Prince William County, Virginia, shows these are very real concerns. In December 2023, rezoning for the massive Prince William Digital Gateway—a 2,200-acre data center campus to be built on the border of Manassas National Battlefield—was rammed through in a vote by supervisors at the end of their term that the Virginia Court of Appeals later found illegally violated public notice laws. Pete Candland—the supervisor in whose district the Digital Gateway was to be situated and who had campaigned for election as an advocate of rural preservation—resigned in disgrace after it came to light that data center developers bought property he owned at a handsome premium. The project was opposed by local citizens, but also by the American Battlefield Trust and such luminaries as Robert Duvall and Ken Burns.

In 2026, Prince William County supervisors continue to advocate for increased data center buildouts, and have often rejected carefully constructed regulations around noise and zoning for data centers. The most pro-data-center supervisors are from the heavily Democrat eastern half of the county, while the data centers themselves are concentrated in the western, rural-suburban Republican-leaning half of the county, effectively offloading the negative externalities from their neighborhoods to rural Republicans while politically and professionally reaping the rewards.

As Peter Thiel has pointed out, California has its own version of the oil curse or “resource curse,” where incompetent governance can survive indefinitely on tax revenues from big tech—these tax revenues untether the California government from the classic need to govern well to maintain power and budgets. Data centers threaten to extend the “Tech Curse” to nearly every county in the country. In addition to broken local incentives, AI can change national incentives.

AI gives its owners data on the deepest insights of the American mind, far beyond what Google had before LLMs. This data can be used to subtly alter someone’s opinion, remove sources, or promote a politician. Many Silicon Valley researchers have long and honorably explained some of these risks, before the current “Takeoff,” and deserve credit.

AI is “trained” on nearly every piece of text ever written that we have access to now: the internet, old books, and Reddit. It is really trained on a lot of Reddit. AI necessarily will have a bias of some kind, and most AIs are tilted toward the progressive narrative. It makes sense: these models were largely trained by Bay Area rationalist and effective altruist-adjacent engineers and executives, and these engineers and executives tune the model to agree with their simplistic, malicious world framework, specifically. 

Earlier in 2026, the leading model company Anthropic accidentally leaked the source code for its Claude Code programming tool, revealing how it assembles the hidden instructions—the “system prompt”—it feeds the model. The instructions tell the model how to interpret inputs it receives, and how to change the outputs it generates in response. The leak showed Claude’s instructions told Claude everything from how to avoid helping commit crimes, to how to avoid emojis in a response unless explicitly requested by the author. It is very possible the system could be changed with a few simple, hidden prompts to favor one political candidate, CEO, or country. If the leading tech companies were to secretly bias their models toward a candidate, it is unlikely to be a candidate whose politics align with rural Republicans.

“Ah,” counters Silicon Valley, “but what we are creating is more powerful than a nuclear bomb. We need it to defeat China!” In January 2026, Anthropic CEO Dario Amodei said that selling AI chips to China is “a bit like selling nuclear weapons to North Korea.”

Silicon Valley leaders like Dario Amodei and Sam Altman really have intentionally promoted comparisons between AI and the nuclear bomb. This is an old Valley trick; if a CEO can convince investors and the public that the technology they are building is potentially dangerous, the technology is assumed to be powerful. And everyone wants to invest in a company building powerful technology! However, Silicon Valley is arguing it is creating a nuclear bomb, and it, not the United States government, should choose how the bomb is deployed. This is already causing friction in Washington, DC. Red states are unlikely to have any say over how this technology is developed and influenced as things stand now, no matter how many data centers are built in red states. How would a San Francisco AI company be forced to care what a Missouri senator thinks?

Historically, this is one of the many reasons why US military bases and nuclear weapons were located across the US, instead of concentrated on the coasts. While I am not saying there are not legion valid geopolitical reasons to think of China as an enemy, China so far is far less effective than Silicon Valley at influencing American elections and destroying American jobs, directly affecting a 40-year-old welder in Box Elder County with four kids.

Most of the main users of AI right now are what Musa Al-Gharbi calls Symbolic Capitalists, not blue-collar workers. Even as more and more rural Republicans adopt AI, especially entrepreneurs and farmers, their revenue share of (and thus importance to) the AI labs will not meaningfully change the companies’ trajectories—so their opinions won’t matter much.

AI labs have done their best to stoke fears of mass job loss and displacement, like this new commercial from Anthropic that looks like an ad for a society recovering from, well, being hit by a nuclear bomb. New technology really does cause mass job loss and displacement—especially during the tumultuous reorientation period when the tech is first deployed. In 2025, Amodei argued that AI could wipe out half of all entry-level white-collar jobs in America and spike unemployment to 10–20% in a relatively short amount of time. Meanwhile, AI threatens to replace blue-collar jobs at an even faster pace—self-driving cars are already ubiquitous in San Francisco and Austin, and China is running fully automated “dark factories” where lights and humans aren’t needed for their AI-driven robotic manufacturing.

Factories are traditionally placed in relatively focused areas like the industrial parks of major cities. Data centers are springing up everywhere, especially on the outskirts of small and mid-sized cities in the Midwest and South, where power and water infrastructure are established but regulations and taxes are lower than in the city core. Land is different from other capital. It is nearly impossible to create new land. Data centers are usually built on beautiful farmland, instead of the blighted areas of cities like Detroit or Gary, Indiana, which have power infrastructure, a need for tax dollars, and areas where infill is dearly needed.

Everywhere that data centers are built, their exteriors are very ugly. I referred earlier to the controversial Prince William Digital Gateway Project next to Manassas National Battlefield Park. The site where so many Americans lost their lives is preserved today as beautiful, rolling Virginia countryside. Imagine that hallowed ground juxtaposed with a grey, unlabeled data center building. Now instead of one data center, imagine hundreds of these buildings on 2,200 acres, surrounding the battlefield. This vision does not honor the memory of Americans who died in the Civil War. Beauty really matters. Early American industrialists understood this, and built truly beautiful industrial facilities, many of which have now been converted to lofts in cities like Chicago and Milwaukee, where it is not unusual to see an old Art Deco or even Egyptian Revival industrial building. Now, data centers are an eyesore, what millennial grey design would be if it were a building.

Once a data center is built on land, the land is rubbish for most other uses, and the view is even more rubbish. And yes, sometimes the water quality or availability does suffer, and electricity costs can go up. They generate noise, traffic, and more. Why should residents be forced to take on a local cost, for a “national” benefit? Everyone in earshot or eyeshot should be compensated with the benefits of these data centers. Complimentary subscriptions, tokens they can use or resell, training in the tech—many rural residents in the US are intelligent but have been locked out of the wealth creation of the coasts. The data center buildout as presently constituted threatens to widen the gap.

Elon Musk understands this, and SpaceX has begun to offer deeply discounted Starlink internet access to local residents near xAI’s Memphis data center, a good start.

There are downsides, but these tools have created a lot of wealth, including multiple trillion dollar public companies, right? A 40-year-old welder in Box Elder County with four kids has not had the chance to invest in Anthropic or OpenAI. Instead, the wealth directly created by equity in the companies will enrich the endowments of wealthy universities; universities which the welder’s smartest child will never have a chance to enter, no matter his test scores or IQ, and it will enrich venture capitalists, who are overwhelmingly graduates of those universities. By choosing to IPO late, the labs are asking Americans to support their data center buildout while excluding Americans from investing during the time of highest upside. Of course, this doesn’t apply to every AI company. Google and Microsoft have been public for a long time, as has Nvidia.

***