Elena Schlossberg spent 12 years forcing Amazon to pay for its own transmission lines in Virginia. Now she argues that cutting the power supply to data centers is the most practical brake on runaway AI development
In For Humanity episode 84, John Sherman speaks with Elena Schlossberg, founder of the Coalition to Protect Prince William County, about 12 years of fighting data center expansion - and why she believes stopping the power supply is the most actionable path to slowing down AI.
When most people think about AI risk, they picture software. Algorithms. Models that might one day outpace human oversight. Elena Schlossberg thinks about concrete.
She thinks about transmission lines cutting through three states to feed a single county in Northern Virginia. She thinks about diesel generators running around the clock next to residential neighborhoods, emitting some of the most dangerous air pollutants known. She thinks about the 38-acre family garden center that used to employ dozens of people in Prince William County until Amazon bought the property and announced what was coming next.
Schlossberg founded the Coalition to Protect Prince William County in 2014 after Amazon Web Services proposed a data center campus in the area and, in her telling, expected the surrounding community to absorb the cost of the transmission line it required. Not just the visual blight of the facilities. The actual utility bill.
"Your electric utility can exercise eminent domain over your property," she told host John Sherman on For Humanity #84, "and then make you pay for it, because it's public infrastructure."
That fight - four years long, against Amazon and Dominion Energy - became a model for a growing national movement. And in this conversation, she connects it to something larger: the question of whether AI development can be slowed down at all, and what the most practical lever for doing so actually looks like.
The central argument Schlossberg makes is structural, not emotional. The data center industry, she contends, found a weakness inside how public utilities work in the United States. Infrastructure that serves private corporations gets classified as public infrastructure - and the costs get passed to ratepayers.
The coalition's approach in 2014 was to prove that the triggered transmission line existed almost entirely to serve Amazon, then develop a cost allocation mechanism to make the company pay. They lost initially. They kept going. They eventually won, and the precedent they set is now being cited in data center fights across the country.
"The only way to make them innovate is to make them pay for their own stuff," she said. Her analogy is clean: if someone else is buying your car, you pick the Porsche. If you're paying, you pick the Ford Focus with good gas mileage.
Today, that question of who pays is no longer a local issue in Prince William County. The regional grid operator, PJM, covers 13 states. Transmission lines are being routed through Pennsylvania, Maryland, and West Virginia to serve data centers in Virginia. Property is being seized in communities that will never see economic benefit from the facilities being built.
Virginia alone already has more data centers than China. "They're laughing at us," Schlossberg said, "destroying our utility grid and risking reliable and affordable energy."
One of the more durable industry talking points is economic development. Data center projects are pitched to local governments with promises of jobs - construction workers, electricians, plumbers.
Schlossberg dissects this carefully. Construction work is finite. Long-term employment inside a data center facility is minimal. The tell, she says, is the parking lot - usually small, usually empty.
Meanwhile, the expansion is actively displacing the businesses that do create sustained local employment. In Prince William County, a 38-acre family-run garden center called Maryfield - complete with a cafe, a dog park, native plant offerings, and a real staff - was bought by Amazon. The land is now slated for data centers. Light industrial businesses, plumbing suppliers, and electricians' shops that were part of the local economy are also leaving, squeezed out by an industry buying up every available parcel.
"Just one Target superstore will create 10 times as many jobs as a data center," she said.
Sherman pushed the job displacement argument further. The work happening inside those facilities - the analysis, the writing, the customer service calls, the design - is work that was previously done by people. A woman Elena knew recently spent a phone call with a Verizon AI system that couldn't resolve her problem, kept shifting accents mid-call, and appeared to be learning from the interaction in real time.
"If you replace every job," Schlossberg said, "what are we as humans if we are not working?"
For much of the conversation, the risks Elena describes are immediate and physical - property rights, air quality, electricity costs, community displacement. Then Sherman introduced a different register of concern.
He walked her through the basic case for AI extinction risk: that the companies building frontier models publicly acknowledge their systems could pose catastrophic dangers, that leading scientists agree, and that the developers themselves admit they cannot fully understand or control what they are building. His framing centered on the logic of learning systems - something designed to explore and optimize, becoming vastly more intelligent than the humans theoretically overseeing it, will not indefinitely accept slow, limited human gatekeeping.
Schlossberg had not encountered the argument laid out this way before. Her response was careful.
She pushes back on the self-awareness framing. Coming from a background in counseling and psychology, she holds a specific definition of intelligence - one that includes self-awareness - and she does not believe current models meet it. But she does not dismiss the concern. She points to a different path to catastrophe: not a model that chooses to destroy humanity, but one that makes consequential errors at a scale and speed humans cannot track or correct.
"I don't know that it becomes self-aware," she said. "But I do believe that you could rely on this kind of AI that could trigger something that ends up being the end of mankind."
She cited WarGames - the 1983 film in which a military computer, playing out scenarios, nearly launches nuclear war - as closer to her mental model than Terminator.
What she found striking was the convergence of concerns. Whether the worry is accelerating climate change from data center energy use, AI systems being given authority over critical infrastructure, or the specific extinction risk scenarios Sherman described, the practical response points in the same direction. Slow down. And her lever for slowing down is the one she has been pulling for over a decade.
The argument is more precise than it sounds. Schlossberg is not arguing for an end to technology. She describes herself as someone who would support the industry if she believed the benefits were real and the costs were being honestly distributed.
What she is arguing is that the data center industry is already financially fragile - revenue to debt ratios are badly lopsided - and that the most effective intervention is to stop subsidizing their infrastructure costs. When corporations bear the actual cost of what they build, the calculus changes. They build differently. They build less.
She also raises an immediate public health dimension that rarely surfaces in AI coverage. The industry's response to insufficient grid capacity has been to generate power on-site using gas turbines. "Bring your own generation," she called it. Those turbines run around the clock next to residential communities and emit the class of pollutants that carry the most serious health risks. This is not a future scenario. It is happening now.
There is also the obsolescence question. Sherman raised the example of an AI-designed rocket engine - generated, 3D-printed, fired, and functional, looking unlike anything a human engineer would have drawn. The data centers being planned in 2024 and 2025, built through 2028 and 2029, may already be designed around hardware that will be superseded before the facilities come online. The industry is potentially racing to build last generation's infrastructure on debt it cannot service, at community expense.
"The way to make this whole thing slow down," Schlossberg said, "is to say no."
The final stretch of the conversation turned to organizing strategy. Sherman asked directly: if he showed up at one of Elena's data center community meetings and asked for 10 minutes to speak about extinction risk, how would that land?
Her answer was grounded in experience. She has already been putting herself into rooms she was not obviously welcome in - Northern Virginia Technology Council conferences, industry-adjacent gatherings - and found that when the full picture gets laid out, the response from people inside the industry is not dismissal. It is recognition.
The political conditions she describes are, in her view, genuinely unusual. The data center issue does not sort along partisan lines. It touches property rights, energy costs, air quality, water supply, and local economic stability - concerns that cross every demographic and political affiliation. Her own household splits its votes. They agree on clean water.
"The data centers are afraid of exactly you and I talking," she said.
She closed with something close to optimism. Twelve years into a fight she started against a single Amazon transmission line, she can see the national conversation catching up to what communities in Northern Virginia have known for years. Elected officials are beginning to engage. The infrastructure reality is too visible to dismiss.
The table has been set, she said. The question is who shows up to sit at it.
Watch the full episode: For Humanity #84 on YouTube
Subscribe to our Substack: substack.com/theairisknetwork
The AI Risk Network team