adults have soured fastest. Majorities in most countries surveyed expect net job losses. Proposals have followed: mandatory reporting of AI-related layoffs, public "AI dividend" or equity-sharing schemes, and even a proposed one-time 50% equity tax on large AI companies. None is close to law, but they show where the debate is heading.
A cost backlash inside companies. Chief AI Officer roles are spreading as companies try to govern AI use. Some large firms have reportedly burned through annual AI budgets within months, and each new frontier model tends to cost more per use than the last. The labor savings used to justify cuts are partly eaten by compute bills. The political cost of the cuts remains.
Power bills feed opinion faster than supply can respond. By mid-2026, more than half of US voters blamed data centers "a lot" for rising power bills, and about seven in ten would oppose one near their home. New gas turbines are booked out to around 2031, so relief is slow and the frustration has time to compound.
The split moves to the model layer. Chinese open-weight AI models have overtaken US ones in global downloads and carry most of the traffic on at least one major model-routing platform. That caps what US AI labs can charge, whatever happens with chip exports. US labs have also accused Chinese firms of copying their models at industrial scale, which points toward tighter controls on who can access AI models at all.
Energy markets. Price caps are holding grid costs below where markets would otherwise clear. If demand keeps outrunning new generation, the pressure could harden into permanent moratoria and cost-allocation rules that raise the all-in cost of compute in the US and Europe. There's a real counterweight: much of the headline demand is speculative. Texas's queue of large-load connection requests is several times the size of its entire grid, and US reliability watchdogs have trimmed their near-term warnings.
Financial markets. A handful of technology stocks make up roughly a third of the S&P 500, so an infrastructure-and-politics problem can become an index-level one. A slower build-out, or policies that make developers carry more of the cost, would hit a very concentrated pool of value.
Labor markets. The mainstream forecast sees AI job losses as real but temporary. The bigger tail risk is a policy overshoot: disclosure mandates, AI-specific taxes or hiring rules driven by graduate distress and public anger, arriving before the broad data shows the disruption they are meant to fix.
Technology and data infrastructure. The US and China openly accuse each other of hacking critical systems, from telecoms to national time services. Some intrusion groups now run AI models inside the networks they've breached, so they no longer depend on a US provider's tools. Meanwhile, the world's most advanced chips are still overwhelmingly made in Taiwan.
Supply chains. Rare earths remain China's most direct lever over the physical AI stack. Beijing's sweeping export controls have been suspended rather than removed, so access is exposed every time a short-term truce comes up for renewal.
Politics. Opposition to data centers is bipartisan at the local level, which makes it hard to reverse in a single election. The most likely political outcome is not a big federal AI law but a steady spread of state-level moratoria and cost rules. These need no federal action and sit largely outside Washington's efforts to override state AI laws.
Want to stay ahead of such risks? Check out Caminos, Chantico Technology’s risk modeling tool for institutional investors and portfolio and asset managers. Chantico integrates real-time data analysis with intuitive tools like scenario-based risk modeling and conviction-building visualizations.
