Research report — August 23, 2026

Executive summary

COMMENTS FROM STREET TALK NETWORK ABOUT THIS POST-

1. Do your own research on the information provided and then come up with your own conclusions.

2. This article contains many names and other information that can be researched.

3. Trust your gut feeling on information provided as this post may contain slanted information.

Data centers are the physical factories of the digital economy. They contain servers, networking equipment, storage systems, cooling plants, backup generators, batteries, transformers, and security systems. Traditional facilities serve websites, streaming, banking, government, and cloud computing; newer AI centers add extremely power-dense clusters of graphics processing units (GPUs) and specialized accelerators.

The expansion is producing four connected controversies:

1. Electricity: U.S. data centers consumed about 176 terawatt-hours (TWh) in 2023—4.4% of national electricity. Lawrence Berkeley National Laboratory (LBNL) projected 325–580 TWh, or 6.7%–12%, by 2028. Its 2025 update placed the central 2030 estimate at 11.8%, with a 9.5%–15.3% range. These are scenarios, not certainties, but they show the scale of the possible change.

2. Water: LBNL estimated 66 billion liters (17.4 billion gallons) of direct on-site water consumption in 2023, plus nearly 800 billion liters (211 billion gallons) indirectly consumed in producing the electricity used by the facilities. Hyperscale centers alone were projected to consume 60–124 billion liters directly in 2028. Local effects vary enormously by cooling method, climate, watershed, power source, and whether reclaimed water is used.

3. Public costs and benefits: Data centers bring enormous capital investment and construction activity, but relatively few permanent jobs per dollar invested. At least 38 states offer incentives. Virginia’s data-center sales-and-use-tax exemption was worth an estimated $928 million in fiscal 2023. The disputed question is whether new tax revenue, jobs, and infrastructure benefits exceed foregone revenue and grid, water, road, and environmental costs.

4. Special treatment: There is no credible evidence that ordinary data-center companies have acquired, or can acquire, sovereign-state status. They remain subject to federal and state law. But governments have created special regimes that may resemble fragments of governmental privilege: sales and property-tax exemptions; fast-track and consolidated permitting; access to federal land; special utility tariffs; private or “behind-the-meter” microgrids; confidentiality for security or proprietary information; and, in West Virginia, state preemption of much local zoning, permitting, inspection, noise, and lighting authority for certified projects.

The accurate conclusion is therefore: not sovereignty, but in certain jurisdictions a form of legally privileged industrial enclave. The distinction matters. A legislature can amend or repeal these privileges; courts can review them; environmental and civil law still applies; employees and residents remain citizens of the surrounding jurisdiction; and the company has no independent territory, criminal law, citizenship, diplomacy, or military authority.

1. What a data center is

Data centers range from small enterprise server rooms to colocation buildings serving many customers and hyperscale campuses operated by companies such as Amazon, Google, Meta, Microsoft, and Oracle. AI training and inference require dense clusters of accelerators. Nearly all electricity entering the computers becomes heat, which must be removed continuously.

The two standard efficiency measures are:

• Power
  Usage Effectiveness (PUE): total facility electricity divided by
  IT-equipment electricity. A PUE of 1.2 means that every 1 kWh used by
  computing requires another 0.2 kWh for cooling, power conversion,
  lighting, and support.
• Water
  Usage Effectiveness (WUE): on-site water consumed per unit of IT
  electricity, usually liters per kWh. This should not be confused with
  water withdrawal, much of which may be returned. “Consumption”
  means water not immediately returned to the source, commonly because it
  evaporates.

An efficient PUE does not necessarily mean a low water footprint. Evaporative cooling can save electricity while consuming more water; dry cooling can save water while using more power, particularly in hot weather. The responsible comparison is therefore energy, direct water, indirect power-sector water, carbon, local scarcity, and reliability together, not one number.

2. Electrical demand

National scale

The 2024 LBNL report found that U.S. data-center electricity use rose from roughly 58 TWh in 2014 to 176 TWh in 2023. The latter equaled 4.4% of U.S. consumption. It projected 325–580 TWh in 2028, equivalent to 6.7%–12% of national use. The wide band reflects uncertain GPU shipments, utilization, operating hours, cooling systems, and efficiency. The 2025 update estimated a central 11.8% share in 2030, ranging from 9.5% to 15.3%.

The rapid change reverses the efficiency-dominated trend of the 2010s. Virtualization, better servers, and migration to efficient hyperscale facilities once allowed computing output to grow much faster than electricity use. AI accelerators and the scale of new campuses are now overwhelming part of those savings.

