Data centers. Data centers everywhere.

For most of the generative AI era, artificial intelligence has had a human face. Sam Altman became the face of OpenAI. Dario Amodei became the face of Anthropic. Elon Musk became the face of Grok and xAI. Jensen Huang, with his leather jacket and seemingly endless supply of GPUs, became the face of the infrastructure powering all of it.

But I think something has changed.

The most recognizable face of the AI era may no longer belong to any of those people. It may not be a face at all. It may be a giant, windowless building sitting beside a substation, surrounded by transmission lines, cooling equipment, backup generators and security fences.

The data center has become the physical face of AI. Unfortunately for the AI industry, most Americans increasingly do not like what they see.

The speed of that change is remarkable. A Heatmap poll in August 2025 found Americans almost evenly divided over whether they would support a data center near where they lived: 43% supported one and 42% opposed it. By the spring of this year, support had fallen to 21% while opposition had climbed to 71%.

That is not a gradual shift in public opinion. That is a revolt.

The New York Times recently asked whether this might be one of the fastest changes in American political opinion we have seen. I don’t know whether it qualifies for the record book, but when 50 points of public sentiment move against an industry in less than a year, something more significant than NIMBYism is happening.

The industry’s reflex will be to say that people do not understand AI. They are afraid of progress. They have been stirred up by activists and politicians. They want the benefits of artificial intelligence but do not want the infrastructure required to produce it.

There may be some truth in pieces of that, but it is a lazy and dangerous explanation for what is happening.

I don’t think this is primarily an anti-AI revolt. I think it is an anti-deal revolt. Americans are looking at the bargain being offered to the communities hosting these facilities, and a growing number have decided that it is not a very good one.

To understand why, we need to leave Silicon Valley and visit the places where the AI era is actually being built.

Let’s begin in Broadview, Montana.

Broadview has roughly 140 residents, one bar and restaurant, and a school gymnasium that doubles as the center of the town’s social life. A 43-year-old rancher named Kassi Solberg has been driving into town to attend council meetings and ask a fairly basic question: What exactly is being built beside her family’s property, and who decided it could go there?

The proposed Quantica Infrastructure campus could cover about 5,000 acres and eventually consume as much as a gigawatt of electricity. That is more power than the local utility currently supplies to its entire customer base across Montana.

The project could require a transmission line approximately 21 miles long. Now the argument is over who should pay for it: The company that needs it or the people already paying electric bills.

Kassi Solberg is not attending those meetings to debate transformer architectures, artificial general intelligence or the latest model benchmarks. She wants to know what will happen to the wells, roads, property values and night sky. She wants to know how much electricity the facility will consume, what it could do to local utility rates and what Broadview will receive in exchange.

Those are not anti-technology questions. They are the questions people ask when something enormous is being built beside their homes.

The frontier of AI no longer runs only through a model laboratory. It runs along a rancher’s property line in Montana.

Now let’s head south to Richland Parish, Louisiana, where the calculation looks very different.

Richland Parish is cotton country. Farmers there can produce strong crops and still lose money because the cost of equipment, seed, fertilizer and fuel has risen faster than what they receive for what they grow. Families that once considered their land untouchable are reconsidering assumptions that survived for generations. Young people have been leaving, and the economic-development projects that were supposed to change the region’s trajectory never seemed to arrive.

Then Meta showed up with Hyperion.

Hyperion could eventually represent more than $200 billion in investment and grow to approximately five gigawatts of compute capacity. Supporting it could require ten new gas-fired power plants, along with enormous transmission, water and infrastructure investments.

Meta expects the project to create approximately 500 permanent jobs by 2035, reportedly paying around $90,000 a year.

Five gigawatts. Ten power plants. More than $200 billion. Five hundred permanent jobs.

The resources look industrial. The employment model does not.

From a distance, it is easy to look at property-tax reductions that could approach 80%, political maneuvering, nondisclosure agreements and the risks being absorbed by the surrounding community and declare it a bad deal.

