How the data center, the electric bill, the federal contract, and the surveillance state became one machine.

The Bill Arrives at Your Address

There is probably a light on somewhere in your house that you forgot about. A monitor in sleep mode. A phone charger plugged into nothing. A router blinking in a closet. The kind of small electrical draw nobody thinks about, because thinking about it changes nothing.

Somewhere in Virginia, (or New Jersery) an unmarked building the size of a regional airport is drawing the equivalent of a small city. It does not have employees on most floors. It does not have windows. It does not have a sign. It has servers, and the servers are computing something for someone in a way that requires more electricity than the entire country of Spain will use three years from now.

You will pay for it. The question is only when, and how, and whether you ever connect the bill in your hand to the building you have never seen.

The cloud lives somewhere.

That is the sentence the industry spent fifteen years trying to keep you from saying. The word “cloud” was always a marketing decision, not a technical one. Clouds are weightless and free and belong to no one. The actual infrastructure is heavy and expensive and belongs to four or five companies. It needs land. It needs power. It needs water. It needs permits. It needs the cooperation of governments that have spent the last decade competing to give it everything it asks for.

In the United States right now, data centers account for roughly half of all new electricity demand. Residential electricity prices have risen forty percent since 2021. In Virginia, where data centers consume more than one in four kilowatt-hours generated in the state, the regional grid auction price jumped 833 percent in a single year. A homeowner in Manassas opened his January electric bill and saw $281 where the previous month had been $100. He has lived in the house for forty years. The house did not change. The grid did. The grid changed because something is happening upstream of his meter that he was never asked about and cannot vote against.

He is paying for AI. He just does not know it yet, because nobody has explained the chain.

The chain runs through your statehouse.

It runs like this. At least thirty-six states have built specific subsidies for data center development.

Virginia gives up close to a billion dollars in state and local sales tax revenue every year and does not disclose to the public which companies benefit. Texas revised its annual data center subsidy estimate from $130 million to $1 billion inside the same fiscal year.

Twelve states with data center programs do not publish what they cost at all. The names of the companies receiving the money are often hidden behind project code names and shell entities, the way a real estate developer hides a tear-down behind an LLC.

On top of state subsidies, the federal government in July 2025 issued an executive order designating large data centers as projects of national security importance. That designation does several things at once. It accelerates federal permitting. It opens federal land for data center siting. It directs the Secretary of Commerce to provide loans, loan guarantees, grants, tax incentives, and offtake agreements. It tells environmental reviewers to find categorical exclusions that allow projects to proceed without the usual public process.

The day after the order was signed, the Department of Energy named four federal sites available for data center development. By October the Air Force had opened five military bases to bids. The Idaho National Laboratory, Oak Ridge, the Paducah Gaseous Diffusion Plant, Savannah River, Edwards Air Force Base.

The federal government became a real estate broker for the AI industry inside three months.

A second federal action followed in December 2025. That executive order directed federal regulators to identify state AI laws considered “obstructive” and to deny those states federal broadband funding.

The federal government was now telling states that if they passed their own rules on AI or data centers, they could lose money the federal government had previously committed to give them. State sovereignty was reframed as a problem to be solved.

When the public sector spends this much energy clearing a path for a private industry, the path is not really private anymore. It is something else. The word “subsidy” understates it. The word “partnership” overstates it. The accurate word is “capture,” but capture sounds like the conclusion of a long story when it is actually a description of where we’ve been heading for a long time.

The wires carry more than electricity.

The same companies building the data centers are not selling only computation. They are selling infrastructure.

Palantir, which has provided pattern-finding software to U.S. Immigration and Customs Enforcement since 2008, signed over $81 million in new contracts with ICE inside the first year of the second Trump administration. The total Palantir-ICE relationship is now north of $145 million. The flagship product, ImmigrationOS, is a platform that pulls data from passport records, Social Security files, IRS tax data, license-plate readers, and the Department of Health and Human Services to map neighborhoods, build dossiers, and rank deportation targets in something close to real time. A separate tool, ELITE, lets agents enter a zip code and receive a heat map of where suspected noncitizens are likely to be found.

The reporting on this is now well-documented. The Guardian has obtained internal ICE training materials showing how agents are taught to upload field data back into the Palantir system, which means every traffic stop becomes a permanent enrichment of the database, regardless of whether the stop resulted in an arrest. The contracts are sole-source, meaning no other vendor is allowed to bid. The justifications cite Palantir’s “deep institutional knowledge” of ICE operations, which is another way of saying that the agency now depends on a private company to run its core enforcement function.

