Two local women in Utah posted a video mocking Kevin O’Leary. Not because of anything he said on Shark Tank, but because his $100 billion data center project threatened to swallow 40,000 acres of their backyard. That’s roughly two Manhattans.
O’Leary fired back on social media, accusing protesters of being paid operatives, even hinting that China was behind the opposition. The community packed planning meetings to capacity and filed lawsuits. The state government scrambled to revise its approval process.
By June 2026, O’Leary caved. He agreed to slash the project’s footprint by 75 percent. His public admission: “I screwed up.”
That concession tells a bigger story. One that the AI industry’s breathless growth narrative has tried very hard to ignore.
The Numbers That Nobody Wants to Talk About
In 2025, community opposition stalled or killed $156 billion worth of data center projects across the United States. In just the first half of 2026, another $98 billion in projects have hit the wall.
A freshly published United Nations report lays out the trajectory: by 2030, AI data centers will double their consumption of both electricity and water. The per-facility numbers are staggering. A mid-sized data center drinks as much water daily as a small town. Large facilities gulp five million gallons per day, enough to supply 50,000 people. AI-driven rack power density has jumped from 8 kilowatts in 2021 to north of 50 kilowatts in 2026.
These aren’t problems you solve with a software patch.
In Howell, Michigan, a Fortune 100 company (name undisclosed) saw its data center proposal rejected for the simplest of reasons: so many residents showed up to oppose it that the meeting room couldn’t hold them. In Virginia, a columnist admitted in the Washington Post what many privately feel: “I have to be honest, on data centers, I’m a NIMBY.” Florida went further and sued OpenAI outright.
The regulatory posture has shifted from “should we govern this?” to “how do we contain it?”
When “Bring Your Own Power” Becomes the Business Model
Electricity has become the single biggest constraint. The industry’s response has a name that sounds almost parodic: BYOP, Bring Your Own Power.
The concept works like this. A data center developer first locates a site where power generation is possible. They build their own power plant and substation. Only then do they construct the actual computing facility. The data center tail now wags the energy dog.
O’Leary’s other project, Wonder Valley in Alberta, Canada, plans to run on 7.5 gigawatts of self-supplied natural gas generation. That’s enough electricity for 7.5 million homes.
The question this raises is uncomfortable: if tech companies are building their own power grids, what exactly is the public grid for?
Nuclear power has entered the conversation with the subtlety of a freight train. Eighteen nuclear-powered AI data center projects are currently in various stages of development, totaling 31.2 gigawatts of planned capacity. Microsoft’s headline-grabbing decision to restart the Three Mile Island nuclear plant (yes, that Three Mile Island, site of America’s worst nuclear accident in 1979) captures the desperation perfectly.
Engineers on forums and social platforms keep pointing out the obvious flaw: nuclear plant approvals take a minimum of ten years. Insurance companies balk at the risk profiles. By the time a reactor comes online, AI architectures may have iterated through three generations. The timeline mismatch is brutal.
One data center investor put the new reality bluntly: “Our strategy now is simple. Control land that has power, no matter where it is. You can run fiber anywhere. You can grade any terrain. But electricity? Money alone can’t conjure that.”
Power has become the new land grab, and the balance of leverage is tipping toward whoever controls physical resources.
The Conflict Isn’t What You Think
It’s tempting to frame this as progress versus preservationists. That framing is wrong.
The push side is obvious: tech giants (Alphabet announced $80 billion in AI infrastructure investment), data center developers, and local governments hungry for tax revenue and jobs.
The opposition is more interesting. In Utah, the people blocking O’Leary’s project weren’t environmental activists. They were ordinary residents worried about water (Utah is already drought-prone), disappearing farmland, and the erasure of community character.
Astra Taylor and Saul Levin wrote in the Guardian: the $156 billion blocked in 2025 will be a smaller number than what gets blocked in 2026. They argued this isn’t merely a technology question. It’s a democracy question.
The structural tension is almost designed to produce gridlock. The federal government pushes AI competitiveness (the Trump administration issued executive orders streamlining data center approvals). But land use, zoning, and utility regulation sit at the state and local level. Washington says “national strategy.” A Utah resident says “my well is running dry.”
The interest map is straightforward: tech companies need expansion to justify valuations. Investors need infrastructure returns (Anthropic’s IPO filing is the signal). Local residents need to protect existing quality of life. State governments are caught between tax revenue and ballot boxes.
Fortune reported that in some Republican strongholds, opposing data centers has become a political asset for candidates. One community organizer captured the dynamic: “This was never red versus blue. It’s people who live here versus people who want to industrialize here.”
With 2026 midterm elections approaching, the political weight of this issue is still growing.
