What the AI Industry Could Learn from the Three Mile Island Meltdown
By the time anyone in the control room figured out what was happening, a stuck valve in a Harrisburg, PA, reactor had melted half the core.
Thankfully, nobody died. And studies conducted in the years afterward found no detectable health effects on the public.
By almost every technical measure, Three Mile Island was a system that failed, creating a disaster, and then worked, by containing the disaster.
Public support for nuclear energy in America fell from 69% before the accident to 46% within months of it. Regulators froze the licensing of new reactors. Utilities canceled orders they had already placed. Construction on a new nuclear plant did not begin anywhere in the United States again for more than three decades.
I think about Three Mile Island a lot right now, because I believe artificial intelligence is a close parallel. And I do not think most investors have priced that in.
A Boom Built on Borrowed Patience
For three years, the AI trade has run on a simple assumption: chips get built, power gets found, and the buildout continues. My tech researcher and I spend most of our time worrying about the technical risks to that assumption. Does the industry run out of chips? Out of power? Out of money?
The risk keeping me up at night is that the country hosting most of this buildout no longer wants it in the neighborhood.
A recent Heatmap News survey asked a simple question: Would you support or oppose a data center being built near where you live? A year ago, Republicans answered with a net rating of positive 14%. Independents sat at negative 5%. Democrats at negative 8%. Practically a flat line, and nobody particularly cared.
Ask that same question today, and Republican support has collapsed to negative 43%. Independents have fallen to negative 65%. Democrats sit at negative 75%. The average has gone from roughly zero to somewhere around negative 60 in a single year.
That is not a data center story. That is a populist movement, and it is one of the only things left in America that Republicans and Democrats agree on.
I talk about this in-depth in my latest episode of Being Exponential With Luke Lango. Check it out at the video link below:
The Nine States That Decide Everything
Roughly 90% of the data centers under construction in America sit in just nine states: Texas, Virginia, Georgia, Illinois, Arizona, Ohio, Indiana, Louisiana, and North Carolina. Every other state’s legislation is background noise. Those nine states are the whole ballgame.
Right now, only one of them, Texas, has passed a statewide freeze. Governor Greg Abbott ordered every data center project in the interconnection process audited before it can move forward, and Texas happens to be the largest data center market in the country. That is a real, immediate cost.
The other eight states are doing something slower and, I would argue, more concerning. Virginia added a per-kilowatt-hour tax. Illinois stopped accepting new incentive applications while still approving projects. Arizona froze tax incentive qualifications through 2029. Ohio has more than a dozen municipalities running temporary pauses with no statewide policy behind them.
None of that stops the buildout today. It is a patchwork, not a wall, and hyperscalers know how to route around a patchwork. If Texas throws up friction, you build in Arizona. If Arizona throws up friction, you build in Wyoming.
The reason I do not treat this as an emergency for 2026 or 2027 is that federal support for data centers remains firmly intact, and President Trump has been openly dismissive of the anti-data center movement. As long as Washington keeps the door open, hyperscalers can absorb state-level friction and keep building.
That changes if the federal backdrop changes. And I do not see that happening before 2028, because the executive branch stays Republican regardless of what happens in the midterms this fall. A shift in the House or Senate does not touch this dynamic.
My honest answer: I believe this movement could meaningfully threaten the AI trade by the end of 2028, once a new administration with a different view on data centers takes office and the state-level dominoes that are wobbling today finally fall together.
I do not consider that inevitable.
Attitudes shift, media attention moves on, and politicians follow incentives that could easily point somewhere else by then. But if you put a gun to my head and asked whether this movement ends the AI rally within two years of that election, I would say yes.
Why the Backlash Is Not About Elon Musk
It is tempting to read this as an anti-Musk story. I think that misreads what is happening.
This movement pulls in the environmentally conscious voter worried about land and water use. It pulls in the cost-conscious voter watching electricity bills climb in data center towns.
It pulls in a broader anti-tech current that grew louder the moment AI labs started talking openly about automating jobs away.
Remove any single company or founder from the equation, and the coalition does not shrink. If anything, it becomes easier to hold together, because you stop blaming a person and start blaming a category.
Americans currently express the least support for data centers of any developed country surveyed, well behind South Korea, Germany, and other major economies.
That gap tells you this is not incidental. Consumer confidence in the U.S. recently hit a seven-month low, inflation continues to outpace income growth, and AI stocks have carried nearly the entire market for three years running while everything outside the AI blast radius has struggled.
When one group of companies wins decisively and everyone else loses, the losing side eventually organizes.
That is how populist movements form every single time.
Where the Money Goes If I Am Right
You could read all of that and conclude I am bearish on AI. I am not. I recommend the AI supercycle as the defining trade of this decade.
What I am doing is positioning for the specific fight the industry is now walking into: power.
Every state pushing back on data centers is pushing back on the same grievance: AI infrastructure strains the power grid and drives up electricity costs for everyone else.
The companies that solve that problem for hyperscalers, rather than fight the politics of it, sit in a different category entirely.
On-site power generation, the bring-your-own-power model that lets a data center supply its own electricity instead of straining the public grid, turns a political liability into a non-issue. The same logic extends to nuclear power, a sector written off for years that could be entering a genuine second act.
That is the trade I am building around. Not a bet against AI. A bet on the companies that let the AI boom keep growing even as the politics around it get louder.
The full episode of Being Exponential goes further than I could fit here.
We walk through the complete state-by-state legislative map, the specific companies positioned to benefit from the bring-your-own-power shift, the nuclear names I am watching most closely as that sector re-emerges, and the single ETF I recommend if you want one diversified way to stay long the AI supercycle while this political story plays out in the background.
Three Mile Island failed because the public stopped trusting it before the industry ever got the chance to prove otherwise. I do not think AI is broken, either. I think the industry has a limited window to make its case to the public before politics decides the case for it.
Watch the full episode to see exactly how I am positioning for it.
The post What the AI Industry Could Learn from the Three Mile Island Meltdown appeared first on InvestorPlace.
