America Just Discovered the Cloud Has Buildings
Scott Bessent says the AI industry has done a horrendous job explaining data centers to the American people. He's right. But the failure is bigger than bad PR.
There is something inherently funny about organizing a protest against data centers from an iPhone. You pull out the phone, connect to the internet, start the WhatsApp group, build the Facebook page, design the signs in Canva, ask ChatGPT to make the slogans better, schedule the meeting in Google Calendar, start a GoFundMe, Venmo somebody for the bullhorn, Uber to the protest, livestream it to TikTok and then hold up a sign that says:
WE DON'T NEED DATA CENTERS.
I'm sorry. That's funny.
It's also not a serious rebuttal to the people holding the sign.
Because somebody can use cloud services every day and still have perfectly legitimate questions about a 500-megawatt industrial project being built down the road. Where is the power coming from? Who is paying for the new transmission? What does it do to local rates? What kind of cooling system is being used? How much water does that specific design actually consume? What does the community get back for hosting it?
Those are reasonable questions.
The problem is that we've somehow managed to turn the entire subject into two people screaming past each other.
One side sees giant warehouses full of GPUs inhaling electricity and water while Big Tech collects tax incentives. The other sees critical national infrastructure, economic growth and the physical foundation of the AI economy. Both of those things can be true.
Treasury Secretary Scott Bessent got surprisingly close to the real problem this week when he said everyone needs to take a deep breath about AI data centers, and then immediately put a good chunk of the blame on the AI industry itself.
His complaint was that the industry has done a "horrendous job" explaining what it's doing to the American people. He also made the larger argument: America intends to be an AI superpower, and the benefits of that should accrue to Americans.
I think he's basically right. But I also think the communications failure is stranger than he makes it sound. Because the technology industry has actually become incredibly good at explaining AI. They're spending fortunes doing it.
OpenAI acquired TBPN. OpenAI and Anthropic are working with creators and influencers. Microsoft is pushing Copilot education. Google is putting Gemini everywhere it can. Meta is weaving AI into products already used by billions of people.
Then you move one layer up the stack and get companies like Base44 essentially telling everyone, Forget learning to code. Tell the machine what you want and build it. Base44 ran a Super Bowl commercial.
A freaking Super Bowl commercial for vibe coding.
So this is not an industry quietly waiting for adoption. They're flooring it.
Use AI at work. Use it at school. Build software with it. Make videos with it. Research with it. Automate your business with it. Give agents jobs. Put intelligence into every workflow you can find.
And I happen to think a lot of that is fantastic. But there's a physical consequence to all of it.
Compute demand goes up.
That connection is almost completely missing from the public conversation.
ChatGPT feels like software. Claude feels like software. Gemini feels like software. Base44 feels like software. Then somebody announces a 600-acre campus with enough electrical demand to make an engineer stare quietly into the distance, and suddenly everybody asks where the hell that came from.
It came from the same place.
The industry did a very good job explaining the product.
It forgot to explain the factory.
The cloud was never actually a cloud
We've spent twenty years deliberately hiding computing infrastructure from ourselves. Your email lives somewhere. Your bank records live somewhere. Your photos live somewhere. Netflix, TikTok, Amazon, Uber, DoorDash, Zoom, Salesforce, Google Docs, Microsoft 365 and the thousand invisible services running the average company all live somewhere.
We just put a little cloud icon on the screen and stopped thinking about it.
There was never a cloud. There were buildings. Buildings full of servers, batteries, fiber, electrical switchgear, cooling equipment, generators and enormous amounts of capital. We got extraordinarily good at abstracting all of that away. Then AI arrived and demanded so much computation that the abstraction started breaking.
The buildings got bigger. The substations became visible. The power requirements became visible. The cooling became visible. The politics became visible. America suddenly discovered that the cloud has an address.
And now we're arguing about the address. That part is healthy.
What isn't healthy is having that argument with an outdated mental picture of what's actually being built.
The data center itself is changing
One of the most frustrating parts of this debate is how quickly the technology underneath it is moving.
Take water.
You constantly hear that data centers consume enormous amounts of water, and in plenty of cases that criticism is justified. Older facilities built around evaporative cooling can consume a tremendous amount of it.
But that doesn't mean every modern facility is simply a larger version of the same thing. We are moving rapidly toward direct-to-chip liquid cooling, closed-loop systems and non-evaporative designs. Microsoft has already been deploying architectures where cooling water circulates through a closed system rather than constantly being evaporated and replaced. Oracle and others are moving in the same direction. And this is where language gets us into trouble.
