AI Data Centers: Community Impact & The National Debate

The training infrastructure gap. Large language models like GPT-4, Gemini, and LLaMA require clusters of tens of thousands of specialized chips running for months to complete a single training run. That computation has to live somewhere. The buildout that started in 2022–2023 to support training is now being followed by an even larger buildout for inference — the compute required to run AI models at scale for users. Inference demand grows with every user added.

Why the Midwest specifically. The coastal data center markets (Northern Virginia, Silicon Valley, Phoenix) are hitting power grid constraints. Utilities in those areas have waitlists measured in years for new large-load interconnections. Midwest markets — Michigan, Ohio, Indiana, Wisconsin — have available grid capacity, lower land costs, cooler ambient temperatures (reducing cooling costs), and proximity to fiber corridors. They are the path of least resistance for the next wave.

Why communities are unprepared. Zoning codes are written for the land uses communities have experienced. A township that has permitted industrial uses for decades has rules for manufacturing plants, warehouses, and distribution centers. It almost certainly has no rules for a 100 MW electrical load, no noise standards for 24/7 industrial cooling equipment, and no water use disclosure requirements. The permit application arrives before the rules exist.

Sources: Goldman Sachs 2024 AI infrastructure report; U.S. DOE data center energy estimates; Dominion Energy rate filings; Bridge Michigan (April 2026).

When a data center permit application arrives at a township board, the community is effectively making several decisions simultaneously:

The fundamental question isn’t whether data centers are good or bad. It’s whether communities are writing the rules before or after the permit is issued. The evidence from Virginia, Arizona, and Texas is clear: communities that wrote rules first got better outcomes than communities that improvised in real time.

Sources: Virginia Code §15.2-2295; Texas Tax Code §151.359; Mesa, AZ water reporting ordinance; Loudoun County VA planning records.

Three states have the largest data center concentrations in the US: Virginia, Texas, and Arizona. Their experiences form the baseline for what communities in newer markets like Michigan can anticipate.

Virginia: Loudoun County “Data Center Alley”

Roughly a third of all global internet traffic routes through Northern Virginia. Virginia eventually codified statewide data center development standards (Virginia Code §15.2-2295), but only after communities had already permitted facilities without uniform rules. Dominion Energy requested rate hike from all customers partly attributed to data center load growth. The Virginia General Assembly has introduced legislation to slow new approvals as grid constraints become acute.

Lesson: Even early movers who wrote rules eventually outran their frameworks. Statewide coordination came too late for communities that made early decisions without standards.

Arizona: Phoenix Metro Water Crisis

Arizona’s data center growth collided with one of the most acute water crises in US history. Communities that approved large facilities without binding water use caps found that actual consumption far exceeded developer projections. Mesa eventually required annual water disclosure from large facilities. Arizona’s Department of Environmental Quality issued guidance, but communities that had already permitted facilities had no enforcement mechanism.

Lesson: Water consumption projections submitted by developers tend to reflect optimal conditions. Real-world use in hot climates is significantly higher. Binding limits must be in the permit, not in promises.

Texas: Tax Abatement Lessons

Texas’s permissive development environment means townships have less zoning authority than in states like Michigan. But Texas Tax Code §151.359 created generous sales tax exemptions for data centers, and county-level property tax abatements further reduced actual revenue to host communities. Projected economic benefits routinely overstated actual community benefit.

Lesson: Economic impact studies submitted by developers use optimistic assumptions. Property tax revenue projections must account for the likelihood of abatement applications. Communities should require independent review of economic projections.

Sources: Virginia Code §15.2-2295; Texas Tax Code §151.359; Mesa, AZ water use reporting ordinance; Goldman Sachs 2024 data center infrastructure report.

When a community is deciding whether and how to permit a hyperscale data center, these are the questions that matter — and what the evidence from existing markets shows about each.

Jobs — construction vs. operations. A hyperscale campus generates hundreds of construction jobs for 18–36 months. Permanent operational jobs typically number 20–50 per facility. Data centers are capital-intensive, not labor-intensive. Economic benefit projections that emphasize employment are often misleading about the long-term jobs picture. The honest number to ask for: permanent FTE staff employed after construction completion.

