Nashville just used eminent domain to kill a data center

What happened

On August 5, Nashville's Metro Council voted to authorize the use of eminent domain to acquire a 32-acre parcel in the Flatrock area of southeast Nashville, next to the Nashville Zoo. The site had been quietly optioned for a data center development, and the council's move — passed on the second of three required readings — is designed to convert the land into public parkland before ground can be broken.

This is not a zoning fight. It is a taking. The distinction matters. Zoning denials get litigated as land-use disputes; the developer typically retains the parcel and either sues, redesigns, or sells. Eminent domain flips the ownership. The city pays fair market value, the parcel becomes public, and the project is dead — not delayed, not relocated on the same lot, dead. The developer, Nashville-based FerrisBuilt, had reportedly been in discussions with a hyperscaler tenant but had not disclosed the end user publicly. Local reporting from CoStar and the Tennessean pegs the planned facility at roughly 250 MW, which puts it in the same weight class as the campuses AWS, Google, and Meta have been dropping into Virginia, Ohio, and central Georgia.

Councilmember Sean Parker, who sponsored the resolution, framed the vote in the language of community land use — proximity to the zoo, traffic, noise, the visual footprint of a windowless warehouse the size of a shopping mall. What went almost unmentioned in the floor debate, but sits underneath every one of these fights nationally, is power draw. A 250 MW campus is roughly the residential electricity consumption of 180,000 homes. Nashville Electric Service has not disclosed what interconnect commitments, if any, were on the table.

Why it matters

Eminent domain in the United States was built for the opposite use case. The canonical justifications — railroads, highways, transmission lines, water — are all about *building* infrastructure over the objections of individual landowners. Nashville is running the machinery in reverse: using a public-purpose taking to stop private infrastructure that a city government has decided it doesn't want. The legal theory is that a park is itself a public use, which is uncontroversial on its own. The novel part is the timing and the target. Cities have condemned blighted lots to build parks for a century. Condemning an actively-optioned data center site to preempt a hyperscaler is new territory.

The backdrop is that the community-level backlash to data centers has gone from a Virginia-and-Northern-Ireland curiosity to a national pattern in about eighteen months. Prince William County put a moratorium on new campuses in the Prince William Digital Gateway area. Fauquier County rewrote its comprehensive plan to require a special-use permit for anything over 25 MW. Chandler, Arizona, and Peculiar, Missouri, both killed proposed campuses this year on noise and water grounds. Groveport, Ohio, forced a redesign after residents documented 24/7 sub-bass from an existing Meta facility. What was previously friction — longer entitlement timelines, tighter setbacks, water restrictions — is now hardening into outright denial. Nashville is the first council I'm aware of that has said, essentially, *we will pay you to go away*.

The relevant question for anyone running capacity planning is no longer "can we get power" but "can we hold the site." Every hyperscaler capacity model I've seen treats land as the cheap, solved input in the stack — power, water, and fiber are the constraints. That assumption held when local governments viewed a data center as a tax-base win with a modest permitting cost. It stops holding the moment a council can decide, mid-development, to swap the parcel for a park. The pricing implication is subtle but real: a site that is optioned but not yet under construction is now carrying political risk that behaves like a call option written against the developer. Someone is going to have to model it.

The community complaints are not, incidentally, wrong. Modern hyperscaler campuses run continuous chiller loads that emit low-frequency noise in a band that residential construction is very bad at attenuating. Air-cooled designs — the ones that avoid the water fights — are noisier than liquid-cooled designs. And the AI training buildout has pushed rack densities from 10-15 kW to 60-120 kW, which means more cooling, more noise, and a much bigger transformer yard. The industry's public story about being a "good neighbor" has not kept pace with what a 2026-vintage GPU hall actually feels like from 400 meters away.

What this means for your stack

If you are a builder shipping software on top of one of the big three clouds, the short-term impact is nothing. Capacity that's already under construction in Loudoun, Columbus, Atlanta, and the Dalles will land on schedule. What changes over the 18-36 month horizon is where the *next* wave lands, and how much premium the hyperscalers pay to lock in sites that survive political review. Expect more announcements in states with pre-emptive statutes that block local overrides — Georgia's SB 34, Texas's Chapter 313 successor regime, Virginia's evolving fight over the same. Expect fewer surprise builds near dense residential areas.

For teams evaluating self-hosted or colo footprints, this reads two ways. On one hand, the same NIMBY dynamics apply to a 5 MW enterprise colo, just with less news coverage. On the other, the hyperscaler power crunch that community denials are amplifying is the thing that finally makes on-prem GPU inference pencil out again in some verticals. If you are a mid-size shop with a rack of H100s in a Tier III facility your CFO thought was legacy debt, that facility is suddenly a strategic asset. The spread between on-demand cloud GPU pricing and amortized on-prem is not going to compress on the schedule anyone forecasted two years ago.

For AI-adjacent startups, the second-order effect is training-region volatility. Reserved capacity contracts in emerging data center metros are now exposed to the risk that the metro itself decides to stop being a data center metro. This is a real diligence question to ask a cloud provider: not just where the capacity is, but what the entitlement status is and whether the local jurisdiction has any pending anti-data-center legislation. Two years ago that would have read as paranoid. Today it's underwriting.

Looking ahead

The Nashville taking will get challenged. Tennessee has a relatively pro-property-rights bench, and the developer has both the standing and the money to litigate for years. The interesting question is not whether this specific parcel becomes a zoo expansion — it probably does, eventually — but whether the *strategy* gets copied. If half a dozen more councils try this in the next twelve months, the industry will get a federal preemption fight, probably attached to whatever the next round of energy-and-permitting reform looks like. If it stays a Nashville one-off, it becomes a cautionary tale in site-selection decks and nothing more. My read: this gets copied. Data centers are the new highways, and cities have remembered that highways can be routed around neighborhoods that push back hard enough.