It was only back in May that Ben Thompson and I recorded an episode of Sharp Tech titled “Much Ado About Data Centers.” That was a fairly comprehensive look at the brewing backlash to building local infrastructure to sustain the American AI ecosystem, and three months later, our conversation looks quaint. I’m amazed by how much more widespread the anger has become, and how quickly it’s happened.
This has, needless to say, become a messy topic, with a bunch of relevant details to wade through. At the end, though, I’m going to explain why I’m less alarmed by the long-term implications of this fight than most of the tech community seems to be. I think the speed and virality of the backlash reflects the internet’s role as its primary driver, and social media-addled political movements tend to be less durable than moral panics of the past (compared to, say, 40 years of opposition to nuclear power). But before we get there, let’s survey the chaos.
Midterms and Moratoriums
Midterm elections are upon us, and data centers are now a political flashpoint all over the country. As the New York Times reports:
One new ad in Texas begins with someone typing into a chatbot, “Why is my electric bill going up?” Another in Ohio blames one of the state’s Republican senators for a proliferation of data centers, calling them “a total scam.” And in top House battlegrounds in Michigan, Wisconsin and Pennsylvania, the Democratic challengers are all running ads against the server warehouses powering the artificial intelligence boom.
Meanwhile, the National Republican Senate Committee warned AI companies that support for data centers is collapsing nationwide and will cost Republicans races in Ohio and elsewhere, and politicians from both parties are updating their platforms accordingly.
In total, 15 states across the political spectrum have now enacted or proposed temporary bans on data center construction. While a blue state like New York was the first to issue a moratorium on hyperscale data center construction, a purple state like Pennsylvania now sees both sides of its governor race fighting to outflank one another in the race to distance themselves from past support for data center projects. Even a red, builder-friendly state like Texas has announced a pause to all new data center approvals amid contested races for governor and the Senate, where Democratic candidates are hammering their Republican opponents on the issue. For reference, Texas is home to 335 data centers with “at least 248 more in the works,” and just 10 months ago was hailed by Governor Greg Abbott as the “epicenter of AI development.” That branding is no longer a political winner.
Last week, Heatmap News released polling results on a series of questions related to the data center issue, and the arrows are all pointing the same direction. Keep in mind that all of this happened across only 12 months:
This week, Bernie Sanders—who favors a federal moratorium on AI data center construction—took a victory lap as the unrest swirls:
Lies, Half-Truths and China
Building more data centers and computing capacity will bring down the cost of AI for everyone and likely lead to more innovation in the long run, in addition to incentivizing the buildout of new electricity resources that will benefit Americans all over the country, including with lower prices. Pausing AI data center construction in the U.S. won’t stop AI from transforming our economy; models will still be trained and deployed, and there will be enough inference to run them at America’s biggest companies. That said, while building one new AI data center won’t meaningfully alter the environment, building hundreds of them, and bringing more power online to sustain that growth, will likely lead to more emissions. Hardcore environmentalists who oppose that outcome are doing so for logically coherent reasons (with the caveat that stopping pollution in one country doesn’t stop it in another).
Beyond that fundamental tension, one obvious observation about the contemporary furor is that a few of the most popular arguments are trivially disproven. The claim that data centers use dangerous amounts of water has been debunked so many times that it’s now cliche to call it misinformation, and in general, the claims that data centers impose unique environmental harms relative to other industrial projects are wrong and misguided (in fact they impose much fewer). Researcher and Substack author Andy Masley has done a terrific job tethering this discussion to facts, and for anyone who’s skeptical, I recommend this piece on water, this piece on land use, and this tweet responding to a variety of stated concerns from data center opponents.
Alongside that now-robust economy of data center myths and mythbusting, there’s also the actual economy. Obviously AI spending and speculation has almost single-handedly powered a growing stock market (and millions of retirement accounts) for the past several years, and that presents its own set of systemic questions. But regardless of how precarious the financing becomes, every inch of compute that’s built is being utilized, and on the ground, data centers themselves have created a variety of winners beyond Nvidia. Last week Reuters highlighted a wide spectrum of attendant construction industries that are booming, while this week the New York Times ran an op-ed from an electricians union whose membership has doubled in the past 10 years.
As to everyone else, alongside the $27 billion in nationwide tax revenue that data centers are projected to generate for counties and states by the end of the decade, another popular concern is the risk of rising electricity costs for residential customers. On that one, the answer is tricky. Yes, the insatiable demand for computing power is adding strain to already-strained electric grids, but many new data centers are being built with their own power sources, rendering some of these concerns moot.
