Run the Data Center Watch arithmetic first, because it is the bigger of the two numbers. Roughly $130 billion of projects blocked or delayed by local opposition in the first quarter, against roughly $156 billion disrupted across all of 2025 [8][9], is 83% of a full year's disruption booked in three months [11]; annualize it flat, with no further acceleration, and you get $520 billion, about 3.3 times last year's figure [12]. It is also the softer number. A blocked project is capital that cannot be deployed on the schedule it was underwritten on, while the up-to-$17-billion Meta agreed to this week [1] arrives with a ceiling and a payer.
Which is what makes the smaller number the useful one to everybody who is not Meta. Set $17 billion against the nearly $1 trillion valuations Anthropic and OpenAI are carrying into their offering preparations [10] and you get 1.7% [13] - not a rounding error, but a quantity that survives contact with an underwriting model. Compare that with what the sector had before: Pew has more than half of Americans more concerned than excited about AI in daily life, up from 37% in 2021 [4], which is a rise of at least 13 points, or at least 35% in relative terms [14]. Nobody reserves against 13 points.
The counter-thesis, and on a five-year view I think it is the stronger one, is that the settlement prices the addictiveness of consumer apps [1] and nothing else, while the thing the public actually fears about AI is job loss and the facilities going up nearby [2]. Sentiment does not convert into liability without a cause of action, and the reporting supplies no legal theory that carries a jury from a displaced worker to a payable claim. The read-across is an assumption, not a schedule.
Then there is the disagreement about what any of this costs to fix. Dario Amodei's own account is that this is a crisis of trust going back decades and that AI is the latest iteration of it [5]; Sarah Myers West of AI Now told CNBC that treating it as a public-sentiment problem would be misguided and that improving sentiment requires contending with what the public wants [3]. Those diagnoses imply different budgets. The first is answered by shipping things people like, which is cheap if you are believed; the second is answered by conceding something, which is not. And the credibility to run the cheap version is the scarce input here, given that more than 75% of 18- to 34-year-olds in CNBC's Generation Lab survey said they do not trust Amodei to act responsibly, with around 70% saying the same of Sam Altman and Mark Zuckerberg [6].
This is probably wrong, but I would hold the $130 billion rather than the $17 billion as the number that reshapes 2026 capital plans, because siting is where compute physically stops and a settlement is a transfer that has already happened. What would falsify it: a quarter of approvals that comes in near zero blocked, or a second nine-or-ten-figure settlement landing on an AI product rather than a social feed. Either one moves the liability from the land to the model, and the two are not budgeted by the same department.