"Liquid cooling" doesn't necessarily mean "using huge quantities of drinking water."
A closed loop can circulate the same coolant repeatedly. Depending on the architecture, that fluid might be water, a water-glycol mixture, a dielectric fluid or something else entirely. The important innovation isn't that somebody discovered a magical new liquid. It's that the industry is getting much better at moving heat away from increasingly dense compute without constantly consuming water to do it.
The chips are helping force that transition. Rack densities are reaching levels where traditional air cooling simply stops making sense. The amount of heat coming off modern AI hardware is pushing the entire industry toward better thermal engineering whether it wanted to move that quickly or not.
The power side is changing too.
There's no universal federal rule saying that every data center above some magic number has to build its own private power plant. Reality is messier than that.
But the policy direction is becoming pretty obvious: if you're going to show up with the electrical appetite of a small city, increasingly you're expected to bring new generation, contract for it, finance the transmission and substations you require, or otherwise make sure normal ratepayers aren't simply handed the bill.
Which seems reasonable.
If you're bringing the appetite of a small city, bring lunch.
That doesn't mean every new data center is environmentally wonderful.
Some aren't.
Some will still use too much water. Some will be built in stupid places. Some will rely on bad power arrangements. Some developers will overstate the number of permanent jobs. Some communities will get offered lousy deals.
That's exactly why the conversation needs more specificity, not less.
Instead of arguing about some mythical average data center, talk about the actual project being proposed. Explain where the power is coming from and who is paying for the upgrades. Explain the cooling architecture and what its real water consumption looks like in that climate. Be clear about the tax incentives and equally clear about what flows back to the county.
Separate the thousands of people who may build the facility from the smaller permanent workforce that stays afterward. Explain what happens as the hardware ages, how the site gets refreshed and what useful infrastructure remains behind.
People can handle nuance.
What they don't like is feeling as though a giant industrial project arrived with a glossy brochure and half the numbers missing.
There is another strange thing happening here
We are treating the construction boom as though somebody suddenly developed a fetish for enormous concrete buildings. That isn't what's happening.
We are changing what computing is.
For the last twenty years, most of the growth was about moving software and storage into the cloud. Now we're adding intelligence. And intelligence is expensive to compute.
The difference between opening Gmail and asking a frontier model to reason through a complex problem is not trivial underneath the hood. Neither is generating video. Neither is running autonomous agents continuously. Neither is having millions of businesses build AI into workflows that previously required almost no machine intelligence at all.
And we are not doing this reluctantly.
The entire industry is actively trying to increase consumption. That isn't sinister. It's what every technology industry does when it believes it has created something broadly useful. Apple wanted everyone carrying a smartphone. Microsoft wanted a computer on every desk. Amazon wanted businesses moving into AWS.
Now the AI industry wants intelligence available everywhere. But you can't simultaneously tell the public that AI should become ubiquitous and then act surprised that ubiquitous AI requires a staggering amount of physical infrastructure. Those are the same story. We just marketed one of them and zoned the other.
None of this means communities should shut up and take it
This is where the pro-data-center argument can become just as lazy as the anti-data-center one.
Saying "national security" doesn't give every developer a magic wand.
America needing AI infrastructure doesn't mean every proposed site is a good one. If the economics only work because residential ratepayers subsidize the power, fix the deal. If the cooling design makes no sense for the local water situation, change the design or change the location.
If the tax incentives are enormous and the community receives almost nothing in return, renegotiate them. If a developer is selling thousands of permanent jobs when most of those jobs disappear after construction, call that out. If the project puts all of the externalities in one county while almost all of the upside flows somewhere else, that's a legitimate political problem.
The answer to those problems isn't "data centers are bad." It's build better data centers and structure better deals. And that matters because stopping new construction doesn't somehow cause the old infrastructure to improve itself.
We already have legacy facilities built around older cooling systems, older power systems, less efficient hardware and much lower compute density. At some point, workloads have to move. The new infrastructure has to exist before the old infrastructure can be retired. That's not unique to data centers. We understand that basic transition everywhere else. You don't modernize infrastructure by refusing to build its replacement.
In fact, AI may end up accelerating the modernization of the entire data-center industry simply because the old architecture increasingly can't handle what we're asking it to do.