A federal function has been outsourced so completely that the company providing it can describe itself as essential to national security and the agency has to agree, because the agency no longer has the in-house ability to do the work alone.

This is the part of the story the AI conversation almost never includes.

The data centers compute, and they also warehouse the digital substrate of a state that increasingly cannot tell which functions are public and which are private. The same servers that train a chatbot for a corporate customer are running pattern-matching for a federal agency that is being used to detain and remove people. The same energy your electric bill is helping pay for is keeping that pattern-matching online.

The grid is the policy.

When a government extends national security designation to a private industry, it inherits the obligation to keep that industry running.

A data center designated as a critical national asset cannot be allowed to go dark. That means the utility serving it cannot be allowed to fail. That means the public utility commissions in thirteen states served by PJM, the regional grid operator, are now operating inside a problem they did not create and cannot solve at their level.

The math is straightforward. The country needs an estimated forty-nine additional gigawatts of generation through 2028 to meet projected data center demand. That is roughly five percent of total U.S. generation. It is not being built fast enough.

To keep the lights on, coal plants from the 1970s are being held open past their planned retirement. New gas turbines are being ordered with multi-year wait times. Nuclear plants that were shuttered are being restarted, including the second reactor at Three Mile Island, the same site where the first reactor melted down in 1979. The Inflation Reduction Act, passed under a different administration, is now subsidizing the restart through federal tax credits for nuclear power. The same act is also subsidizing the renewable buildout that the current administration is trying to slow.

The result is that whatever you believe about climate, energy, or industrial policy, your electric bill is now an instrument of all three.

Coal stays online because AI needs it. Gas plants get built because AI needs them. Nuclear plants get restarted because AI needs them. Solar and wind projects get delayed because the current administration prefers it that way.

None of these decisions are made by you. They are made by public utility commissioners, FERC, the Department of Energy, regional transmission organizations, and the executive branch. They show up on your bill as a line item you cannot read.

The numbers say who pays.

Between 2020 and 2024, residential electricity rates rose twenty-five percent. Commercial rates rose three percent. Industrial rates fell two percent.

Households absorbed the increase. The data centers driving the increase received discounts.

Public utility commissions in Georgia just lost two incumbents in the November 2025 election over rate hikes, which is an early sign of what happens when the bill becomes visible enough to vote against.

Capacity is the new currency.

You can buy capital. You can borrow it. You can print it. You cannot conjure a transformer. You cannot wish a substation into existence. You cannot retrain an electrician overnight.

The physical bottleneck of grid buildout means that whoever gets the watts first wins, and the federal government has decided that the companies most worth winning are the ones designated as critical national assets.

Microsoft has pledged to cover its own electricity costs at its sites. Amazon, Google, Meta, and Anthropic have signed a non-binding ratepayer protection pledge. The pledge has no enforcement mechanism. The same companies are simultaneously asking utilities to deliver gigawatts on schedules the utilities cannot meet without raising rates.

Capacity is also a labor problem.

The AI Action Plan published by the White House in July 2025 includes a section on workforce development. It calls for federal apprenticeships in skilled trades, especially electrical work, HVAC, controls engineering. The reason is simple. The country does not have enough electricians to build out the grid the industry needs.

The same federal government that is incentivizing the rapid replacement of professional workers with AI is also acknowledging that it cannot build the infrastructure for AI without dramatically expanding a labor force whose work cannot be automated, at least not yet. The hands that wire the data center will be human. The work being done inside it, increasingly, will not.

Which means the country is sorting itself into two economies. One economy builds the boxes. The other economy is replaced by what is inside them.

The tax code rewards the second economy by making capital and equipment deductible at one hundred percent in the year of purchase. The same tax code does not reward hiring a human at any rate. Hiring a human creates payroll tax obligations, benefits costs, training time, legal exposure. Hiring a server creates a deduction.

This is not new. The asymmetry has existed for forty years. AI did not create it. AI is just the first replacement technology fast enough and cheap enough to make the asymmetry visible to people who used to think they were safe.

The database knows where you live.

A surveillance state used to require informants. It required a knock on a neighbor’s door, a notebook, a filing cabinet, a clerk to file the paper, a courier to deliver it, a supervisor to read it, a decision to act. The cost of finding a person was high. The cost of building a list was higher. The cost of acting on the list was higher still. That cost was the friction inside which civil liberties survived.