Property tax revenue. This is the most legitimate economic benefit. A $500M–$2B campus generates significant assessed value — and unlike manufacturing, it doesn’t depreciate quickly. The caveats: (1) developers routinely apply for Industrial Facilities Tax Exemptions (IFT abatements) that reduce actual revenue. Cascade Township’s February 2026 decision to tighten IFT policy was directly motivated by this risk. (2) Tax revenue projections should be calculated on the abated value, not the assessed value.

Power — ratepayer impact. A 100–300 MW load requires new substation and transmission infrastructure. In markets like Michigan, the utility (Consumers Energy or DTE) builds that infrastructure and recovers the cost through rate proceedings that affect all customers in the service territory. In Virginia, data center load growth was cited as a factor in Dominion Energy’s residential rate increase. Michigan communities should ask Consumers Energy directly: who pays for grid upgrades, and when is that cost recovered?

Water consumption. Evaporative cooling towers consume water proportional to ambient temperature and server load. In Michigan’s climate, consumption is lower than in Arizona or Texas — but it is not zero. A 100 MW facility may consume 500,000–2,000,000 gallons per day depending on design. Michigan’s Great Lakes Compact (MCL §324.32701) triggers review at certain withdrawal thresholds, but Compact protections are aggregate and state-level. They do not replace local permit conditions. Communities that want binding water limits need to write them into the zoning ordinance before the permit is issued.

Noise. Data centers run 24 hours a day, 7 days a week. Cooling towers, backup generators, and transformer equipment produce continuous noise. Industrial zoning typically allows higher noise levels than residential zoning, but the question is whether the community’s industrial noise standards were written anticipating this type of continuous, overnight noise source. Communities near residential areas should require third-party noise impact assessments and binding nighttime noise limits as permit conditions.

Visual impact and setbacks. A hyperscale campus involves large low-rise buildings, cooling towers, generator pads, and transformer yards. Without setback and screening requirements in the zoning code, the facility can be built at the property line facing a residential street. This is not hypothetical — it has happened in communities that approved facilities before writing visual standards.

Traffic. Unlike manufacturing, a data center generates almost no daily traffic once operational. Construction traffic is significant but time-limited. Long-term neighborhood traffic impact is minimal — this is one area where data centers compare favorably to other large industrial uses.

Sources: Lawrence Berkeley National Laboratory data center energy report (2024); Michigan DEQ water withdrawal data; Dominion Energy SCC rate case filings; Cascade Township Board packet (Feb 25, 2026).

The federal government has designated AI infrastructure development as a strategic and national security priority. That priority is creating pressure on state and local governments to approve data center projects quickly — while communities lack the tools to evaluate them effectively.

What the executive branch is doing:

• EO 14110-A (January 2026): Directed DOE and DOD to identify and fast-track permitting for AI infrastructure. Directed FERC to prioritize interconnection for AI data center projects. Does not override local zoning authority on private land.
• DOE data center efficiency program: The Department of Energy operates a voluntary data center efficiency program with energy use and water use benchmarks. Participation is voluntary — most facilities do not publish their actual consumption data publicly.
• FERC interconnection queue reforms: FERC has proposed reforms to the grid interconnection queue process that would affect how quickly new large loads (including data centers) can connect to the transmission grid. These reforms are proceeding through notice-and-comment rulemaking.

What Congress is considering:

• Data Center Transparency Act (HR 4471): Would require data centers above 100 MW to file annual public disclosures of total energy consumption, water withdrawal, water recycled, and backup generator fuel use with DOE. No state currently mandates this disclosure. Bipartisan co-sponsors from Virginia, Arizona, and Texas — states that have already seen grid and water impacts. In the House Energy & Commerce Committee.
• Grid Modernization Act (HR 5213): Would provide $28 billion in DOE grants for transmission and distribution grid upgrades, including provisions for federal co-funding on large load interconnections. If passed, could reduce the amount data center grid costs are recovered from residential ratepayers through utility rate cases.
• AI Infrastructure Investment Act: Would create a new category of federal loan guarantees for AI data center campuses that meet minimum energy and water efficiency standards. Intended to direct investment toward more efficient designs by making financing cheaper for qualifying facilities.