Moreover, using tech as a bogeyman to explain America’s steadily rising electricity prices oversimplifies the problem and risks forestalling potential solutions. In a comprehensive op-ed published by the New York Times in April, the authors found “[AI] is actually a small part of a much bigger problem,” because “[o]ur grid is too old and our supply of electricity too small.” Indeed, we inexplicably stopped building new electricity generation 15 years ago, well before the AI boom started. In this environment, data centers can benefit both short and medium term prices by spreading a grid’s operating costs across industrial clients (as opposed to residential customers) while also incentivizing the construction of new power. As that Times op-ed continues:
The concern around these data centers is understandable. But at least when thinking about the power system, we need to walk a fine line: We should welcome the growth of electricity demand, but seek to slow electricity inflation. Although it sounds counterintuitive, when a big new electricity customer shows up on the grid, it can sometimes help keep power costs down for everyone else, because it buys so much power that it can cover much more of the system’s costs, keeping a lid on rising power rates.
Cases will vary by market, but those counterintuitive benefits are real. And as a general matter, short-circuiting U.S. data center construction won’t solve our electricity inflation problems. To the contrary, the current AI investment cycle may be our best hope in a generation for expanding grids and building new power that will benefit businesses in any industry, as well as residential consumers.
Finally, returning to the category of viral claims that lack supporting evidence and tend to confuse the issues, I’m skeptical of the popular argument that the violent shift in American public opinion is being driven by Chinese propagandists. On that point, let me recommend one more piece from Andy Masley, because earlier this month he wrote exactly what I’ve been thinking all year:
There have been a few high-profile reports and claims purporting to show foreign involvement, and lots of people are sharing them as if they’re smoking gun evidence that China’s had significant impact. I’m worried that a lot of people are reading those headlines and nodding along without poking at the claims. When you do, they mostly collapse into a few specific examples of tiny organizations who happen to be based in China and other countries running some ineffective bot farms and behaving in ways that make them seem unlikely to be directed by the CCP, some negative coverage of data centers from Chinese state media, and a CCP-aligned billionaire funding networks of goofy goober fringe groups in the US like CodePink … All of these examples seem to be completely dwarfed by the organic local backlash driven by American media coverage, social media, and ordinary land-use politics. […]
All this made me pretty temperamentally skeptical of the idea that the data center panic is too intense and targeted to be an organic result of local American politics. American politics is pretty wild right now! I had also been following the panic from the start, and knew that many of the most popular talking points and ways of thinking had come from specific overconfident or clumsy reporting from established US papers, and spread quickly because their claims were shocking and alarming when read in isolation, and also sometimes exciting to talk about on social media (it’s very easy to buy a water bottle to hold in front of a camera on TikTok and compellingly repeat a fake statistic).
China’s most effective propaganda tool is buying off elites, and while some have argued that various climate organizations belong in that category and many of those same orgs now oppose data centers and grid expansion, we’re getting awfully attenuated at that point. I’ve omitted citations from most of that Masley excerpt, but I retained the link to the fake statistic he highlights—in which the Washington Post uses graphics to incorrectly tell readers that one AI search query uses more water than a 16.9 oz bottle of water—because I find it wonderfully illustrative of the point: the American information ecosystem is definitely screwed up enough on its own, independent of China, to power a months-long moral and political panic over data centers. Regardless of China’s intent to interfere in U.S. politics (I have no doubt they’re trying), I don’t think it’s a primary driver of the backlash. Nor do I think it’s useful to exoticize Chinese capabilities in lieu of better understanding our own politics.
How This Story Will End
In broad strokes, I do see how we got here. Data centers are massive, hulking buildings that can make dystopian humming sounds. In the face of rampant warnings about their environmental harms and the consequences of widespread AI adoption, voting no on constructing more of these buildings is a straightforward and honestly pretty rational response to both the past 35 years and the past 35 months. Tech has already atomized and disrupted our lives in ways that no one ever voted for and many people don’t love, while AI’s most visible leaders spent the three years after ChatGPT’s launch promising on one hand to remake the economy without humans, while warning of dire safety risks on the other. In April I said on Sharp Tech that I can sympathize with anyone who instinctively opposes the continued buildout of data centers on these grounds, and Ben echoed the point in both our May episode and a Stratechery Update from the same week.
Furthermore, to the extent a large portion of the opposition is not that principled or rational, and in fact this movement has spread rapidly across the American electorate thanks to the proliferation of half-truths or outright lies that citizens have seen on the internet, that’s of course poetic in its own way. Tech is now getting the pointy end of the stick after a generation of profiting from disruption of both the literal economy and our information economy. The latter has been upended, probably irrevocably, at the hand of algorithms that amplify conspiratorial thinking and erode faith in institutions across the political spectrum. This is the water we’ve been swimming in for a while now; anyone who’s been awake for the past 15 years shouldn’t be surprised that an informative op-ed from USA Today is less powerful than a hundred Instagram videos about water and noise pollution, or that a local government desperate for revenue now encounters a woman who’s convinced data centers will make everyone sick.