The cost is now near zero. ELITE renders a map in seconds. ImmigrationOS files the paperwork. Sole-source contracts mean the company providing the tool cannot be replaced and therefore cannot be held to its mistakes. The data flowing into the system is constantly enriched by other branches of government, including agencies that were never authorized to share their data this way. HHS data, originally collected to administer Medicaid, has reportedly been routed into ELITE. The IRS data that helped catch a tax fraud is now helping locate a deportation target. The Social Security record that tracks a retirement benefit is feeding a system designed to remove the person who earned it.

The constitutional theory underneath this is that the government is allowed to know what it already legally knows. The practical effect is that the government can now act on what it knows at machine speed, against millions of people simultaneously, using private infrastructure shielded from FOIA, governed by its own confidentiality clauses, and run by companies whose largest customer is the federal government.

There is no chain of accountability for this. There is a chain of contracts. The contracts are sole-source. The vendor is essential. The agency is dependent. The voter is paying for it through their utility bill. The fingerprints are everywhere and nowhere.

The state is the customer.

The largest single buyer of frontier AI services in the United States is now the United States government.

That includes the Defense Department, ICE, USCIS, the IRS modernization contracts, the Department of Health and Human Services, the Veterans Administration, and a long list of agencies whose procurement is happening under expedited authorities that bypass normal contracting timelines.

The companies selling those services are also the companies negotiating the rules under which they will be regulated. They are the same companies whose data centers are sitting on federal land. They are the same companies whose energy contracts are being underwritten by ratepayers in twenty-six states. They are the same companies whose CEOs sit on advisory councils to the same agencies they are selling to.

This is what people mean when they say “regulatory capture” without quite knowing why the phrase fits.

Capture used to describe a regulator becoming friendly to an industry. The new arrangement is bigger than that. The regulator, the supplier, the customer, the landlord, the financier, and the strategic partner are now overlapping institutions.

The boundaries that used to separate state capacity from private capacity have been redrawn in real time, mostly by executive order, mostly with no congressional debate.

Lockheed sells weapons. Palantir sells the operating system of enforcement. Microsoft sells the cloud the operating system runs on. NVIDIA sells the chips the cloud runs on. The Department of Energy supplies the power. The IRS supplies the data. The state supplies the legitimacy.

Each layer depends on the layer below it. None of them can be replaced without taking the whole stack offline.

The old military-industrial complex had two parties to it. This one has six, and one of them is you.

Innovation was the cover. Capacity was the project.

AI is the rationale for the largest peacetime expansion of state-funded private infrastructure in American history.

The technology revolution sits on top of it as the cover story. Roads were once built this way. Railways were built this way. The internet was built this way. The difference is that this time the public underwrote the buildout and the private companies kept everything.

There is no public option. There is no public utility model for compute. There is no national data center, no national AI lab, no FCC equivalent that licenses the spectrum and demands service in return.

The federal government built the highway and gave the keys to four companies and is now paying them to use it.

Everything else flows from that. The electric bill flows from it. The surveillance flows from it. The job changes flow from it. The political economy of small towns near data centers flows from it. The water table flows from it. The grid emergencies flow from it. The state laws being preempted flow from it. The deportation maps flow from it. The procurement contracts flow from it. The election of utility commissioners flows from it. The next round of layoffs at a Fortune 500 company flows from it.

This is the question of who owns the country now, dressed up as a technology story.

The answer is in the building you cannot see.

If a stranger lands tomorrow in your county and asks who governs here, the honest answer is that you do not entirely know anymore.

You know who was elected. You do not know who runs the procurement contract that determines whether your grid stays online. You know who sits on the school board. You do not know who decides whether your local utility raises rates to serve a data center you cannot enter. You know who your congressman is. You do not know which company holds the database your government uses to find you.

The work of the next decade, for anyone who wants to live in something recognizable as a democracy, is reconnecting those threads.

The bill is the database. The database is the contract. The contract is the executive order. The executive order is the campaign donation. The campaign donation is the lobbying budget. The lobbying budget is the share price. The share price is the data center. The data center is the bill.

The wheel was a sentence. The wing was a sentence. The bomb was a sentence. The data center is a sentence too, and we are still arguing about how to read it.

The light is on somewhere in your house. You forgot about it. Someone, somewhere, is counting on you to keep forgetting.