The Outsider Perspective Issue 493 | The Beltway Outsiders

If you’d like to read this issue on my website, click here! If you’d like to sign-up and receive this in your inbox each week, click here! Read past issues here.

Good Friday Morning! The World Cup is here, the planet is pouring into the United States for the 2026 tournament, and the early returns are a delight. For weeks the usual voices warned foreign visitors that the American South was a frightening place to steer clear of. Then a German fan named Freddy started live-tweeting his road trip through Georgia, Tennessee, and Alabama, and fell head over heels.

His first Waffle House landed at 1 a.m.: “Great food, great prices, and friendly staff. 10/10, we will be coming back.” Buc-ee’s got three words: “The holy land.” A Walmart snack aisle nearly undid him: “How am I supposed to choose from all of this?” And a packed college stadium for Argentina and Iceland was, in his words, “the most ‘The European mind can’t comprehend this’ moment of my life.”

J.J. Watt offered to host him in Houston. As one roundup put it, the English fans are online moaning and making dark jokes while a German tourist is having the best week of his life. America is not a hard sell. You just have to show up.

This week, I’m writing about the fight over AI data centers, and why the people trying to stop them are wrong on the water, wrong on the jobs, and wrong about the country they think they’re protecting.

Quick Hits:

Buckle up for the end of the Supreme Court term. June is when the big ones land, and the Court is saving the heaviest for last. It has roughly two dozen opinions still to hand down before it breaks for summer. Among the ones still waiting to drop: birthright citizenship in Trump v. Barbara, the biggest Fourteenth Amendment case in a generation; the transgender-sports bans in West Virginia v. B.P.J. and Little v. Hecox; and Trump v. Slaughter, which could finally undo the 1935 precedent that lets independent agencies sit beyond a president’s control. NPR has a rundown of the rest. Expect the fireworks over the next two weeks.

Where you can find me this week

Please subscribe, rate, and review The Horse Race on YouTube — the reviews help listeners, and readers like you find me. Make sure to sign up for the Conservative Institute’s daily newsletter and The American Almanac.

Socialism Doesn’t Work. Mamdani Just Proved It Again in Six Months. – Conservative Institute

The SPLC’s Own CEO Couldn’t Defend the Hate Map. Neither Can the Press. – Conservative Institute

Graham Platner Told the Conservative One Nazi-Tattoo Story and the Leftist Another. Both Were Lies. – Conservative Institute

The Scarcity Lobby Comes for the Data Center

The fight over AI data centers is sold as a fight about water, jobs, and the electric grid. It is really a fight about whether a free country still believes in building things. Our adversaries have already picked a side.

This spring a member of Congress held up two mason jars of brown water in a hearing room and blamed a data center. The same season, China broke ground on new coal plants at the fastest pace in a decade, poured roughly half a trillion dollars into electricity, and kept building data centers as fast as it could pour concrete. Its state media ran articles telling American families that data centers would wreck their water and spike their power bills. OpenAI caught and shut down a Chinese influence operation seeding that exact message online.

One country is holding hearings with jars of water. The other is building the future, and would very much like us to keep arguing.

I want to make the case the jars are meant to stop. We should build more data centers, not fewer. More compute. More power to run it. The objections you are hearing fall into two piles, and both lead to the same place. Most of them are false, and the few that are true are arguments for building smarter, not for building less. Underneath the loudest opposition is something older and worse than a water-quality complaint. It is a worldview that treats human flourishing as the problem. It has been wrong about everything for sixty years, and it is wrong again now.

More compute means cheaper intelligence, and cheaper intelligence reaches everyone

The point of a data center is compute, and more compute drives down the cost of intelligence. Every new facility, every new chip, every new megawatt makes the next model cheaper to train and cheaper to run. The cost of intelligence is the difference between AI that only a handful of trillion-dollar labs can afford and AI that a small business, a student, a solo developer, or a person who cannot pay for a subscription can use.

Two things drive the cost of intelligence down. One is raw compute, which is what data centers provide. The other is open models, the kind whose weights are published for anyone to download and run. Cheap compute plus open models is how a kid with a laptop ends up holding what only a national research lab held a few years ago. The scarcity of intelligence is what keeps power concentrated in a few hands. Abundance is what breaks it open.

The technology companies talk about this part badly, so here it is plainly and from the right. Julian Simon, the economist who spent his career demolishing doom, called energy the master resource, because energy is what lets us turn one thing into another. Cheap energy and cheap compute are the master resources of this century. A free people armed with both will solve problems faster than any people in history. The question is whether we let them build the engines, or whether we talk ourselves out of it while our rivals build theirs.

