THE MISSING SEAT • DEMOCRACY & GOVERNMENT • INFRASTRUCTURE

The Cloud Has an Address. Somebody Else Is Paying the Bill.

AI feels weightless on a screen. The buildings behind it are anything but.

We call it “the cloud,” which makes the whole thing sound almost magical. Ask a question, store a photo, stream a movie, run an AI model and the answer appears. But the cloud has an address. It sits on land. It connects to power lines. Some facilities use water for cooling. Somebody builds the substations, transmission and generation needed to keep the machines running around the clock.

That does not make data centers bad. They support services people and businesses actually use, and the growth of artificial intelligence is creating real demand for more computing capacity. The problem starts when we talk about the economic benefits as though the physical costs disappeared somewhere between the server and the screen.

The newest federal numbers make the scale hard to dismiss. Lawrence Berkeley National Laboratory’s 2026 update estimates that U.S. data centers could use about 649 terawatt-hours of electricity in 2030 in its reference case, roughly 11.8 percent of total U.S. electricity use. Its broader scenarios range from 521 to 843 terawatt-hours, or about 9.5 to 15.3 percent. Those are projections, not guarantees. They depend on how quickly equipment is deployed, how efficiently it operates and how demand develops. But even the uncertainty is telling us something: utilities and communities are being asked to plan for very large new loads. [1]

That is where the bill gets interesting. Electricity infrastructure is not free, and a new data center can require investments that last much longer than the first contract. Berkeley Lab’s 2026 review of electricity-rate designs for large customers says regulators and utilities have to manage risks including insufficient energy supply and underused infrastructure investments that can affect other customers. In Regular Joe language: if a utility builds expensive capacity for a giant customer and the deal does not protect everybody else, ordinary customers can be exposed to costs they did not create. [2]

This is not an argument that every data center automatically raises everybody’s electric bill. Rate structures differ. Utilities and regulators can require large customers to make minimum payments, commit for longer periods, provide financial security or directly cover infrastructure costs. That is precisely the point. The outcome depends on the rules written before the facility plugs in.

Water deserves the same honesty. Some data centers use evaporative cooling and can consume substantial amounts of water. Others use closed-loop or air-cooled systems that can sharply reduce direct water consumption. Berkeley Lab’s earlier national modeling estimated direct U.S. data-center water consumption at roughly 66 billion liters in 2023 and projected higher use under several 2028 scenarios. DOE has also emphasized that newer closed-loop designs can reduce water demand substantially. So “data centers use huge amounts of water” is too crude. The questions that matter are local: What cooling system is being built? What water source will it use? How much will it consume during the hottest and driest periods? What happens if the forecast is wrong? [3][4]

Then there are the tax deals. States compete for data centers because the projects can bring investment, construction work, tax base and technology infrastructure. But many states also offer sales-tax exemptions or other incentives. Good Jobs First, a subsidy watchdog, reported in 2026 that 36 states had subsidies specifically designed for data centers and that disclosure remained weak in many of them. Its estimates are advocacy research, not a federal audit, so they should be treated as such. But the underlying accountability question is legitimate regardless of the exact total: if taxpayers are giving up revenue to attract a project, the public should be able to see what was promised in return and whether the promise was kept. [5]

This is where I think the conversation gets backwards. We keep asking whether America should build the infrastructure needed for AI. Of course it should build infrastructure for technologies people are actually going to use. The better question is whether the people who own the demand are paying a fair share of the infrastructure needed to serve it.

If a company needs a new substation, who pays? If transmission has to be expanded, who carries the risk if the customer leaves or uses less power than expected? If a county gives up tax revenue, what measurable benefit is guaranteed? If a facility will use local water, what happens during drought restrictions? If residents are told the project will create jobs, how many permanent jobs are actually in the agreement rather than in the press release?

Those are not anti-technology questions. They are basic business questions. If I open a restaurant and need special equipment, I cannot simply tell the people next door that my expansion is good for the neighborhood and therefore they should quietly absorb my costs. Scale does not make the accounting disappear.

Who benefits? Data-center operators, cloud and AI companies, customers who use the services, property owners, construction firms, utilities and governments can all benefit in different ways. Communities can benefit too through investment, employment and tax revenue. None of those benefits should be assumed; they should be measured.

Who pays? Companies clearly pay enormous construction, equipment and operating costs. But communities may also contribute through tax exemptions, public infrastructure, land-use decisions, water allocations or utility investments. Whether those public contributions are reasonable depends on what the community receives and how financial risk is divided.

Who is missing? Too often it is the ordinary ratepayer or resident who learns the important details after officials, utilities and developers have already decided what the project is worth. The missing seat belongs to the people whose electric system, water system, tax base and neighborhood are part of the deal.

The recipe is not “stop building data centers.” It is much simpler: make the deal visible before the public is committed to it. Utilities should publish the rules that determine who pays for new large-load infrastructure. Governments should disclose tax incentives, expected permanent employment and performance requirements. Water-intensive projects should disclose the cooling method, expected consumption and drought plan. Regulators should make sure ordinary customers are not quietly guaranteeing investments built primarily for one very large user.

And companies that believe these projects are good for communities should welcome that accounting. A genuinely good deal should survive daylight.

The cloud is not floating above us. It is sitting somewhere, plugged into something, using resources somebody manages and infrastructure somebody paid to build.

Before we celebrate the next billion-dollar announcement, there is one receipt worth demanding: Who gets the benefit, who carries the risk, and who pays if the forecast is wrong?

Ally = Action. 🪑

Sources and Receipts

Verification notice: Final evidence review completed October 7, 2026. Electricity figures are projections, not guaranteed outcomes. Water use varies materially by cooling technology, climate and facility design. Good Jobs First is an advocacy organization; its subsidy estimates are identified as such rather than treated as government findings. No claim is made that every data center raises residential rates or creates a net public loss.

Return to all editorials