The Cloud Isn't a Cloud

You type a question into ChatGPT. Maybe you ask it to summarize a document, plan a vacation, explain your child's math homework, or help write an email. A few seconds later, an answer appears.

The entire interaction feels almost weightless. No machinery moves in front of you. No delivery truck arrives. No factory starts humming in your kitchen. You type words into a box, and words come back.

That's one of the remarkable things about the digital world: it has trained us to forget that it is physical.

We store photographs in "the cloud." We stream movies from nowhere in particular. Our email, banking records, work files, music, social media, and increasingly our artificial intelligence seem to exist somewhere beyond the devices in our hands.

But there is no cloud. There are buildings.

Inside those buildings are rows upon rows of computers. Those computers contain specialized chips. The chips consume electricity and generate heat. The heat has to be removed. The buildings need connections to enormous electrical grids and high-speed fiber networks. Depending on how they're cooled, they may consume substantial amounts of water.

And as artificial intelligence becomes more powerful—and we ask it to do more—the infrastructure behind our seemingly invisible digital lives is becoming much harder to ignore.

In fact, one of the biggest technology stories of the next few years may have surprisingly little to do with software. It may be about electricity, water, land, and who pays for all of it.

The Cloud Was Always a Metaphor

The term "cloud computing" is wonderfully effective marketing. It suggests something light, distributed and almost magical.

Upload a photograph to the cloud and it simply goes…up there. Of course, your photograph actually goes somewhere very specific. It is stored on a physical computer in a physical facility operated by a company such as Amazon, Microsoft, Google, Apple or one of thousands of other providers.

Those facilities are data centers.

They have existed for decades, and they aren't exclusively—or even primarily—an AI invention.

Every Netflix movie you stream, Google search you make, website you visit, cloud document you edit and online purchase you complete depends on computing infrastructure somewhere.

For most of the internet era, however, ordinary users had little reason to think about it. That is changing because the scale is changing. The U.S. Energy Information Administration now expects American electricity consumption to reach record levels in 2026 and 2027, with data-center development among the major drivers of that growth. 

AI isn't creating the physical internet. It's making the physical internet much, much bigger.

Why AI Changed the Equation

Artificial intelligence requires enormous amounts of computing power.

First, an AI model has to be trained.

Very simply, training is the process through which a model analyzes enormous amounts of data and adjusts itself to become better at recognizing patterns and producing useful outputs. Training the largest models can require huge clusters of specialized processors working together.

But the infrastructure requirement doesn't end when training does.

Every time millions of people ask those models questions, generate images, analyze documents, write computer code or eventually send AI agents off to complete tasks, computers have to process those requests.

That's called inference.

So the AI infrastructure challenge has two sides:

Build increasingly sophisticated models. And then: Run those models for potentially billions of users and devices.

The resulting electricity projections are striking.

A 2026 update from Lawrence Berkeley National Laboratory estimates that data centers could account for about 11.8% of total U.S. electricity consumption by 2030, with plausible scenarios ranging from 9.5% to 15.3%. The Electric Power Research Institute produces an even wider range, projecting 9% to 17%, compared with roughly 4% to 5% today. 

Those are forecasts, not destiny. AI chips could become more efficient. Models could require less computing. Companies could build more power. Demand could grow more slowly—or more quickly—than expected.

But the direction is difficult to miss. Computing is becoming a major energy story.

What Actually Happens When You Ask AI a Question?

Let's return to that ChatGPT prompt.

You type: "Help me plan a five-day trip to Italy."

The request travels over the internet to computing infrastructure running the AI model.

Inside a data center, specialized processors perform an extraordinary number of calculations to interpret your request and generate a response.

Those processors require:

• Electricity to operate.

• Servers to house them.

• Networking equipment to move information.

• Cooling systems to remove the heat they generate.

• Buildings to contain all of that equipment.

• Fiber networks connecting the facility to the wider internet.

• And an electrical system capable of supplying enormous amounts of reliable power.

Your individual question isn't the problem.

The scale is.

Multiply seemingly trivial digital interactions by millions of users, thousands of companies, always-on services, increasingly capable AI models and eventually fleets of AI agents performing work continuously.

Suddenly the invisible world starts becoming very physical.

Why Are These Giant Buildings Appearing Everywhere?

Data centers can't simply be dropped anywhere. Large facilities need access to tremendous amounts of reliable electricity.