Regional concentration and reliability

National percentages conceal local stress. Data centers cluster where fiber routes, land, tax incentives, skilled contractors, and grid connections are favorable. Northern Virginia is the leading U.S. cluster. PJM, the regional grid operator serving parts of 13 states and Washington, D.C., said its 2025 forecast attributed 30 GW of 32 GW of load growth from 2024 to 2030 to data centers. For perspective, 30 GW is the output of roughly thirty large power reactors when operating at full rating, although annual energy and peak capacity are not the same measurement. PJM also projected upward price pressure and resource-adequacy risk.

Large campuses create several grid problems:

• Connection
  queues:
  a proposed 500-MW or 1-GW campus may require new substations, transmission
  lines, transformers, and generation.
• Forecasting
  risk:
  developers may submit overlapping or speculative requests, causing
  utilities to plan for loads that never arrive.
• Stranded
  assets:
  if a utility builds infrastructure and the customer delays, leaves, or
  uses less power, ordinary customers could be left paying for it.
• Capacity
  prices:
  sharply rising demand can increase the price paid to ensure generation is
  available during peaks.
• Retirement
  conflicts:
  demand may grow faster than replacement generation after coal, gas, or
  nuclear retirements.
• Reliability: a huge,
  inflexible load can worsen a shortage, while a flexible facility capable
  of curtailing demand can help the grid.

In June 2026, FERC ordered all six federally regulated regional grid operators to justify or reform their large-load connection rules. FERC framed the effort around faster connections and ratepayer protection. Important safeguards include minimum bills, long contract terms, upfront deposits, exit fees, construction contributions, proof of project readiness, and curtailment obligations.

Who should pay?

The core principle is cost causation: customers that cause new costs should bear them. A special rate is not automatically a subsidy. A properly designed large-load tariff can protect households by charging the data center for dedicated facilities and the risks of unused capacity. Conversely, discounted electricity, socialized transmission upgrades, weak exit charges, or unrealistic forecasts can shift cost to residents and small businesses.

Public utility commissions should publish, with genuine security redactions only, the forecast load, required upgrades, allocation method, minimum bill, term, collateral, curtailment rights, and effect on residential bills.

3. Water consumption

Direct and indirect water

LBNL’s national model estimated:

These figures are from the LBNL national study. They are modeled national totals, not a rule for every building. One facility can use very little potable water with dry or closed-loop systems; another can consume millions of gallons per day during hot periods. “Millions of gallons per day” claims must always specify maximum versus average demand, potable versus reclaimed water, withdrawal versus consumption, and campus buildout stage.

Why location matters

A gallon consumed in a water-rich basin during a wet season is not equivalent to a gallon consumed from a stressed aquifer during drought. Relevant questions include:

• Is
  the source a river, reservoir, groundwater aquifer, potable municipal
  supply, wastewater, or reclaimed industrial water?
• Does
  the utility possess enough treatment and pipe capacity at summer peak?
• Are
  household growth, agriculture, ecosystems, and drought reserves included
  in the plan?
• Will
  warmer discharge affect streams?
• Does
  the cooling process concentrate salts or treatment chemicals in blowdown
  water?
• Are
  actual monthly use and WUE publicly reported?

Virginia’s nonpartisan JLARC review concluded that statewide use was then sustainable but growing, while warning that some localities and groundwater-management areas have less water and that oversight of how available water is allocated among competing users was limited.

Better approaches

Mitigation includes climate-appropriate dry cooling; closed-loop liquid cooling; higher server inlet temperatures; air- and water-side economizers; direct-to-chip cooling; workload shifting to cooler hours or less-stressed regions; reclaimed wastewater; metering; leak detection; and drought curtailment plans. The EPA’s Quincy, Washington case study describes a city–Microsoft reuse utility that treats and recirculates cooling water, reducing reliance on potable groundwater.

“Water positive” corporate pledges require careful reading. A restoration project elsewhere cannot necessarily repair depletion of the specific aquifer or municipal system serving a campus. The best accounting is watershed-specific, time-matched, independently verified, and based first on avoiding consumption.

4. Air, carbon, land, noise, and equipment

Data centers can have a large carbon footprint when supplied by fossil-heavy grids. Renewable power-purchase agreements can add clean generation, but an annual contract does not prove that the center runs on carbon-free power every hour or that local grid constraints disappear.

Diesel backup generators emit nitrogen oxides and particulate matter. They normally run for tests and outages, not continuously. Virginia JLARC found that generators represented less than 4% of regional nitrogen-oxide emissions and 0.1% or less of regional carbon-monoxide and particulate emissions in Northern Virginia, while also recommending attention to local effects. Proposals to operate generators during grid shortages or to build dedicated gas plants change that calculation and require separate air-quality analysis. EPA maintains Clean Air Act resources for data centers.

