Photo courtesy of ZNetwork
Every prompt a chatbot receives has to travel somewhere. Contrary to popular belief, that somewhere is not a cloud but a warehouse filled to the brim with servers that draw water, power, and money from the surrounding towns. These warehouses are data centers, and in 2026 they’ve become the most contested buildings in American politics. According to the International Energy Agency (IEA), a typical AI data center uses as much electricity as 100,000 homes. As tech companies rush to build ever-more of the building that underpins the AI revolution, utility customers and a growing number of politicians from both parties want to know who is paying for it.
For Upper School History teacher Michael Whalen ’08, the question reaches students long before they can vote. “I think it is important for anyone to stay on top of current events because it is one of the responsibilities of citizenship,” he says. “While they can’t vote yet, AI will continue to impact the country and the world they live in. It will impact their future careers and, in some respects, the geopolitical balance of power. Data centers are connected to the efficiency and growth of AI.”
This conflict has proven to be an immediate reality. In May, residents of East Whiteland Township in Chester County put up signs opposing a proposed data center; in August, Upper Marion Township rejected plans for five data center sites around King of Prussia, after residents gathered nearly 20,000 signatures against them; and a statewide Muhlenberg College poll in March found that 64% of Pennsylvania adults see data centers as a crisis or problem.
Many critics have complained about the constant noise. Data centers run around the clock, and their low-frequency hum penetrates walls, carrying hundreds of feet beyond the source. Kalabe Sebsibe ’27 shares his concerns, saying, “I’ve read that data centers give off a constant low hum that people can hear inside their homes.” He continues, “There have been proposals about 20 minutes from here. I’d be lying if I said I wasn’t nervous about one ending up close to home.”
Voters aren’t the only ones divided on data centers; Pennsylvania’s governor Josh Shapiro has completely flipped his public stance. During a 2025 press conference, Shapiro celebrated Amazon’s $20 billion plan for data center campuses in Bucks and Luzerne counties. But this May, amid growing voter backlash, he signed an executive order requiring data centers to run on at least 32% clean energy and to sign community benefit agreements to keep their tax breaks, among other constraints. His Republican opponent, Pennsylvania Treasurer Stacy Garrity, favors a natural-gas-first approach and points to Homer City, where a former coal plant is being converted to power data centers, as proof that abandoned industrial land can work again.
Still, data centers offer benefits that appeal to everyday citizens, including jobs, tax revenue, and an edge in what President Donald Trump calls the AI race. Because there is no single federal policy on data centers, each state decides for itself whether to welcome them. Some states have rejected proposals in order to avoid the costs that come with them, such as sustained power grid use and heavy water use. However, a rejected data center is often built in a neighboring state instead, resulting in a state that declined the offer suffering the costs without reaping the benefits. This phenomenon incentivizes states to approve the construction of data centers because ones built within their borders contribute to tax revenue. On the other hand, skeptics argue that residents who never asked for data centers shouldn’t have to pay for their power lines and that towns, not developers, should control zoning decisions.
At the federal level, the response has relied largely on voluntary measures, chiefly President Trump’s Ratepayer Protection Pledge, which asks companies to regulate themselves. In September, the House passed the Ratepayer Protection Act, which encourages state regulators to consider requiring data centers to cover their own costs without strictly mandating it. Supporters consider it to be flexible, while critics call it a suggestion with no real protection. The bill has since been stalled in the Senate. In the meantime, governors, including Texas’s Greg Abbott and New York’s Kathy Hochul, have moved to pause new data center approvals on their own.
For most families, the first encounter with data centers comes through their electricity bills. Based in Valley Forge, PJM Interconnection, the grid operator serving Pennsylvania and 12 other states, purchases future power supply through auctioning. In July, that auction reached its price cap for the second consecutive year, with data center demand serving as the largest driver. Comparatively, an earlier auction, held in 2024, led to a 10% increase in PECO bills.
Data centers also contribute to rising costs by competing with other buyers for copper, electricity, and electricians, so much so that Federal Reserve (Fed) officials now name the buildout as a source of inflation. In September, the Fed raised interest rates for the first time in about three years, a move intended to slow spending by making borrowing more expensive. However, the largest technology companies finance much of their expansion in cash, so the increase falls instead on families with mortgages, car loans, and small businesses hoping to expand. As a result, the Fed’s contractionary approach slows AI growth by a virtually negligible amount while devastating much more fragile parts of the economy.
Photo courtesy of MIT News
Not every AI company is as insulated. CoreWeave, which rents out AI computing power, pays about 8.3 percent interest on roughly $35 billion of debt. If AI’s promised profits arrive slowly, highly indebted firms such as this one would be the first to struggle.
For now, the clearest winners in the situation are construction personnel. Electrical work makes up 45% to 70% of a data center’s construction cost. Many Pennsylvanians, though, doubt these jobs will outlast construction.
Available water is the concern residents raise most often, and in certain locations, it is a serious one. A large data center can use one million to five million gallons a day, often in regions already facing shortages. Nationally, US data centers use about 449 million gallons a day for cooling, less than a quarter of what the country’s golf courses use, and agriculture consumes far more than either.
A more pressing issue is the energy required to keep the servers running. To meet data center demand, utilities have delayed coal plant retirements and proposed a surge of new natural gas plants, which typically operate for 30 to 40 years. A plant built to serve today’s demand could still be burning fuel into the 2060s.
The community, however, has shown significant concern for the land itself. Max Leung ’28 says, “As someone who cares a lot about ecology, what worries me isn’t only the electricity; it’s the land.” He continued, “Every new data center clears hundreds of acres that animals already live on. It feels like another round of ‘we take more land, and then more after that.’ I don’t know where we’re supposed to stop.”
The fight over bills and zoning rests on an older and more fundamental disagreement about what AI is for. In The New Yorker, staff writer Becca Rothfeld argued that much of the resistance to AI comes down to one idea: the value of learning, making things, and caring for others lies partly in the fact that a person does them. Through this lens, data centers are not a neutral supply of computing power. They make AI cheaper and easier to reach for, and make it easier to hand off work that was never meant to have a shortcut, like the essay draft that teaches you what you believe or the article that questions those beliefs. They may even threaten the very idea of a vocation.
The opposite case starts from the same premise. If the process is what matters, then AI that clears away the menial work returns hours to the things only people can do. On that view, AI’s purpose is not to replace what humans do, but to give them time to do it. Benjamin Heo ’28 explains, “I am genuinely excited about AI…It helps me learn faster and get through the tedious parts of my work, which leaves me more time for things I love and that feel more human.” He concluded, “If data centers mean less busywork, that’s hard for me to argue against.”
The question was never whether AI saves time. It was always whether humans can be trusted to know what to do with the time.
On November 3, Pennsylvanians voted for a new governor, with control of both chambers of the legislature at stake and data centers on the ballot in all but name. Whoever takes office in January will decide how much the state asks of the companies that want to build here. The prompts will keep coming regardless. What remains open-ended is who pays for them, where they are answered, and what they take.