• They need high-capacity fiber connections.

• They need land.

• They need roads and other infrastructure.

• They need favorable zoning and permitting.

• They may need access to water.

And because speed matters for many digital services, geography can matter too.

This helps explain why data centers tend to cluster. Northern Virginia provides perhaps the best example.

The region has become one of the world's largest concentrations of data centers, aided by its history as an internet-networking hub, extensive fiber connectivity, available infrastructure and proximity to major population and government centers.

The scale is now large enough to reshape Virginia's electricity system. The U.S. Energy Information Administration reported this year that Virginia's commercial electricity sales increased by nearly 30 million megawatt-hours between 2019 and 2025, with data centers driving much of that growth. 

If you live in Northern Virginia, then, the physical infrastructure behind the digital world isn't somewhere far away. It's increasingly part of the landscape.

And That Creates a New Question: Who Pays?

Suppose a technology company wants to build an enormous new data center.

The company obviously pays to construct its building and purchase its computers. But supplying that building with electricity can require much more.

The utility might need a new substation. Transmission infrastructure may need upgrading. Additional generating capacity may eventually be required.

The regional electrical grid may need investments to ensure enough power remains available during periods of peak demand.

Those projects can cost enormous amounts of money. And electricity infrastructure has traditionally been financed partly by spreading costs across large numbers of customers.

That's where a technology story becomes a kitchen-table story:

If a data center requires billions of dollars of additional electrical infrastructure, who should pay for it?

The company creating the demand?

The utility?

The government?

Or everyone who receives an electric bill?

That question is no longer theoretical.

In the PJM electricity market serving 13 states and Washington, D.C., the independent market monitor calculated that existing and forecast data-center demand accounted for about 9% of wholesale power prices through July 2026. The monitor also identified data-center load growth as a major factor in tightening capacity conditions. 

The issue has now reached Congress. A bipartisan bill called the Ratepayer Protection Act would require state utility regulators to consider strategies designed to make data centers bear the costs they create rather than shifting those costs onto other electricity customers. The House Energy and Commerce Committee advanced it 52–0 in July, and House leaders are preparing a floor vote. 

In other words, the AI boom has become large enough that Congress is debating whether your electric bill should help pay for it.

Electricity Isn't the Only Resource

Then there's water.

Computers generate heat. Putting thousands of powerful processors together in a building and removing that heat becomes a major engineering challenge.

Different data centers solve the problem differently. Some cooling systems use relatively little water on site. Others rely more heavily on evaporative cooling. Climate, facility design, technology and location all matter.

That's why claims such as "one AI question uses exactly X amount of water" should be treated cautiously. There isn't one universal number that accurately describes every model running in every data center.

But the underlying issue is real.

A recent North Carolina State University review notes that some large facilities can use millions of gallons of water for cooling, and U.S. data centers directly consumed an estimated 17 billion gallons for cooling in 2023. Their total water footprint can be considerably larger because producing electricity and manufacturing semiconductors can themselves require water. 

And here again, location matters.

A gallon of water consumed somewhere with abundant water resources isn't necessarily equivalent to a gallon consumed in a drought-prone region.

The useful question isn't simply: "Do data centers use water?"

They do.

It's: "How much water does this particular facility use, where does it come from, and what does that mean for the community around it?"

That's a much harder question—and a much more useful one.

The Community Is Part of the Technology Stack

For years, discussions about AI infrastructure tended to focus on companies.

• Who has the best chips?

• Who has the biggest model?

• Who has the most computing power?

But communities are becoming another important participant.

A proposed data center can raise questions about land use, electricity, water, transmission lines, backup generators, noise, tax incentives and local infrastructure.

The potential benefits can also be significant.

Data centers represent enormous capital investments. They can generate substantial local tax revenue. Construction creates jobs. New infrastructure can bring economic development.

But communities increasingly want to understand the bargain. And governments are responding.

Just this week, Massachusetts established new requirements for large data centers that explicitly address energy, water, environmental effects and community benefits. The state's policy says residents and businesses should not bear the costs and burdens of new data-center development without sharing in its benefits. 

That's a revealing shift. The data center is no longer being treated merely as a building full of computers. It is increasingly being treated as major infrastructure.

There Is Another Problem: What If We Build Too Much?

There's an interesting complication hiding inside all these forecasts.

Everyone is racing to build.

• Technology companies want computing capacity.

