The Industry That Shut Itself Down, and the One That Won’t

In February 1975, about 140 of the world’s leading molecular biologists checked into a conference center on the California coast and spent four days arguing about whether they should be allowed to do their jobs. They had already stopped. Seven months earlier, a group of them had published a letter asking their own colleagues, worldwide, to voluntarily halt certain recombinant DNA experiments until the field could figure out what it was doing. The remarkable thing is that their colleagues did. There was no law. There was no agency. There was a new technology with risks nobody could yet measure, and a set of practitioners who decided that the moment to draw lines was before anything went wrong rather than after.

The industry building the physical substrate of artificial intelligence is being handed the same invitation right now, at scale, in public, in hundreds of council chambers across forty-nine states. It is answering with shell companies.

That opportunity will not stay open for long. Every industry that has ever grown on the logic of infinite demand has eventually met a limit it did not choose. The only variable that has ever mattered is whether the industry chose first.

The scale of the invitation

Data Center Watch counted roughly $130 billion in data center projects blocked or delayed in the first quarter of 2026 alone, across more than seventy-five projects — matching the total for all of 2025 in ninety days. Organized opposition groups grew from 396 at the end of 2025 to 833 by the end of March, and they now operate in forty-nine states. An earlier tranche of the same research found the political opposition splitting almost evenly, 55 percent Republican and 45 percent Democrat.

On June 16, 2026, the city council of San Marcos, Texas voted 4 to 3 to make data centers ineligible for every parcel of land inside city limits — the first Texas city to do it, in a state that has spent three years passing preemption laws designed to prevent exactly this. Council member Lorenzo Gonzalez explained his vote in a single sentence: “Ultimately, I concluded that the promised benefits remained speculative while many of the concerns raised by residents remained unresolved.”

Speculative. Unresolved. Those words point at evidence, not water or noise, and they describe a case that was never actually made.

The pattern is older than the technology

Strip out the specifics and this story is one of the most familiar in American industrial history.

The railroads spent three decades expanding on the belief that demand for freight was inexhaustible, setting rates community by community according to whatever the traffic would bear, and treating the resulting farmer revolts as ignorance. They got the Interstate Commerce Act in 1887, written by people who had stopped listening to them.

The patent medicine trade sold whatever it liked, disclosed nothing, and argued that consumers were free to choose. It got the Pure Food and Drug Act in 1906, on terms set in the aftermath of a novel about meatpacking.

The chemical industry disposed of waste at the lowest available cost for half a century, entirely legally, until a school and a subdivision in Niagara Falls were found sitting on top of some of it. Love Canal produced Superfund in 1980 — a retroactive, strict, joint-and-several liability regime, which is roughly what a legislature writes when it has concluded that an industry cannot be trusted to account for itself.

In each case the industry’s real mistake was timing, not the harm itself.

Asilomar is the counter-example, and it worked for one specific reason. The biologists skipped the pledge and went straight to a technical classification: tiers of experiment matched to tiers of physical and biological containment, specific enough to be enforced and specific enough to be argued with. Those tiers became the NIH guidelines the following year. Fifty-one years on, molecular biology remains a field that largely governs itself, produced an enormous amount of medicine, and never had its Superfund moment. The scientists in that room paid a real price in lost time and lost freedom of action. What they bought with it was the next half-century of the field.

What the current playbook really admits

The way these projects arrive in a community tells you what the industry believes about its own case.

Meta’s Rosemount, Minnesota project was known locally as “Project Bigfoot,” with the rezoning application filed by a shell entity called Jimnist LLC and a second shell appearing on the state utility petition. A Fortune 50 company approached Hermantown, Minnesota as “Project Loon” through a construction firm. In North Mankato, the community development director signed a nondisclosure agreement within five minutes of being asked. A University of Mary Washington survey found that 25 of 31 Virginia officials contacted had signed data center NDAs.

The justification is straightforward, and it’s honest, as far as it goes. Neila Wilson, who runs regional data center engineering and operations for Visa, put it plainly at an industry event: “The minute the person selling the land knows who’s going onto the land, the price goes up exponentially. There is, from a business perspective, some intentionality there.”

That is a real commercial interest worth a real and measurable amount — the spread between the anonymous price of a parcel and the price a seller charges once he knows a hyperscaler is across the table. Call it a few million dollars on a large site.

Now price the other column. Sentinel Grove Technology Park near Port St. Lucie, Florida was a $13.5 billion investment drawing up to a gigawatt, and it came with self-contained water facilities and 60-foot setbacks — the kind of engineering concessions that answer the exact objections residents raise. It lost its planning board vote in October and was withdrawn in February. A few million dollars of land price leverage on one side; a project of that magnitude on the other. The industry has been trading the permit for the parcel, and the exchange rate is grotesque.