And all of that could, of course, be understood as a crushingly depressing reality check. Our leaders have been lamenting the disappearance of American factories for as long as I’ve been alive, and now that a new and enormously valuable category of factory is spurring blue collar job growth and incentivizing brick and mortar buildouts across the country, America is hand-wringing over electricity and bizarre lies about water. (Please, no one tell Instagram about the environmental costs of a rare earth refinery).
But I wouldn’t over-read this moment, either. This comes back to my lack of alarm about the long run implications. While it’s certainly true that a nihilistic sort of populism is on the rise in both parties and de-growth impulses are especially common on the left—two trends that aren’t going away—it’s worth keeping in mind that policy movements that draw their momentum from stuff people are saying on social media often lose steam just as quickly as they seemed to accelerate to the forefront of every political conversation in the country. Or to put it differently: of all the things that have taken their turn as The Current Thing, the track record for sustained outrage or support is not great. No one talks about de-funding the police anymore, Donald Trump is not building the wall, no one is flying Ukrainian flags today (well, except my parents), the MeToo movement is now seen as excessive in a variety of ways, and Alexandria Ocasio-Cortez, who today sponsors a bill to impose a federal moratorium on data center buildouts, recently made news for blithely admitting, “Woke 1 was crazy.”
Let me be clear that I’m not offering prescriptions to tech companies and telling them to just wait it out; they all need to calibrate their messaging a lot more carefully, and recognize that American towns are no longer falling all over themselves to hand out tax breaks and land. But I imagine incentives will continue to point the tech industry in the right direction without my help. Siting these plants needs to be done thoughtfully and power costs need to be borne by the companies themselves, and all of that is achievable, particularly when the benefits are undeniable.
In lieu of offering obvious advice, I’m more interested in cultural observation. Many have cited the 5G panic as a parallel for the kind of mass delusion around data center harms that has taken root this year, but personally, I think back to the conversations we used to have about fracking for oil and natural gas. That, too, was once branded as a uniquely harmful practice that imperiled our water supply, and while there were isolated incidents in which that was true, Obama’s EPA found that as a general matter those fears were overblown (although the agency later softened its exculpatory language after receiving criticism from activists and scientists).
Those EPA findings and decades of mostly safe fracking didn’t stop Elizabeth Warren and Bernie Sanders from promising to impose a federal ban on fracking when they ran for president in 2019 (which could have cost millions of jobs), so sure, there’s your reminder we will always have radical ideas to contend with. But the salience of fracking in political life has subsided, and along the way, while many companies engaged in the practice lost astounding amounts of money, the consumer surplus they delivered is undeniable. The shale revolution has made America the biggest energy supplier in the world and led to the mass adoption of natural gas as a replacement for coal, which went from supplying 50% of U.S. electricity generation in 2005 to 15% in 2026. And most importantly, Americans who once opposed fracking don’t care or even notice. There was never any grand public debate where one side won and the other lost; the work got done in spite of vocal critics, and then the critics mostly forgot about it.
Today those sorts of shifts in sentiment and attention happen across months and years, not decades. An infinite supply of memes means every meme has a shorter life span, and a constant flood of news and opinion means it’s easier than ever to wash away political movements that aren’t rooted in the lived experience of voters. To that end, I think what will undermine the cause of activists opposing new data centers is not anecdotes about their positive impact on local communities, but the simple reality that the vast majority of people are not affected by local data centers one way or the other. The endgame might be different if environmental risks really were dire, or tax benefits were all a sham, but all things considered, the objects of today’s populist outrage are arguably the cleanest and most profitable factories that America has ever seen.
So midterms will come and go. Opportunistic states like Texas will find a way to return to building. As entertaining or distressing as the current panic has been, the future of American AI infrastructure will not in fact be relocated to the UAE or outer space. Some of this controversy will probably be revived for a presidential election in 2028, but I expect the incentives for both companies building data centers and local governments attracting them will stubbornly prevail. And over time, voters won’t care.
I could cite polling as evidence that a shift in sentiment is foreseeable (support swings by 31 points when voters learn how modern data centers use water), but in truth, this argument is all instinct and experience. Everything on the internet is ephemeral, and that includes most of the political movements that start there.
Sharp Text is an extension of the Stratechery Plus podcasts Sharp Tech, Greatest of All Talk, and Sharp China. To subscribe and receive weekly posts via email, click here.