• Developers propose data centers.

• Utilities receive enormous requests for electricity.

• Grid operators plan for future demand.

But not every proposed data center will necessarily be built.

Developers can submit requests in multiple locations while deciding where to construct a facility. Plans change. Financing changes. Technology changes.

Reuters recently found that very large electricity requests—mostly associated with data centers—across parts of the Midwest, Mid-Atlantic and South exceeded 700 gigawatts, (by contrast, in Back to the Future the Delorean only needed 1.21 gigawatts for the flux capacitor) more than ten times estimates of current U.S. data-center power use. Texas has begun scrutinizing proposed connections partly to determine which projects represent genuine future demand. 

That creates a strange problem. Underbuilding infrastructure could leave the country without enough electricity to support AI growth. But overbuilding based on projects that never appear could leave somebody paying for expensive infrastructure that wasn't needed.

Once again, the important question becomes: Who carries the risk?

We Should Admit Something: We Want What Data Centers Provide

It would be easy to end the story here.

AI companies are building enormous data centers. They consume electricity and water. Communities are pushing back. Case closed.

Except that leaves out one important participant. Us.

• We stream movies instead of buying DVDs.

• We keep tens of thousands of photographs online.

• We expect our files to appear instantly on every device.

• We conduct video calls with people around the world.

• We store documents in the cloud.

• We ask search engines billions of questions.

And now we're beginning to ask AI to write, research, translate, analyze, create images, generate video, write software and help us make decisions.

Soon we may expect AI agents to monitor information and perform tasks for us around the clock. We don't particularly want data centers.

We want everything data centers make possible.

That's an important distinction because it makes the debate much more complicated than technology companies versus local communities. The infrastructure exists because digital services have become infrastructure for our lives. AI is simply accelerating the demand.

So Can We Build This Better?

The meaningful debate isn't whether America should stop building data centers. That's neither realistic nor necessarily desirable.

The more useful questions are about how we build them.

• Should companies creating enormous new electricity demand pay for the infrastructure required to serve them?

• Should data-center operators disclose more information about their electricity and water use?

• Should facilities be encouraged to locate where energy and water are more abundant?

• Could data centers reduce their electricity use temporarily when the grid is under extreme stress?

• Could more computing be shifted to times when electricity is plentiful?

• Can chips, models and cooling systems become dramatically more efficient?

• Should new facilities bring their own generation—or help finance new clean or reliable power?

• And what should communities receive in exchange for hosting infrastructure that serves people far beyond their borders?

None of those questions requires deciding whether AI is "good" or "bad."

They require acknowledging something much more basic. Artificial intelligence has infrastructure. And infrastructure requires choices.

The Digital World Was Physical All Along

For most of the internet era, we've been encouraged to imagine the digital and physical worlds as separate things. The physical world contains roads, factories, power plants and buildings. The digital world contains apps, websites, streaming services and clouds.

But that distinction was never really true.

Every photograph stored online occupies physical computing infrastructure. Every movie streamed requires servers and networks. Every AI response depends on machines performing calculations somewhere.

The cloud was always made of concrete, steel, silicon, copper, fiber, electricity and water.

Artificial intelligence is simply making that reality much harder to ignore. And perhaps that's useful. Because as AI becomes a larger part of our lives, understanding it will require looking beyond the chatbot on our screen.

We'll need to understand the chips.

The data centers. The electrical grid. The water. The communities. And the economic choices connecting all of them.

The digital future may arrive through our screens. But somebody still has to build—and power—the world behind them. 

Stay ahead of the curve with the latest in Web3 culture and innovation. Subscribe to Hashed Out for exclusive insights, case studies, and deep dives into the decentralized future.

Help Grow Hashed Out And Get Rewarded With Premium Content & Merchandise

If you believe in a more open, fair internet — help us build it, one reader at a time.

Web3 adoption starts with curiosity. Share Hashed Out with someone who’s ready to explore.

You’re not just sharing a newsletter — you’re inviting someone into the future of digital life.

Refer 3 friends and unlock premium content. The more you share the more rewards you unlock, including Hashed Out mugs or tote bags, and exclusive community memberships.

RELATED ARTICLES:

• When AI Meets the Blockchain: Smarter Systems, Fairer Foundations

• When AI Goes to Work

• Can You Trust AI?

• CHECK OUT ALL OUR PREVIOUS EDITIONS