Hook
Dominion Energy’s latest filing in Virginia is data I cannot ignore. Buried in a routine regulatory update is a line that confirms what I have suspected since 2021: the AI industry’s physical expansion is not a matter of demand—it is a matter of seizure. Power companies are now wielding eminent domain to acquire private land for transmission lines feeding AI data centers. This is not a story about chips or models. This is a story about the fragility of centralized infrastructure when it meets the hard constraints of land and law.
Context
The narrative around AI’s growth has focused on training compute, GPU availability, and power consumption curves. All valid. But the overlooked variable is the right-of-way. To supply a 500 MW data center, you need a high-voltage line that cuts through dozens of privately owned parcels. The grid was not designed for the load density of modern AI clusters. In states like Virginia, Ohio, and Texas, utilities are now invoking a 19th-century legal doctrine—eminent domain—to force landowners to sell. The justification: the data center serves the public interest. The reality: it serves a handful of hyperscalers with deep pockets and aggressive timelines.
Core
Let me break this down with the same lens I used when I dissected the EOS genesis contract in 2017. There, a race condition allowed infinite token minting under specific block producer configurations. Here, the race condition is between land acquisition speed and energy delivery. The bug is not in code—it is in the legal system.
First, incentive asymmetry. The utility company profits from construction and rate base expansion. The hyperscaler gets subsidized power and tax breaks. The landowner receives compensation calculated on “highest and best use” before the data center announcement—often far below the real market value after. The community absorbs the visual and environmental cost. This is not a market. It is an extraction mechanism.
Second, systemic fragility. The eminent domain process is legally contested. Landowner lawsuits can delay transmission line construction by 12–24 months. Meanwhile, GPU orders are placed based on projected data center energization dates. A single legal challenge in a key county can cascade into a supply chain bottleneck that no one in the hype cycle accounts for. I saw this pattern in the Terra/Luna collapse: a feedback loop that appears stable until the threshold is crossed. In 2022, I calculated that UST’s death spiral would trigger at a $10 billion market cap. Here, the threshold is a legal ruling that redefines “public use.” A bug is just a feature that hasn’t yet caused a crash.
Third, regulatory alignment failure. The SEC’s regulation-by-enforcement in crypto is mirrored here by state public utility commissions allowing eminent domain without explicit legislative amendment. Both are forms of deliberately withholding clear rules. The front-runner didn’t read the code—they read the political tea leaves. The same cynical logic applies: players who know the system best exploit the ambiguity.
From my 2020 analysis of Uniswap V2 front-running, I learned that extractable value is not a bug—it is a feature of the protocol’s incentive design. Eminent domain is the same: it is a feature of a legal protocol that prioritizes rapid infrastructure deployment over property rights. The question is whether that feature is sustainable.
Contrarian
Now the angle the bulls will throw at me. They say: eminent domain has been used for highways, pipelines, and power lines for a century. AI data centers are just another critical infrastructure. Without it, the US loses the AI race to China. The scale of demand requires extraordinary measures. They have a point. I cannot deny the urgency. The compute growth curve is exponential. The grid is linear. Something has to give.
But here is what they miss. The justification relies on a definition of “public use” that was originally intended for projects that benefit all citizens—roads, schools, utilities. A private data center that feeds mostly enterprise AI workloads and shareholder returns does not fit. Stretching the doctrine creates a precedent that can be applied to anything—crypto mining farms, private prisons, luxury data centers. The consequence is a degradation of property rights that erodes the very trust that underpins market economies. Code doesn’t lie, but people do. And so does their interpretation of law.
Takeaway
The AI industry is about to learn what DeFi learned in 2020: scaling via coercion comes with hidden liabilities. The legal challenges will mount. The costs will be socialized. The latency between code deployment and physical reality will expose every flaw. Whether you are building a Layer2 or a data center, the lesson is the same. Trust the incentive structure, not the narrative. And check the mempool—because that is where the real action is.