Other effects include large land footprints, transmission corridors, loss of farmland or habitat, construction traffic, visual impacts, continuous fan and chiller noise, low-frequency hum, light pollution, and eventual electronic waste. JLARC found that about one-third of Virginia centers were near residential areas and called data-center industrial use largely incompatible with nearby homes without adequate siting and mitigation.

5. Economic benefits, jobs, and tax incentives

The genuine benefits can be substantial: billions of dollars in construction; demand for electricians, pipefitters, engineers, security, maintenance, and suppliers; property and utility revenue where exemptions do not remove it; reuse of industrial sites; and infrastructure essential to modern services and national competitiveness.

The weakness is that a completed hyperscale facility is highly automated. Permanent employment is usually modest relative to capital invested, and construction jobs are temporary. A sound incentive evaluation must report jobs created, median wage, local hiring, total exemption, tax revenue received, public infrastructure spending, electricity and water costs, and the counterfactual—whether the center would have located there without the subsidy.

The National Conference of State Legislatures reported that at least 38 states offer data-center incentives. Common forms are:

• exemption
  of servers, cooling equipment, and construction materials from sales/use
  tax;
• property-tax
  abatements or special valuation;
• reduced
  or exempt taxes on electricity;
• tax-increment
  financing and publicly financed infrastructure;
• workforce
  grants;
• discounted
  land or expedited permits.

Virginia illustrates both sides. Its exemption helped create the country’s largest cluster and a strong local tax base, but the estimated fiscal-2023 state tax relief was $928 million, and about 90% of the industry used it. Incentives should have expiration dates, audit rights, public cost-per-job calculations, clawbacks for missed investment or employment promises, and resource-performance conditions.

6. “Special legal treatment” and the sovereignty question

What sovereignty would mean

A sovereign state normally controls territory, makes general law, exercises police and judicial authority, controls entry or citizenship, conducts foreign relations, and is not subordinate to another domestic government. A U.S. data center possesses none of those attributes. Private ownership, fencing, guards, a microgrid, tax relief, or federal “critical infrastructure” designation does not create sovereignty. Nor does a special economic zone cease to be part of the United States.

What companies and allied officials actually seek

Documented requests and enacted measures fall into narrower categories:

1. Tax preference: exemptions and special property valuation.

2. Speed: one-stop, fast-track, or consolidated permitting and shorter environmental review.

3. Land access: federal or state land leases and redevelopment sites.

4. Energy autonomy: co-located generation, microgrids, nuclear or gas plants, and special interconnection rules.

5. Local preemption: state limits on municipal zoning, building, noise, or permitting authority.

6. Confidentiality: protection of cybersecurity, customer, design, or negotiated economic-development information from public disclosure.

7. Dedicated utility treatment: tailored tariffs, priority studies, interruptible service, or negotiated infrastructure.

These can cumulatively make a campus feel enclave-like, but every power comes from ordinary public law or contract.

The strongest state example: West Virginia

West Virginia’s 2025 Power Generation and Consumption Act created certified microgrid districts and special treatment for high-impact data centers. The legislation included certification, special property valuation and tax distribution, and rules for utility service. Reporting by the Associated Press documented the most controversial feature: the state removed substantial county and municipal authority over land use, permitting, noise, and lighting for qualifying developments, even as residents near a proposed Tucker County gas-powered project sought information and local control.

This is properly described as state preemption and a special development regime, not corporate sovereignty. State environmental permits and federal law remain; the legislature created the regime and can change it. Nevertheless, it is a serious democratic issue because the people bearing local effects may lose their most accessible decision-makers and because confidentiality can prevent informed participation.

Federal fast-tracking and federal land

The July 2025 executive order on accelerating federal permitting of data-center infrastructure directed agencies to facilitate qualifying projects through federal land availability and faster environmental and permitting processes. DOE subsequently selected Idaho National Laboratory, Oak Ridge Reservation, Paducah, and Savannah River Site for private AI data-center and energy-infrastructure solicitations.

This is preferential federal industrial policy. It does not transfer sovereignty or exempt a private operator from all law. The important oversight questions concern lease terms, environmental review, water and power responsibility, nuclear or fossil generation, cybersecurity, public return, cleanup liability, and whether critical-defense designation restricts transparency.

Secrecy versus legitimate security

Data centers have real security needs: detailed floor plans, network architecture, customer identities, vulnerabilities, and some utility information can create physical or cyber risks. But blanket nondisclosure agreements, secret water commitments, undisclosed tax concessions, and closed zoning negotiations prevent the public from assessing public costs. A defensible rule protects narrowly specified technical details while publishing ownership, location, acreage, forecast and actual power and water, emissions permits, incentive value, permanent jobs, public infrastructure obligations, and emergency plans.

