The Progress, Peril, and Paradox of AI Data Centers

Every so often, a societal trend unites the left and the right in shared outrage. At the moment, it’s AI data centers. New York just became the first state to halt permitting for new construction, and now Bernie Sanders and AOC are calling to take that policy national.

While “bipartisanship” is usually invoked to describe a handful of moderate Democrats or Republicans voting with the other party on a particular issue, often something most voters don’t even support, in this case it’s real. Recent surveys show that 63% of Americans support a one year moratorium, including 67% of Democrats, 66% of independents, and 58% of Republicans. Asked in a slightly different way, 71% oppose the construction of AI data centers in their local area.

An important fact to keep in mind is that popularity is not some infallible metric of right and wrong or good and bad ideas. Fads and moral panics come and go all the time. A poll is not an argument. Even when a bipartisan consensus correctly identifies a problem, the rushed “solutions” politicians offer probably won’t help. They’re more likely to be stunts they can tout in their reelection campaigns to prove they’re on the “right side of history.”

According to those surveys, more Americans worry about data centers in their neighborhoods (71%) than data centers in general (63%). So should we be talking about one-size-fits-all national policies or practical problem-solving on the local level?

The Energy Question

Of all the worries about data centers, spiking residential electricity prices have dominated the public conversation. These facilities draw enormous amounts of power, so following the laws of supply and demand, you’d expect prices to rise accordingly. Elizabeth Warren has claimed that in some areas, electricity bills have increased by up to 267% over the past five years. Her solution, as always, is to increase taxes on data centers, since they were “built on American land using our shared resources,” whatever that means.

So are data centers causing a nationwide energy crisis? The truth is more complicated. First, the number Warren cited is extremely misleading. While whole-sale electricity prices have increased by up to 267%, that only represents a fraction of the bills paid by households. It doesn’t account for the costs of transportation, distribution, or taxes.

Her claim also ignores inflation, which peaked during the exact time period she was referencing. If you look at nationwide average prices adjusted for inflation, they’re up about 10-15% from the most recent low point in 2021. However, that was preceded by a roughly equivalent decline during the first Trump administration. Today’s prices are just around where they’d floated from 2007 to 2017.

But that’s just the national average, which could be obscuring localized spikes near data centers. Virginia leads the country in data center buildout, followed by Texas. So those states must have out of control electricity costs, right? In reality, both states are 10-15% below the national average. California, which has a similar number of currently operating data centers as Texas, has double the electricity costs. Whatever the impact of data centers may be, other factors clearly play a much larger role in the bills households have to pay.

Part of why data centers don’t appear to be spiking electricity costs is also a function of supply and demand. Earlier this year, the Trump administration negotiated the Ratepayer Protection Pledge with the biggest players in AI to encourage the buildout of new power sources to cover their electricity consumption. This is a win-win: increased supply and economies of scale mean that both residents and data center operators can avoid massive price spikes.

An Undeniable Problem

So if electricity costs aren’t really the problem they initially appeared to be, what are all the complaints about? Unfortunately, building a power plant next to a data center instead of connecting to the grid has one big downside: it’s loud.

In Sterling, Virginia, there’s a new data center located just across the street from a residential neighborhood. Its generators run 24/7, forcing residents to listen to a constant high pitched ringing, “almost like you have a headache.” This is obviously a disaster for quality of life, and if they want to move elsewhere, they’ll likely take a price hit when they try to sell their homes.

I lived in New York City for many years, and when I later moved to Hoboken, New Jersey, I discovered the simple luxury of quiet. Without the horn honking and trucks driving over metal plates outside of my apartment all night long, I experienced dramatically lower ambient levels of stress and anxiety. That luxury has been stripped from these communities. Living in NYC, I knew that was part of the package. But for this community in Virginia and elsewhere across the country, the noise has radically changed the deal they signed up for when they bought their homes.

Again, however, like Elizabeth Warren’s claims about electricity prices, there remains the question of how prevalent this problem actually is and whether there are fairly straightforward, win-win solutions on the table. It’s not obvious why you don’t just… build a data center away from residential neighborhoods. The United States isn’t exactly known for having too little land. Halting data center construction nationwide is treating a paper cut with amputation.

Our Ancient Anxieties

All of the practical concerns about AI data centers, reasonable if not always accurate, distract from a much more fundamental anxiety: what will be left for humans to do once AI advances far enough?

If we can trust centuries of historical precedent, there’s an economic concept to answer this question. It’s known as the Jevons paradox: when technology allows us to use a resource more efficiently, we use more of that resource, not less. William Stanley Jevons noticed in the 19th century that, as steam engines were improved to run more efficiently, we didn’t consume less coal. That would be the intuitive result, since you could do more work with less. We also didn’t consume the same amount of coal, even though we could again do more work with that. Instead, we consumed more coal, because there now was an economic incentive to find new industrial applications of steam engines.

The question here is: does AI replace human labor, or does it amplify the power of human labor?

In my experience, it’s both. Over the past year, I’ve been building apps to improve my everyday workflow and doing deep research for Dad Saves America that I never could have done before. I’m not just not spending time on tasks I had before. I’m doing completely new tasks, while getting everything else done in less time, because of AI.

The uncomfortable reality is that, yes, many jobs will disappear as we know them today, but the concept of a job will not. New ways to mix human labor with technology will continue to be discovered, as they always have been.

This is the story of every technological revolution in human history. It’s the story of the Luddites, the 19th century English textile workers who smashed the new machines. The irony is that they themselves relied on stocking frames, a technology that had put the old hand-knitters out of work less than 200 years earlier. The people fighting the new machines were, as usual, living off the second-to-newest ones.

Their anxiety was understandable, just as ours is now. But that doesn’t mean we shouldn’t step back, see the historical pattern, and realize that “this time is different” is what we’ve always said.