The water argument collapses the moment you open the USGS tables

The water argument has a face now. In a May hearing, Rep. Alexandria Ocasio-Cortez held up two jars and said one held clean water and the other held the drinking water in a Georgia county after a Meta data center was built. “The only difference between the clean water and this was that data center,” she said. “Neither one of these things are drinkable.”

It was a good piece of theater. It was also false, and the government’s own witness said so at the same table. The Environmental Protection Agency’s top water official testified she was not aware of any data center causing a drinking-water quality problem, and that the agency’s concern is water availability, not contamination. Cooling a server does not poison a well. What Ocasio-Cortez held up as proof of a poisoned town was a claim about construction dust and blasting at a building site, and the federal official whose entire job is the nation’s drinking water would not confirm a word of it. The socialist from New York was simply wrong.

The scale is where the water argument dies for good. The U.S. Geological Survey keeps the books on American water. American farms irrigate with 118 billion gallons a day. Every data center in the country uses roughly 17 billion gallons of cooling water in a year. One day of irrigation is more water than a year of data-center cooling, many times over. The California Water Impact Network puts the state’s almond orchards alone at more than a trillion gallons a year, dozens of times the cooling water of every data center combined. Nobody is holding up a jar of almond milk in a hearing room.

The activists also leave out where the water actually goes. Most of what gets counted as data-center water is not the cooling tower at all but the power plant upstream. By the Energy Information Administration’s figures, a coal plant burns through almost nineteen thousand gallons of water for every megawatt-hour it generates; a gas plant a fraction of that; solar and wind almost none. The water footprint of a data center is mostly the water footprint of its electricity. The honest fix is the same fix as for everything else here. Build clean, efficient power, and the number falls on its own.

None of this denies that a few dry counties feel a real pinch. Northern Virginia’s cluster draws real water, and in a drought-prone county a single large facility is a real load on the system. That is a siting problem and a sourcing problem, and the industry is already solving it. Companies increasingly cool with reclaimed wastewater and closed-loop systems that recycle the same water instead of evaporating it. Google already runs much of its fleet on non-potable water. Air-cooled designs use almost none. The answer to a data center straining one county’s water is to build it smarter, not to stop building data centers in a country with more fresh water than almost any nation on earth. “They are stealing our water” collapses the second you open the USGS tables.

The men who sold the jobs panic walked it back the month they filed to go public

The second pile of objections is about work. I used to believe the job-apocalypse story myself. I do not anymore, and it is worth saying who sold it.

The doom came from the people building the technology. Anthropic’s Dario Amodei warned that AI could wipe out half of all entry-level white-collar jobs within five years and push unemployment to twenty percent. OpenAI’s Sam Altman said human customer service would be “totally, totally gone.” Elon Musk told a conference that “probably none of us will have a job.” These were the salesmen, and they were selling fear.

Nvidia’s Jensen Huang told reporters he disagreed with almost everything Amodei said. He laid out the logic underneath the doom. AI is so scary that only a lab or two should build it, so expensive that no one else can, so powerful that everyone will lose their jobs. That is exactly why, the story goes, a couple of labs should be trusted to control it. The apocalypse was a moat. It kept regulators worried and competitors out.

This May, the same month both companies were reportedly preparing to go public at a trillion-dollar valuation apiece, Altman walked his version back. He said he was “delighted to be wrong,” that he had expected more damage to entry-level jobs than had actually happened, and that he had been “pretty wrong.”

Amodei walked nothing back. This week he published a new essay arguing that the government should have the legal power to block or reverse the release of AI models it judges unsafe. Even Axios noted the essay would “stir up a new set of accusations that Anthropic is proposing strict rules to lock in its own dominance or using frightening future scenarios as a marketing ploy.” It is the moat Huang described, now in Amodei’s own hand. The danger is so great, he says, that only a few approved labs, his among them, should be trusted to build. I read it as doom used to sell a wall that keeps competitors out, and I do not buy a word of it.

The dividing line at work runs between the people who use AI and the people who refuse to. We have watched this movie before. When the personal computer arrived, the clerk who learned the spreadsheet pulled ahead, and the one who refused to touch it got left behind. The economist James Bessen documented the cleanest version of this with bank tellers and the automated teller machine. The ATM cut the number of tellers a branch needed from twenty to thirteen. Everyone assumed that meant the end of the teller. Instead, cheaper branches meant banks opened far more of them, teller jobs grew, and the human skills that machines could not do became the valuable part of the job.

The data on AI points the same way. A study of customer-support agents found that an AI assistant raised output fourteen percent on average, and thirty-four percent for the least experienced workers, because the tool handed the newest people the best workers’ playbook. In GitHub’s own test, developers given an AI coding assistant finished a task more than fifty percent faster. The tool lifts the person who picks it up.