Up close, the choice is harder.

Five hundred good jobs mean something very different in a parish that has spent decades watching farms consolidate, children leave and opportunity pass it by. Hyperion may be a lottery ticket with a considerable amount of fine print, but when a community has been offered very few other tickets, buying this one is understandable.

Farmers know risk. They can do everything correctly and still lose because the weather changes, costs rise or commodity prices fall. Meta has capital, financing structures and other options if the economics of AI change. Richland Parish has its land, schools, roads, water and hope that this particular crop will finally pay off.

Broadview and Richland Parish are not opposites. They are two sides of the same story. One community fears that something enormous is being imposed upon it. The other fears what will happen if its one enormous opportunity disappears.

Now we go to Boxtown in southwest Memphis.

In January 1961, newspaper photographer Saul Brown captured two Boxtown residents, James Threadford Jr. and Albert Lee Wright, riding a horse-drawn cart piled with firewood. Behind them stood a power plant.

That photograph contained an entire economic system in a single frame. Two men were collecting fuel while a facility almost within reach generated electricity for people somewhere else. Some homes in Boxtown still did not have electricity.

More than six decades later, Boxtown again sits beside infrastructure producing one of the most valuable resources in the world. This time, it is machine intelligence being manufactured inside the massive Colossus AI complex originally built by xAI to train Grok.

Residents have described sitting on their porches close enough to smell fumes coming from the facility beyond the woods. Once again, the community is asking what it receives for living beside infrastructure that creates tremendous value for people somewhere else.

The corporate tenants have protected themselves rather well. Anthropic’s arrangements give it access to hundreds of thousands of GPUs across Colossus and Colossus II. Google has also contracted for a large block of compute. Their agreements include provisions that allow the companies to leave after their initial commitments with 90 days’ notice.

Anthropic can leave. Google can leave. SpaceX can redirect the GPUs, reclaim them for Grok or rent them to the next AI company.

Boxtown cannot leave.

The models will change. The contracts will change. The logos on the buildings may change. Boxtown will still be beside the land, turbines, emissions and infrastructure.

That is the moral question beneath the entire AI data-center buildout: What does a community receive when it hosts infrastructure producing something indispensable for everyone else?

Then there is Fort Worth, Texas, where the new arithmetic of economic development becomes almost impossible to ignore.

A planned data-center project there carries a projected investment of approximately $2.1 billion. The number of permanent full-time jobs expected in its first phase appears to be 37.

They are good jobs, reportedly paying around $150,000 a year. But there are still only 37 of them. Tax abatements of as much as 70% were reportedly under discussion.

For generations, towns accepted the noise, traffic and pollution that came with factories because those factories employed hundreds or thousands of people. The bargain was visible. The plant supported families, local businesses and sometimes an entire county for generations.

Data centers are different. They are designed around automation, efficiency and minimal human intervention. A data center consuming the electricity of a small city might employ a permanent workforce that could fit inside Broadview’s only restaurant.

That does not make the facility worthless. It may generate tax revenue and enable enormous economic activity somewhere else. But we need to stop selling these facilities to communities as if billions of dollars in capital investment automatically produce thousands of local jobs. They do not.

The secrecy surrounding some of these projects makes the situation worse.

In Tucson, Arizona, residents discovered that a mysterious development called Project Blue belonged to Amazon only after a document was mistakenly released through a public-records request. The city ultimately canceled the project.

In Mason County, Kentucky, a family reportedly refused millions of dollars above market value for its farm because the prospective buyers would not reveal whom they represented.

In Virginia, a public-records survey found that 25 of 31 localities with existing or proposed data centers had signed nondisclosure agreements. Those agreements can prevent residents—and sometimes even elected officials—from knowing how much electricity and water a project will require or what effects it may have until the consequential decisions have effectively been made.

If you want public suspicion, conducting public policy behind private nondisclosure agreements is an excellent way to create it.