7. Arguments on both sides

Industry and government proponents argue that data centers are indispensable infrastructure; AI leadership is a national-security and economic priority; campuses bring investment and tax base; efficient hyperscale centers can replace less efficient server rooms; private generation can reduce grid burden; and slow, inconsistent permitting risks losing projects to other states or countries.

Residents, consumer advocates, and environmental critics argue that the permanent job count is too small for the subsidies; household ratepayers may finance grid expansion; potable water and land are being committed without transparent accounting; fossil plants and diesel fleets undermine climate and air goals; local democracy is weakened by preemption and NDAs; and companies can abandon or downsize plans after public infrastructure is built.

Both sides can be partly right. The decisive issue is the contract and regulatory design: who pays, who bears failure risk, what is disclosed, which limits apply during scarcity, and whether communities retain a meaningful voice.

8. A public-interest policy framework

Before approving a large facility, governments should require:

1. Resource disclosure: projected and actual monthly electricity, peak MW, direct water withdrawal and consumption, source, WUE, PUE, and fuel mix.

2. Independent cumulative review: include other proposed centers, housing, agriculture, drought, retiring generators, transmission constraints, and climate effects—not merely one project in isolation.

3. Ratepayer protection: dedicated-facility payment, minimum bills, collateral, long-term take-or-pay contracts, exit fees, and no recovery from ordinary customers for speculative capacity.

4. Water safeguards: reclaimed water where safe, drought stages, potable-water caps, aquifer monitoring, chemical and thermal-discharge limits, and watershed-specific replenishment.

5. Local land-use standards: setbacks, noise frequency limits, lighting, landscaping, height, generator hours, traffic, emergency response, and residential buffers.

6. Transparent incentives: public agreements, cost per permanent job, expiration, annual audits, wage and investment conditions, and clawbacks.

7. Clean and firm power: additional generation and storage matched as closely as possible by place and hour; honest accounting of backup and co-located fossil generation.

8. Flexible-load obligations: compensated curtailment during emergencies where technically feasible, without compromising essential services.

9. Decommissioning security: bonds for dismantling, contaminated equipment, batteries, fuel, and site restoration.

10. No blanket secrecy or immunity: narrow security exemptions; ordinary labor, environmental, civil-rights, building-safety, and judicial rules remain enforceable.

9. How to evaluate claims

When encountering a dramatic claim, ask:

• Does
  “use” mean withdrawal or consumption?
• Is
  the number average, peak-day, permitted maximum, or full future buildout?
• Does
  “power” mean MW capacity or MWh/TWh energy?
• Is
  an announced campus actually financed and connected?
• Is
  a tax exemption being confused with complete freedom from tax?
• Is
  local preemption being confused with exemption from all state and federal
  law?
• Does
  “self-powered” mean electrically islanded at all times, or merely
  co-located generation connected to the grid?
• Is
  a company pledge measured in the affected watershed and during the
  affected season?

Conclusion

Data centers are becoming one of the most important infrastructure and public-governance questions of the decade. Their national water footprint is material, their local footprint can be decisive, and their electricity growth could reshape generation and transmission planning. They can deliver real economic and technological benefits, but the benefits do not justify undisclosed subsidies or shifting costs and risks to residents.

The evidence does not support the proposition that data centers are literally gaining sovereign-state status. It does support a narrower and important warning: governments sometimes assemble tax, land, energy, secrecy, permitting, and local-preemption privileges that create quasi-enclave treatment. West Virginia’s removal of significant local control is the clearest documented U.S. example. Federal fast-tracking and federal land access add another layer, but neither creates an independent state.

The proper democratic response is not to treat every data center as inherently harmful or to accept every project as inevitable. It is to require measurable resource limits, full public accounting, fair utility rates, enforceable community protections, and a clear rule that private digital infrastructure remains subordinate to public law.

Principal sources

• Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report
• LBNL, United States Data Center Energy Usage Report: 2025 Update
• U.S. Department of Energy summary of the 2024 report
• Virginia JLARC, Data Centers in Virginia (2024)
• National Conference of State Legislatures, Data Center Incentives
• FERC, Large-Load Integration Action (June 2026)
• PJM, Large-Load and Data-Center Growth
• White House, Accelerating Federal Permitting of Data Center Infrastructure (2025)
• DOE, Federal Sites for AI Data Centers and Energy Infrastructure
• West Virginia Legislature, SB 818 text
• Associated Press, West Virginia Local-Control Investigation
• EPA, Quincy Water-Reuse Case Study