The one study everybody cites for the opposite case does not hold the weight put on it. Researchers at Stanford found that early-career employment fell sixteen percent in jobs most exposed to AI. But the entry-level slump started in 2022, before ChatGPT existed, and newer work pins it on something else. Two economists compared hiring data before and after the pandemic. Once they separated the effect of AI from the effect of remote work, AI’s effect “attenuates sharply and is often statistically indistinguishable from zero,” while working from home kept driving the decline. The Federal Reserve Bank of New York reached the same place. Remote work, not AI, explained roughly two-thirds of the rise in young-graduate unemployment, because managers stopped hiring juniors they could not train over a screen. The machine did not take the entry-level job; the empty office did.

So the cutoff stands where it always has, with the worker who picks up the new tool on one side and the worker who will not on the other. You can fight the change. It will cost you your career. That is not the technology’s fault. That is yours.

China is building as fast as it can pour concrete, and it wants us to stop

While we argue about jars of water, China is building.

In a single year China added more electric capacity than the United States has added in nearly a decade. It poured roughly half a trillion dollars into power. It began building new coal plants at a ten-year high. It approved ten reactors for the fourth year running. And it ran a national program to plant data centers next to cheap inland electricity. On top of all that, it subsidizes up to half the energy cost of its own data-center operators.

Beijing tells Americans the opposite of what it does at home. Its state broadcaster and its English-language papers run a steady stream of content aimed at the United States, warning that data centers are spiking electricity bills and damaging communities. The same government that pays half its own data centers’ power bill is telling American families that ours will ruin them. That is a competitor trying to talk a rival out of building, not concern for the American ratepayer.

It is not only China. Russia’s state outlet RT has promoted campaigns to ban data centers in Maine, New York, South Carolina, and Oklahoma. Iran’s hardline press has run the same line. Three of America’s adversaries, pushing one narrative at one target. When Beijing, Moscow, and Tehran all want you to stop building the same thing, that itself is information.

They are not playing fair on the technology either. The White House science office accuses China of “deliberate, industrial-scale campaigns” to steal American AI. OpenAI has told Congress that the Chinese firm DeepSeek copied the behavior of American frontier models, a practice called distillation: training a cheaper model on a more expensive one’s output to take the capability without paying to build it. It is a way to slip past American limits on advanced chips. They want our chips. They want our models. And they want us to stop building the compute that produces both. None of that is a coincidence.

The influence has reached our own politics. This spring Senator Bernie Sanders headlined a Capitol event on “the existential threat of AI” alongside two academics tied to the Chinese state, one of them from a Tsinghua University program that does research for China’s military. They argued, on a United States Senate platform, that the AI race with China is an inaccurate narrative. Sanders and Ocasio-Cortez then introduced a national moratorium on new data centers, the exact build-freeze Beijing’s propaganda has been calling for, backed by a coalition of left-wing environmental groups. Treasury Secretary Scott Bessent said Sanders was “inviting foreign nationals to tell the United States how to regulate AI.”

The cleanest fact in the whole story belongs to OpenAI. With full view of its own systems, the company identified and banned a Chinese influence operation it nicknamed the “Data Center Bandwagon,” a set of accounts posing as Americans that manufactured the precise claim that data centers were raising electricity prices for ordinary families. The operation did not go viral. It got caught first.

You will read that none of this is proven. NPR ran a piece saying there is no evidence China funds the data-center opposition. On a close read, the piece debunks exactly one thing: an investor’s loose claim that two specific Utah nonprofits were Chinese “cells.” That claim was thin, and it deserved to be knocked down. But in the same article NPR concedes that Chinese influence is not absent, reports the OpenAI ban, and quotes its own expert admitting he cannot rule out that China is quietly paying people to oppose data centers. “No proof of a wire to one local group” is a different sentence from “no Chinese influence in the debate.” NPR established the first and walked past the second.

Whether Chinese money is directly funding the American groups organizing the protests is now the subject of an open congressional investigation. The chairman of the House Ways and Means Committee, Jason Smith, says his committee has tracked Chinese money into nonprofits organizing against data centers; the chairman of the Energy and Commerce Committee has asked the FBI for a briefing on foreign influence in the fight. Those investigations are ongoing. The rest is already on the record. Beijing builds at home, scares us abroad, steals our models, and got caught seeding the exact narrative now echoing in our hearing rooms. Every month we spend on jars of water is a month we hand to the people who want us to lose.

The grid strain is real. The answer is to build, not to freeze.