And then there's China
This part needs some care because it's very easy to turn a legitimate issue into nonsense.
There's documented evidence of China-linked influence operations amplifying anti-data-center narratives in the United States.
That doesn't mean China created the opposition. It doesn't mean everyone holding a protest sign is secretly reading from Beijing's talking points.
And we definitely shouldn't use foreign influence as an excuse to dismiss legitimate American concerns. But influence operations don't usually work by inventing a grievance from scratch.
They find a real one. Then they amplify it. Turn the temperature up. Push people farther apart. Make compromise harder. And from China's perspective, the strategic logic here is obvious.
The AI race isn't just about models. It's about chips, talent, energy, manufacturing, capital, fiber and ultimately the ability to power huge amounts of computation.
If you can't slow America's models directly, slowing America's ability to build the infrastructure underneath them is still useful. That doesn't make the backlash fake. It makes the backlash strategically relevant.
Those two ideas are perfectly capable of existing at the same time.
Which brings us back to Bessent
I don't think the industry needs another campaign telling Americans that AI is awesome.
They're already doing that. What it needs is a much better explanation of the bargain underneath it. If America wants to become an AI superpower, what does that physically require? How much infrastructure are we actually talking about? How is that infrastructure going to be powered? What does the transition from legacy facilities to modern ones look like? How do we keep residential customers from subsidizing enormous industrial loads? How do host communities participate in the upside rather than simply absorbing the inconvenience? And what exactly happens if we decide not to build it?
That last question gets strangely little attention. Because "we don't need data centers" isn't really one of the options anymore. Not while we're simultaneously asking OpenAI, Anthropic, Microsoft, Google, Meta and a thousand companies built on top of them to put increasingly capable intelligence into everything we do.
That demand has to run somewhere. And this is where I think the entire debate eventually lands.
We should argue about where data centers go. We should argue about power, water, taxes, noise, jobs and who pays for what. We should be skeptical of bad projects and insist that the next generation of facilities is dramatically better than the last one. But the choice increasingly isn't between building data centers and not building data centers. It's between old infrastructure, modern infrastructure, and somebody else's infrastructure.
For twenty years we made computing invisible. AI made it visible again.
Maybe that isn't a crisis. Maybe it's finally our opportunity to have the infrastructure conversation we should have been having all along.
Sources
Scott Bessent — AI/data center comments https://rumble.com/v7f09vy-scott-bessent-everyone-needs-to-take-a-deepbreath-about-ai.html
Public opposition to data centers — Gallup https://news.gallup.com/poll/709772/americans-oppose-data-centers-area.aspx
Trump administration — Ratepayer Protection Pledge / “build, bring, or buy” power https://www.whitehouse.gov/ratepayer-protection-pledge/
Microsoft — closed-loop, zero-water-consumption cooling for new AI data centers https://blogs.microsoft.com/blog/2026/06/24/inside-microsofts-two-decade-push-to-cut-water-intensity-while-scaling-for-growth/
Oracle — direct-to-chip, closed-loop, non-evaporative cooling https://www.oracle.com/news/announcement/blog/closed-loop-cooling-in-oracle-ai-data-centers-2026-02-09/
OpenAI acquisition of TBPN / AI communications push https://openai.com/index/openai-acquires-tbpn/
Anthropic Claude Corps — $150M program, 1,000 fellows expanding practical AI adoption https://www.anthropic.com/news/claude-corps
Microsoft Copilot Creators Collective — creator-led AI adoption and literacy https://news.microsoft.com/en-ph/2026/08/10/the-new-media-movement-how-the-copilot-creators-collective-is-shaping-the-future-of-creativity-and-ai-literacy/
Base44 Super Bowl campaign / mass-market AI adoption https://www.wix.com/press-room/home/post/base44-to-air-its-first-super-bowl-ad
OpenAI — China-linked “Data Center Bandwagon” influence operation https://openai.com/index/disrupting-malicious-uses-of-ai-data-center-bandwagon/
Axios — China-linked activity targeting the U.S. data-center debate https://www.axios.com/2026/08/28/china-ai-data-center-backlash-bots
Broader data-center backlash / industry communications problem https://www.axios.com/2026/08/27/data-center-backlash-pr-strategy
Texas backlash, grid concerns and changing data-center policy https://www.reuters.com/legal/government/texas-republicans-turn-against-data-centers-putting-big-tech-notice-2026-09-03/
Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.
Comments