History’s Rhymes and One Critical Difference
Railroad expansion in the 19th century sparked endless legal battles over land acquisition. Rural electrification in the early 20th century took three decades. Fiber optic deployment in the 1990s ran into “last mile” community resistance at every turn.
The script is always the same. Proponents invoke national interest and economic inevitability. Locals ask why they should bear the costs.
But there’s a difference this time that breaks the historical pattern: AI’s resource consumption curve is exponential.
A railroad, once built, is a fixed piece of infrastructure. Fiber optic cable, once laid, serves for decades. AI compute demand doubles roughly every 18 months. A data center built today may need expansion or replacement within three years.
This isn’t a one-time infrastructure investment. It’s a continuously accelerating resource draw.
The Atlantic published a case study from Chile that crystallized the absurdity: a single Google data center consumed one thousand times the annual water usage of the entire local population. Residents only discovered their tap water pressure had dropped because an “algorithm factory” had quietly moved in next door.
The UN report makes the point explicitly: AI is not merely a digital technology. It is “a material system with measurable environmental costs.” The researchers argue that investors should treat power consumption, carbon emissions, water usage, and land occupation as “significant risks within AI infrastructure investment portfolios.”
Wall Street is starting to listen.
The Contrarian Thesis: AI Hits Its Ceiling at the Power Grid, Not the Algorithm
The mainstream narrative says AI’s growth bottlenecks are algorithmic breakthroughs, data quality, and regulatory policy.
That analysis misses the harder constraint: grid capacity.
A 2026 investment report from law firm Ropes & Gray states it plainly: the competitive advantage in data center investment isn’t technology. It’s “BYOP strategy and early site control.” Whoever secures power wins.
The implications cascade outward.
First, AI’s geographic distribution will be forced toward energy-rich but remote regions. O’Leary didn’t choose Utah for its proximity to Silicon Valley talent. He chose it for land availability, solar potential, and low population density (meaning relatively fewer voices to oppose him, or so he thought).
Second, AI companies will be dragged into the energy business whether they want to be or not. Microsoft restarting a nuclear plant is just the opening act. Expect more tech companies acquiring power utilities, investing in generation projects, and lobbying to reshape energy regulation.
Third, and this is the most aggressive prediction: by 2028 to 2030, electricity will overtake chips as the dominant cost in AI training. GPUs are expensive but they’re a capital expenditure. You buy them once. Electricity bills arrive every hour of every day. When that crossover happens, AI competition reshapes around “who has cheap power” rather than “who has the smartest model.”
A Gartner vice president analyst warned that “disruption in energy availability will limit new data center growth for generative AI and other purposes.”
The translation is simple: no matter how advanced your algorithm, without electricity it’s just math on paper.
What This Means for Different Players
For investors: stop fixating exclusively on AI model companies. The quieter bet may be whoever solves the power problem. Data center REITs, power infrastructure plays, and small modular reactor (SMR) projects could offer more durable returns than the next foundation model startup.
For people working at tech companies: recognize that your employer’s AI roadmap may get cut not by budget constraints but by power constraints. “We can’t afford it” is being replaced by “we can’t power it.” This isn’t speculative. It’s already happening in planning documents.
For founders: edge AI and local inference models deserve a hard second look. As cloud compute pricing rises under power constraints, small models that run on user devices gain a structural cost advantage. The open-source AI community’s push toward compact multimodal reasoning systems isn’t just a technical preference. It’s a response to resource reality.
For everyone else: pay attention to data center proposals in your region. This isn’t an abstract tech industry concern. It can directly affect your water bill, electricity rates, and property taxes. The residents in Utah aren’t anti-technology. They’re fighting for a seat at the table when decisions are made about their resources.
The most important takeaway cuts against the dominant narrative: infinite AI scaling is a myth. Physical constraints exist, and they don’t disappear because a company raised another billion dollars or a stock price ticked up.
Where This Lands
O’Leary’s 75 percent concession in Utah may be a preview of what the entire AI industry faces over the next few years.
The past five years demonstrated exponential AI growth in the digital domain. The next five will show what happens when that growth curve collides with physical-world walls: insufficient grid capacity, water stress, escalating community opposition, and tightening regulation at every level of government.
AI development won’t stop. But the mode of development will be forced to change. More efficient models, smaller inference systems, distributed architectures instead of mega-centralized ones. These aren’t idealistic technical choices. They’re the inevitable response to resource constraints that no amount of capital can override.
Three predictions for the next three years:
At least one tech giant will publicly delay a major AI product launch because of power availability issues. Energy efficiency will displace parameter count as the primary competitive metric for AI models. Data center site selection will pivot from “close to users” to “close to power sources.”
AI’s expansion hasn’t ended, but the era of unconstrained acceleration is over.
Physical reality is applying the brakes, and those brakes are harder than anything any regulator could impose.
Stay updated with our latest AI insights