The strongest objection is real. Building this much compute strains the electric grid, and somebody pays for the new power. Virginia’s own legislative analysts estimate that unmanaged data-center growth could push residential power costs up sharply by 2040. A state senator there put it bluntly: your electricity costs will go up because of data centers. That is a real concern, and it has two answers, not one. The first is to build more power, and a lot of it, so the new demand meets new supply instead of bidding up the old. That means more nuclear, more natural gas, more of everything that can run a grid around the clock. The second is to make the data centers pay for what they use, through large-load contracts and rate structures that wall off the new demand so it covers its own supply instead of landing on your neighbors’ bills. Even the Sierra Club asks for that second part, and it is right to. The trouble is that the same people demand the second answer while fighting the first.

But the strain is neither new nor mysterious. By federal energy data, American electricity demand was essentially flat for nearly two decades, as efficiency gains canceled out growth. The grid absorbed air conditioning. It absorbed the electrification of half the country. A return to rising demand is a normal thing a serious country plans for and builds toward, not a crisis that justifies freezing in place. Data centers used a bit over four percent of American electricity in 2023 and, by the Berkeley Lab projection, could reach as much as twelve percent by 2028. That is a build order, not a stop sign.

And the building has started. Microsoft is paying to restart the undamaged reactor at Three Mile Island, roughly 835 megawatts of carbon-free power coming back online. Google signed the first corporate deal for small modular reactors, a newer reactor design small enough to build in a factory, for up to 500 megawatts. Michigan is restarting a decommissioned nuclear plant for the first time in American history. Amazon and Meta are buying reactors too. Add natural gas, add solar and storage, and the supply is there to be built. The binding constraint is not physics or money but permitting, and a connection queue that Berkeley Lab tracks at roughly 2,300 gigawatts of power, nearly twice the entire existing fleet, waiting years for permission to plug in. That is a choice, and choices can be changed.

The upside is concrete, and it lands on the towns that say yes. Loudoun County, Virginia reports that data centers sit on about four percent of its commercial land and pay thirty-eight percent of the county’s general-fund revenue, more than a hundred million dollars a year that has let the county cut homeowners’ tax rates every year for a decade. Build it right and the servers help pay for the schools.

Strip away the arguments and what is left is degrowth

What the loudest opponents are against tells you what this fight is really about.

The same environmental groups fighting the data centers oppose new nuclear power. The Sierra Club readopted its policy against building new reactors in 2024, in the middle of the boom, and that ban covers the small modular designs the tech companies are now betting on. It opposes the new natural gas plants. It fights the transmission lines. Oppose nuclear, oppose gas, oppose the wires, and oppose the data centers, and you have opposed every source that can actually run a modern grid around the clock. That is a less-energy position wearing a clean-energy label. And there is a name for the belief that rich countries should use less energy on purpose.

The name is degrowth. Its leading theorist, the anthropologist Jason Hickel, defines it plainly as “a planned reduction of energy and resource use,” aimed at making wealthy nations “slow down the pace of material production and consumption.” Less energy, on purpose, as the goal. And Hickel does not hide what sits underneath it. He told an interviewer last year that degrowth is “a gateway into socialist thought for the 21st century,” “an anti-capitalist position,” and that “it is capitalism that needs to be overcome.” That is the movement describing itself, not my spin on it.

This is the oldest losing idea in modern politics, and it loses the same way every time. In 1968 the biologist Paul Ehrlich opened The Population Bomb with a sentence the movement has never lived down: “The battle to feed all of humanity is over. In the 1970s hundreds of millions of people will starve to death.” They did not. World population doubled, and people got better fed, not worse. The economist Julian Simon bet Ehrlich that the prices of five metals would fall as the population grew, because human ingenuity would find more and use it better. Simon won every metal, and Ehrlich mailed him a check. Ehrlich never recanted. To his death this year, he was still calling economic growth “the creed of the cancer cell.” The scarcity prophets are always certain, and they are always wrong.

Simon’s answer was the right one, and it is the conservative one. The ultimate resource, he wrote, is people, free people, using their minds. Energy is the master resource that lets those minds turn one thing into another. Abundance is not naive optimism. It is the entire record of the modern world, every forecast of starvation and exhaustion filed and falsified, every limit dissolved by people allowed to build.

So when someone holds up a jar of brown water in a hearing room, look at what is underneath it. Not a water test. A worldview that treats human progress as the disease and human appetite as the sin. Scarcity is not a fact these people discovered. It is a choice they are trying to make for the rest of us.

Build

Build the data centers. Build the reactors and the gas plants and the wires to carry the power. Drive the cost of intelligence down until a kid with a laptop holds what only a national lab used to hold.

The scarcity lobby has lost every argument it ever picked. It lost to Norman Borlaug and the wheat that fed a billion people it swore would starve. It lost to Julian Simon and his five metals. It lost to the shale boom it called impossible right up until it happened. It will lose this one too.

The only real question is whether America builds the future, or hands the lead to Beijing while we stand in a hearing room arguing about a jar of water.