But the deeper cost isn’t the arithmetic so much as what the secrecy gives away. Concealment is what an industry does when it has privately decided it cannot win an argument, so it arranges not to have one. And it does not even work: the Data Center Watch researchers found that “in some cases, opposition mobilized before any project was officially filed; the mere rumor of a data center was enough to trigger organized resistance.” The fight happens anyway. Secrecy only guarantees the developer arrives at it with no standing to describe its own project, while every local official who signed an NDA has been converted into a suspect. St. Louis County Board Chair Annie Harala, who signed one, described the cost in personal terms: “I have some heartache from very dear friends who felt like I was hiding something.”

That is the hidden term in the contract. The developer gets a cheaper parcel; the mayor pays for it with her credibility, and she is the person whose support the project needs most.

Oracle vice president TJ Ciccone offered the industry’s own diagnosis at a conference in Maryland in May: “The problem is, inside the industry, the first rule of Fight Club is we don’t talk about Fight Club.” Asilomar’s entire method was the opposite proposition — that the way to keep authority over a technology is to hold the hardest argument about it in public, early, and on your own initiative.

The harder half of the argument

Transparency is the easy ask, and some of the industry is already there. In March 2026, Microsoft ended its use of NDAs with local governments worldwide, stating that “being transparent with the communities where we operate or seek to operate is paramount.” Google, Amazon, and Meta declined to make the same commitment at an Ohio legislative hearing. David Tolson’s firm, DBT Data, pursued a project in a community that had already rejected five previous data center proposals, committed to sharing detailed plans with residents before filing anything, and got approved: “We now have their support simply by engaging, being present and being transparent.” Five failures under concealment, one success under disclosure, same community, same land use. Watch which approach accumulates approvals over the next twenty-four months; that experiment is running right now.

But disclosure alone is only the first tier of an Asilomar. The harder commitment is the one the industry has not been willing to say out loud: that some compute should not be built where and when the market would build it.

This is the sentence that sticks in the throat, because it concedes that the demand curve is not sacred. And that concession is exactly what the record supports. No industry in history has been correct that its demand was infinite. The railroads believed it until 1893. Telecom believed it hard enough to lay fiber that sat dark for a decade. Housing believed it until 2008. In every case the correction was violent precisely because nobody inside had been permitted to say the word “enough” while it still could have been said cheaply.

Self-modulation, then, is capital discipline wearing a different hat, not simply a civic posture. An industry that can articulate which sites are wrong, which load growth is speculative, and which projects should wait for grid capacity rather than race it, is an industry making better allocation decisions than one that treats every megawatt as self-justifying. The community that asks “do we need this here” is asking a question the investment committee should have asked first.

And there is a specific reason this technology in particular warrants it. Compute is a general-purpose capability, not a product with known effects being deployed at scale. Its second-order consequences for labor, information, energy markets, and the structure of firms are being discovered in real time by everyone, including the people building it. That is the precise condition Asilomar was designed for: not proven danger, but unsized uncertainty in the hands of the only people equipped to reason about it.

The fork

Data centers are worth building. The compute they house underwrites drug discovery, materials simulation, climate modeling, and most of the ordinary digital infrastructure everything else now sits on. A country that zones them out parcel by parcel will discover the consequences slowly and then all at once, in electricity markets and in where the next decade of capital chooses to go. San Marcos has now demonstrated that citywide prohibition is available even in the most preemption-happy state in the country, and 833 organized groups are reading the transcript.

So there are two roads, and the industry is standing at the junction with its hand on the wrong turn signal.

One road is fifty state regimes and several thousand municipal ones, written over the next four years by legislators and council members who were told nothing, negotiated with under NDA, and asked to approve projects filed by companies that would not give their names. Those rules will be blunt, inconsistent, technically illiterate in places, and effectively permanent. They will be written about this industry rather than with it.

The other road looks like a room on the California coast. A published technical taxonomy of facility classes with binding commitments attached — gallons per year, megawatts at peak, decibels at the fence line, grid additionality, water recycling thresholds — specific enough to be enforced and specific enough to be argued with. Tenants named before the application, so the community hears it from the developer rather than from a reporter working a records request. No elected official asked to sign anything. And, hardest of all, a stated standard for where and when the industry will decline to build.

The San Marcos vote was 4 to 3. One council member. The information that might have moved that vote existed the entire time, in engineering specifications and operational data the applicant had every right to publish and chose to withhold. The industry keeps calling the outcome irrational opposition. A more accurate word is verdict — and a verdict is what you get instead of a settlement when you refuse to put your evidence on the table.

That is the pattern worth carrying out of this, well beyond data centers. The organizations that keep authority over their own future are the ones that can show their work — that can put defensible evidence in front of a decision while it is still being made, rather than assembling a defense after the vote. Whether the scrutiny comes from a city council, a regulator, an examiner, or a court, the position you can document is the only one you get to keep.

Asilomar bought fifty years with four days of uncomfortable honesty. It remains, by a wide margin, the best trade anyone in a disruptive technology has ever made.

Sources: