I’ve been thinking a lot about infrastructure lately. Not the romantic kind that gets headlines, but the boring, physical kind that makes everything else possible. Right now, the tech industry is having a massive collective realization: you can’t run cutting-edge artificial intelligence models on dreams and venture capital. You need power. Lots of it. And we’re discovering that local communities, energy grids, and apparently even international adversaries have thoughts about that.
The numbers are staggering enough to feel almost abstract. When fully completed, the planned hyperscale data center in Box Elder County will use 9 gigawatts of power. That’s more than double what the entire state of Utah uses right now. Let that sink in for a moment. One facility. One company’s ambition. More than half the state’s total consumption.
This isn’t an edge case anymore. It’s the new normal.
When Innovation Meets Reality
Here’s what fascinates me about this moment: we built the artificial intelligence boom on the assumption that scale was a software problem. Bigger models, more data, more compute. We got really good at that part. Nobody really wanted to think about where the electrons actually come from.
Now we’re dealing with the consequences. Donna Collins lives about 20 miles from where Meta’s biggest data center is being built in North Louisiana. Her family has owned that land for five generations. “It’s all I’ve ever known as a home. It’s quiet. It’s rural. It is beautiful,” she said. “We can’t imagine the changes that are coming.”
That’s not a quote from an environmental extremist. That’s a person whose literal backyard is about to transform because the tech industry decided it needed to build faster. And she’s not alone. A recent poll shows that 47 percent of voters in Georgia oppose the multibillion-dollar data center developments coming to their state. Republicans and Democrats agree on this. That’s the bipartisan agreement nobody wanted.
The thing that gets me is how genuinely blindsided people seem when this happens. Tech companies show up with economic promises. Jobs! Tax revenue! Progress! And then the power bills start rising, construction tears through agricultural communities, and suddenly you’ve got a political firestorm. Now companies are scrambling to sign “rate payer protection pledges” with the Trump administration, promising to pay for their own power infrastructure upgrades.
It’s essentially an admission that we knew this was going to be a problem and decided to handle it with PR instead of foresight.
The Energy Grid is Screaming
Let me be direct: the US power grid was not designed for this. Winter Storm Fern swept across 34 states and left hundreds of thousands without electricity. Wholesale electricity prices in Virginia, which has more data centers than any other state, soared. This is going to keep happening, and each time it does, data centers will make it worse.
NV Energy stopped selling power to a small utility serving 49,000 customers in Lake Tahoe because data center requests are driving a tripling of expected peak power demand. They literally couldn’t keep the lights on in a tourist community because the power was being allocated elsewhere.
The Energy Information Administration is launching pilot surveys to understand how much energy data centers actually consume. This shouldn’t be surprising. This shouldn’t be something we need to measure for the first time in 2026. But here we are, and senators like Elizabeth Warren and Josh Hawley are pushing for mandatory annual reporting requirements because the transparency doesn’t exist yet.
And as a developer working in this space? This is the stuff that keeps me up at night. We’re writing code that demands infrastructure that doesn’t exist yet, betting that someone else will figure out how to power it.
The Uncomfortable International Dimension
There’s a part of this story that feels almost surreal until you remember how the world actually works. Iran’s Revolutionary Guard published a video threatening OpenAI’s planned Abu Dhabi data center, claiming they’d carry out the “complete and utter annihilation” of US-linked energy and technology infrastructure in the region if the US followed through on threats to attack Iranian power plants.
OpenAI’s $500 billion Stargate project sits in a region where geopolitical tensions are escalating. Energy infrastructure has become an explicit leverage point in international conflicts. We’ve moved past the point where data centers are just infrastructure decisions. They’re now strategic assets in actual warfare.
Elon Musk responded to this exact problem by announcing he was merging SpaceX and xAI to build data centers in space. His reasoning: terrestrial data centers require immense power and cooling at great expense to the environment and with significant community opposition. So the solution is to put them in orbit.
I’m not even sure how to process that proposal. On one hand, it’s hilariously Musk. On the other hand, it tells you how serious the problem has become. We’re running out of places to put data centers on Earth.
The Legal and Regulatory Backlash
The NAACP is suing xAI to block the Colossus 2 data center project outside Memphis, Tennessee, claiming the project is operating 27 gas turbines without an air permit and violating the Clean Air Act. “By looking to evade clear air laws to operate dirty turbines that emit pollution and known carcinogens, these companies are following a shameful, familiar pattern: asking Black and frontline communities to bear the toxic brunt of ‘innovation,’” said Abre’ Conner, NAACP Director of Environmental and Climate Justice.
That’s not fringe activism. That’s a major civil rights organization using the legal system to block what they see as environmental racism, and they’ve got a point. More than a third of new gas power generation in the US is expected to directly power data centers. Gas-fired power in development globally rose 31 percent in 2025, with almost a quarter of that new capacity in the US.
We’re literally building a new fossil fuel infrastructure to power AI. New York’s considering a three-year pause on new data center construction. A Data Center Proposal Tracker exists across 18 states to keep tabs on where new facilities are planned. Communities are organizing. Regulators are paying attention.
What This Means for Builders
Here’s my honest take: if you’re building AI applications or infrastructure, you need to start thinking about the physical implications of what you’re building. Not as a compliance checkbox. Not as a PR problem. As a fundamental constraint on what’s actually possible.
The chip industry is responding. Arm is producing its own inference CPU designed for data center efficiency. Meta is working on “multiple generations” of data center CPUs with Arm. Microsoft is exploring high-temperature superconductors that could reduce transmission losses and the physical footprint needed for power infrastructure.
These aren’t nice-to-have optimizations anymore. They’re competitive requirements. Companies that can build efficient inference systems will have economic advantages over those that can’t. The cost of power is becoming a direct limiting factor on AI capability, not just a business expense.
And there’s a shadow hanging over all of this. As geopolitical tensions rise and nations view data center infrastructure as legitimate military targets, you have to wonder about resilience and redundancy. Can your AI application survive if a key data center goes offline? Should it need to?
The future of AI infrastructure isn’t going to be determined by model architecture or training efficiency anymore. It’s going to be determined by whether we can actually build the power infrastructure to support these systems, whether communities will tolerate the environmental impact, and whether the political will exists to continue this trajectory.
We’re at an inflection point, and I think we’re only starting to realize how constrained we actually are.