With a contentious midterm election less than a week away, one cause has unified voters and candidates across the political spectrum: slowing down AI.
asserted
cause → unify → AI
A friend of mine who directs tens of millions of dollars in midterm election spending says he's never seen voter salience of an issue—AI, broadly—increase so quickly.
asserted
salience → direct → issue
In July, Gallup found that 39 percent of Americans thought AI does more harm than good, up from 31 percent in 2025.
asserted
AI → find → 2025
Only 9 percent of Americans, meanwhile, thought the technology did more good than harm (down from 12 percent last year).
asserted
technology → think → percent
Public sentiment has since soured even more.
asserted
sentiment → sour → ?
An October ABC News/Washington Post/Ipsos poll found that 57 percent of Americans think AI is bad for society, while a staggering 66 percent said they opposed new data center construction in the United States.
asserted
they → find → States
Politicians on the left and right have seized on this political sentiment and offered their own data center stiff-arm policy du jour.
asserted
Politicians → seize → policy
However, even a full e-brake on data center construction would do little to stop AI progress.
asserted
brake → do → progress
In August, Arvind Narayanan, director of the Princeton Center for Information Technology Policy and co-author of AI Snake Oil: What Artificial Intelligence Can Do, What It Can't, and How to Tell the Difference, claimed that a data center moratorium would basically not slow AI development at all.
asserted
moratorium → do → development
Using a back-of-the-envelope calculation, he concluded (under some reasonable assumptions) that a one-year data center moratorium in a "typical U.S. state" would slow down AI progress nationally by only five to 10 hours.
asserted
moratorium → use → hours
Assuming that at least half of data center developers would avoid a moratorium by building their facilities elsewhere, a one-year freeze on construction in larger states like New York would set AI in America back by less than one day, according to Narayanan.
uncertain
freeze → assume → Narayanan
My own modeling of this question reaches essentially the same conclusion as Narayanan's, but rather than assuming computing power will continue growing at its current rate, it treats future growth as dependent on how much computing capacity gets built.
asserted
capacity → reach → growth
I considered three theoretical policy regimes: a full stop on new data center construction and on replacement of depreciated graphics processing units (GPUs); a full stop on new data center construction only; and no restrictions whatsoever.
asserted
I → consider → construction
The point is that, as a federal policy, standing athwart data centers yelling "Stop!" doesn't accomplish the outcomes that moratorium-ites claim to want.
uncertain
ites → stand → that
And, while some local policies could slightly impede center growth, the only way to hamper the progress of frontier models would be to enact an extremely draconian policy regime—one that goes beyond moratoriums for new data centers and prohibits GPU upgrades in existing facilities—which no one is actually proposing or close to implementing.
uncertain
one → impede → which
What does this all tell us about where advocates of any stripe should put their efforts?
asserted
advocates → tell → efforts
For those in favor of accelerating AI, on the one hand, it provides comfort in that data center construction, even in the extreme, is not a binding constraint to AI advancement.
asserted
construction → accelerate → advancement
But on the other hand, the political energy currently concentrated on mobilizing the backlash won't stay focused there forever; an issue where more than two-thirds of people are even tangentially opposed to your "side" is not one that easily disappears.
asserted
that → concentrate → side
For doomers, it means if the goal is a "pause" of some sort on the ability of AI technology to, say, replace you in your job, targeting data center construction isn't the best way to go about it.
asserted
targeting → mean → it
It may be a useful galvanizing node in the broader policy fight, but don't expect it to cause any meaningful change in the current trajectory—whatever you believe that to be.
uncertain
that → galvanize → trajectory
The ineffectiveness of data center moratoriums doesn't make the policy on its own more or less optimal.
asserted
policy → make → own
But if the data center protests are a proxy for broader AI-induced anxiety about large language models crowding out human pursuits, causing unemployment, or destroying the environment, protestors will need to find another leverage point.
asserted
protestors → induce → point
The Aporia