Meanwhile, utilities see the largest growth opportunity they have encountered in generations. Their incentives are not difficult to understand. More power demand means more generation, more transmission, a larger rate base and potentially greater returns.

But ordinary customers are asking whether they will be expected to finance infrastructure built primarily for some of the richest companies on Earth. They want to know what happens if a utility builds billions of dollars of new capacity and the AI demand projections do not materialize. They want to know who receives priority when the grid becomes constrained.

In Texas and elsewhere, data-center developers are responding by creating what amounts to a shadow grid: dedicated, behind-the-meter power generation designed to bypass the limitations and timelines of the traditional electrical system.

Bringing your own power may prevent some costs from landing on the public grid. It may also help projects get built much faster. But it creates another troubling image: A parallel energy system serving private AI campuses while households and ordinary businesses remain dependent on the public system.

When people already believe the game is rigged, building a private grid for billionaires is unlikely to restore their confidence.

Utah shows us how powerful the momentum behind this buildout has become. The proposed Stratos AI campus has been envisioned across approximately 40,000 acres, with potential power requirements reaching nine gigawatts.

Technology companies need compute. GPU manufacturers need deployments. Utilities and pipeline operators want customers. Construction companies want projects. Investors want returns. Politicians want economic-development announcements and ribbon cuttings. National-security officials want the United States to stay ahead of China.

There does not have to be a conspiracy. Every powerful participant has a rational reason to want the same thing.

That is why I have called this the AI Industrial Complex. Not because it is inherently sinister, but because aligned institutions and incentives create momentum that can easily outrun scrutiny and public consent.

Politicians of both parties are addicted to feeding at the trough of investment announcements, campaign money and enormous numbers they can place in press releases. The billionaire tech bros too often appear callous and tone-deaf, telling Americans that the AI future is inevitable while those same Americans worry about their jobs, electric bills and water supplies.

Utilities see growth. Investors see returns. Passed-over communities see what may be their last chance to participate in the next American dream.

Put those interests together and you get data centers, data centers everywhere.

Yet public resistance is now producing moratoriums, zoning fights and political campaigns across the country. New York has paused permits for certain large new facilities while it develops standards governing their environmental, energy and community effects. Local governments in red states and blue states are reconsidering the bargain.

America wants the Intelligence Grid. It just does not want it next door.

There is one final story that reveals how completely the cloud metaphor disguised what we were building.

In 2007, a group of thieves dressed as police officers talked their way into a Verizon data center in London by claiming they were investigating people on the roof. They tied up five staff members and stole millions of dollars of equipment.

The New York Times Magazine recently revisited the heist, and I encourage everyone to read the piece. It has all the makings of a very good movie.

But this is not merely a colorful story from 2007. It may be more relevant today than it was then.

The cloud was never in the sky. It has always had a street address. It has doors, loading docks, guards, power connections and people who can be fooled. As we concentrate enormously valuable GPUs, proprietary models, sensitive data and critical services inside AI factories, physical security becomes part of cybersecurity, national security and economic security.

The data center is not an abstraction. It is a place, and places can be found.

None of this means we can choose not to build data centers. We need them. Every AI agent, enterprise model, sovereign AI program and military system depends upon the infrastructure beneath it. Models run on chips. Chips sit in racks. The racks consume electricity and produce heat inside buildings that must exist somewhere.

The data centers will be built.

The question is whether we build them through public consent or political access. Whether developers pay the actual cost of the power and transmission they need or shift that cost onto ratepayers. Whether communities receive enforceable benefits or ceremonial promises. Whether local people become partners before the deal is finished or obstacles after it is announced.

Public acceptance is not a public-relations problem. It is an infrastructure requirement.

Most Americans will never meet Sam Altman, Dario Amodei or Elon Musk. Their opinion of AI may instead be shaped by the data center proposed five miles from their home, the gas plant constructed to power it, the water it consumes or the increase that appears on their utility bill.

If that becomes their most tangible encounter with the AI economy, the industry should not be surprised when opposition to the building becomes opposition to the technology inside it.