Data Centers in America: What You Should Know

If 2026 wasn’t already weird enough, I found myself agreeing with Tucker Carlson last week. Didn’t have that on my bingo card.

He was interviewing Kevin O’Leary about the Stratos Project, a proposed AI data center in Box Elder County, Utah. Forty thousand acres. More electricity than the entire state currently uses. Billions of gallons of water per year in a region already in severe drought. Approved, with no public referendum, by three county commissioners.

Carlson’s question was simple: why are taxpayers subsidizing infrastructure that will primarily benefit some of the richest companies in the world? O’Leary’s answer was, essentially, “Welcome to America, buddy.”

I’m not linking arms with Tucker Carlson. His concerns are about surveillance and government overreach, which is a different argument than the one I’m making. But when someone I rarely agree with is asking the right accountability question, I’m going to say so.

Let me take the counterargument seriously before I tell you why it's aimed at the wrong target.

Now. I know some of you have been carrying a version of this question yourselves, and it’s worth taking seriously before I get to the part that frustrates me.

Individual creatives using AI tools have been absorbing a steady stream of environmental critique. You’ve probably seen it. Maybe you’ve felt it. Every time you use an image generator or ask a language model to help you draft something, there’s an implication that you’re personally degrading the planet.

That framing deserves an honest answer.

User demand is part of the equation. Aggregate choices matter across any industry. That’s fair, and I’m not going to wave it away. That’s why boycotts can work. Why Target Corporation’s shareholder value has been declining since they made their bed after the election.

But here’s what’s also true.

Amazon, Alphabet, Meta, and Microsoft are collectively committing somewhere between $650 billion and $700 billion to AI infrastructure in 2026 alone. That figure is nearly double what they spent in 2025. It’s larger than the GDP of Sweden. Amazon’s CEO Andy Jassy put it plainly in his April shareholder letter: “We’re not going to be conservative in how we play this.”

That buildout is running ahead of current demand. It’s driven by enterprise contracts, government compute, and a competition between companies who believe whoever controls the infrastructure wins the next decade. Kevin O’Leary’s project doesn’t even have any signed tenants yet. That’s how ridiculous this has gotten.

So let me be blunt: a quilter in Naperville opting out of Midjourney wouldn’t register in that calculation. I wish it were otherwise. It isn’t.

Wait. Isn't Anthropic building data centers, too?

Yes. And that’s worth understanding clearly, because it’s not the same thing.

There are two different kinds of AI infrastructure, and they get lumped together in ways that muddy the conversation.

The first is training infrastructure. This is where a model like Claude or Gemini learns in the first place. Training is extraordinarily compute-intensive, runs for weeks or months, and requires massive, centralized facilities with enormous power draw. Amazon’s Project Rainier, built specifically for Anthropic’s training workloads, cost roughly $11 billion. That’s a large facility by any measure. Anthropic also announced a separate $50 billion infrastructure partnership with FluidStack in late 2025, with data centers coming online in Texas and New York. These are real, significant investments.

The second is inference infrastructure. This is what runs when you open Claude or Gemini and type something. Inference needs are distributed, lower-latency, and smaller in individual footprint. Industry guidance suggests inference facilities typically need 10 to 30 acres. They’re built closer to population centers, not in remote rural areas where land is cheap and oversight is thin.

The Stratos Project isn’t being built to run your chatbot queries. It’s being built for the compute race between the four largest technology companies in the world, competing for dominance in enterprise AI and government contracts.

The scale is categorically different: 40,000 acres, 9 gigawatts of on-site natural gas generation, capacity to double Utah’s entire electricity usage. That’s training infrastructure at a scale the industry hasn’t attempted before.

The distinction matters because conflating them is how individual AI users end up shouldering guilt that belongs to a different conversation entirely.

Utah is not an outlier. It's a preview.

Utah Clean Energy’s preliminary analysis found that the Stratos Project, at full build-out, would consume twice Utah’s peak electricity demand and increase the state’s CO2 emissions by roughly 75 percent. It would use approximately 2 billion gallons of water per year to run the natural gas engines on-site. The Great Salt Lake is already approaching record-low levels. Warmer winters have reduced snowpack. The project was approved in five months, partly through a zoning authority that was created to support a military base in a different county.

When the commissioners voted, residents chanted “Shame.” Nearly 4,000 people filed protests with the Utah Division of Water Rights. A referendum effort started within days.

That’s not the story of a community that didn’t care. That’s the story of a community that wasn’t asked.

So where does the accountability question actually belong?

I’m not here to tell you to use or not use AI tools. That’s yours to decide, with your own values and your own judgment. Intentional use matters. Not because every creative’s choice shifts the infrastructure calculus, but because reflexive use serves no one, including you. If you’re reaching for these tools out of pressure or obligation rather than actual purpose, that’s worth examining regardless of any environmental argument.

But the guilt that’s been aimed at individual creatives is not the pressure that changes what’s getting built.

That accountability belongs to investors, regulators, and the communities where these facilities are going in the ground. It belongs to elected officials using zoning workarounds to approve billion-dollar projects without a public vote. It belongs to companies whose shareholder letters describe infrastructure spending in terms of competitive positioning, not public benefit.

I don’t mean to say this like it is some sort of consolation. It’s just where the actual leverage is.

When O’Leary replied “Welcome to America, buddy” to a question about why taxpayers should subsidize infrastructure for some of the richest companies in the world, he wasn’t defending anything, either. He was admitting that this is exactly what we all see it to be.

The accountability question belongs somewhere specific.

I want to make sure we as creatives are clear on exactly where that is.