The Land and Water of Compute — How AI Data-Center Expansion Is Tearing American Communities Apart

In May 2026, the family of Ansley Brown, a resident of Georgia, made public a notice they had found in their mailbox: Georgia Power, the state's electric utility, needed to build a 35-mile, 500-kilovolt high-voltage transmission line to serve a new cluster of AI data centers. The route would cut across more than 330 private properties in Coweta and Fayette counties, and some houses would be demolished entirely. Tucked inside the letter was a single loaded term — eminent domain, the government power of compulsory taking.

This is not an isolated case. Three months earlier, farmers in Indiana discovered that Amazon was building the largest data center in the United States right next door, dedicated exclusively to Anthropic. What the local community feared was not the transmission lines but the drinking water — the water table was dropping at a pace visible to the naked eye. One resident told the media: "They're going to take what we depend on to survive."

The AI industry's hunger for compute is tearing open the wounds of American communities in the most primitive, physical form.

One: The Legal Machine of Eminent Domain

What Georgia Power is invoking is Section 22-3-20 of the Georgia Code, a statute that allows state-authorized electric utilities to take private land or easements for the construction of "necessary public utility facilities." Transmission lines fall squarely within that category — and the astronomical power demands of data centers naturally count as a "public purpose" as well.

As a matter of legal principle, eminent domain is a power granted to government by the Fifth Amendment of the U.S. Constitution: the government (or entities it authorizes) may take private property for public use. American electric utilities are typically authorized by state law to exercise this power; as regulated entities providing public service, they receive a "delegation" of eminent domain from state legislatures, to be used for building transmission lines, substations, and similar infrastructure.

This institutional framework worked smoothly enough when it came to building power grids and laying highways. But when the electricity demand of AI data centers becomes the underlying driver pushing eminent domain forward, a boundary question emerges — does supplying power to private technology companies count as "public use"?

Georgia Power claims that compulsory taking is a last resort, accounting for less than 1% of its land transactions, and that the company negotiates first and offers compensation above market price. Yet in the Project Wansley transmission-line project, the company has already issued threats against more than 330 properties — and the "above market price" formula itself dodges a more fundamental question: the private value of these lands and houses to their owners cannot be measured by any "market price."

Two: Community Resistance

Property owners have not gone without a fight. The Brown family raised objections and demanded higher compensation; the project is moving ahead regardless. Other families placed on the list have chosen public protest, documenting the process on social media and trying to halt the land grab through public pressure.

But such resistance has limited effect against the combined force of institutions and capital. In Georgia, the eminent domain granted to electric utilities is not selective — for critical infrastructure projects it is close to absolute. The most owners can do is bargain for a larger compensation figure, not stop the project itself.

What makes matters worse is that the pressure communities face is not only legal. Some owners have reported that after refusing buyout offers, they were subjected to persistent, harassment-style "negotiations." This is not the signature style of any particular company, but a pattern that recurs throughout American infrastructure expansion — when negotiation fails, the methods change.

Three: Spreading from Land to Water

If the crux of the dispute in Georgia is "who owns the land," the question in Indiana is "whether there is enough water."

The data center Amazon is building in Indiana is so vast that local residents have begun reporting wells running dry and water pressure collapsing. More Perfect Union, an Emmy Award-winning American media outlet, covered the story in detail. It reveals a deeper problem: the resource consumption of AI data centers is not one-dimensional — electricity means land, cooling means water, and both ultimately point to the limits of what a community can bear.

Place the Georgia and Indiana cases side by side, and a clear trajectory of expansion emerges: the physical costs of AI data centers are spreading outward step by step — taking land, erecting lines, sinking wells, consuming water. Every time one problem is solved, a new conflict is created along the next dimension.

Four: From "Strategic Investment" to "Social Cost"

The narrative of the American AI industry in the first half of 2026 contains one revealing turn.

On the commercial level, the cost-rationality inflection point of AI investment is approaching — companies are beginning to question the return on AI spending, and some of the tech giants are trimming their AI budgets. At the same time, the physical costs of AI are accelerating their spillover to the community level: Georgia's transmission lines, Indiana's well water, and the reshaped face of communities across the country where data centers arrive.

Between these two trends lies a tension that has yet to be adequately discussed: just as companies start to do their books, the costs paid at the societal level have already been booked as "externalities" and left hanging on the ledger long term. The beneficiaries of the AI data-center building boom are the technology companies and the capital markets, while the costs are itemized, line by line, on the communities' bill.

Five: Pinkerton-Style Harassment — The Shadow Side of Land Acquisition

In early June 2026, an American self-working farmer posting under the handle homer4400 published a video describing what he had been through. He owns rural property, and a land-acquisition company had been pressing him to sell the land for the construction of an AI data center. After he refused explicitly, the harassment began: cars driving past his property in the middle of the night; a message at five in the morning telling him that "in the gabion area at the front of your property, there's a gift left for you and your daughters" — it was gift cards and cash. The buyer kept contacting him from changing numbers, repeating over and over: "You'll sell now, or you'll sell later."

This style of operation has a proper name in America — "Pinkerton-style harassment," named after the notorious Pinkerton detective agency of the nineteenth century, which was hired by capitalists to suppress the labor movement and forcibly evict farmers. In its modern version, the core operating pattern is: once negotiation fails, switch to sustained harassment, applying psychological pressure through "legal means," until the exhausted and insecure landholder accepts the buyout offer.

This finding completes a full chain of evidence alongside the Georgia and Indiana cases: land taking (Georgia) → water depletion (Indiana) → harassment (the homer4400 case). The physical costs of AI data-center expansion are extending from institutional compulsory taking all the way into near-gray-zone methods of land acquisition.