What the federal government is not doing:

• No federal minimum standards for noise, water withdrawal, or setbacks at data centers — these remain state and local authority.
• No mandatory federal disclosure of data center energy or water consumption exists (transparency legislation is pending but not passed).
• No federal preemption of local zoning authority for data centers — townships and municipalities retain control over permits on private land.
• No federal reimbursement to communities for grid infrastructure upgrade costs that are socialized through utility rate cases.

Sources: Executive Order 14110-A (January 2026); FERC Docket RM21-17 (interconnection reform); HR 4471 (Data Center Transparency Act); HR 5213 (Grid Modernization Act); DOE Better Plants data center program.

These questions are derived from the experiences of communities in Virginia, Arizona, Texas, and Michigan that have already been through this process. They are not arguments for or against any specific project — they are due diligence that every community is entitled to expect answers to before voting.

About the facility itself:

• Energy: What is the total connected electrical load in megawatts at full buildout? What is the planned source of that power? Has the serving utility confirmed available capacity without new infrastructure? If new infrastructure is required, who pays?
• Water: What is the projected maximum daily water withdrawal in gallons at full buildout and at typical operation? What cooling technology is planned? What percentage of water is recycled? Have all required water withdrawal permits been applied for under state law?
• Noise: What is the projected fenceline noise level at full operation, in decibels, measured at the nearest residential property line? During the day? During nighttime hours? What backup generator testing schedule is planned, and what noise does testing produce?
• Emissions: How many backup diesel generators will the facility operate? What is the total diesel storage capacity? What are the projected air emission impacts from generator testing and emergency operation?

About the economic claims:

• Jobs: How many permanent full-time employees will the facility employ at full operation (not including construction workers)? What is the projected wage range?
• Tax revenue: What is the projected assessed value of the facility? Does the developer intend to apply for any property tax abatements, exemptions, or PILOTs? What is the projected annual property tax revenue net of any abatements?
• Economic impact study: Was the economic impact study submitted by the developer or funded by an independent party? What methodology was used? How do the projections compare to outcomes at comparable facilities in other markets?

About enforcement:

• Standards: Are the commitments the developer is making written into the permit as conditions, into a zoning ordinance as binding standards, or into a voluntary Community Benefit Agreement? What is the enforcement mechanism for each?
• Monitoring: Who monitors compliance with noise, water, and emissions commitments? The township? The developer? An independent third party? At what frequency?
• Remediation: If the facility violates its conditions, what remedies does the community have? What are the practical legal costs of enforcement against a multi-billion-dollar company?

Sources synthesized from: Virginia SCC proceedings; Loudoun County VA planning documents; Mesa, AZ water reporting ordinance; Michigan Zoning Enabling Act PA 110; EGLE Water Withdrawal Assessment Program; Goldman Sachs 2024 data center infrastructure report.

You don’t need to be a zoning expert or an engineer to meaningfully participate in your community’s data center debate. The most effective thing any resident can do is ask specific, factual questions and show up.

Find your community’s specific debate: Most data center decisions are made at township or municipal board meetings. Your community’s website will list meeting dates and agendas. Agendas are typically published 24–72 hours before the meeting.

Get the primary documents before the meeting: The permit application, staff reports, and any supporting studies should be publicly available. Request them from your township clerk if they’re not on the website. In Michigan, public records are available under the Freedom of Information Act (MCL §15.231).

Public comment is your window: Most zoning hearings have a public comment period. You don’t need to be opposed to or in favor of the project to speak. Asking factual questions — the ones in the section above — is a legitimate and valuable form of participation. Boards are more responsive to specific questions than to general positions.

Follow the process after the initial vote: If your community passes a moratorium, the Planning Commission process is where the actual standards are written — and those hearings are also public. If your community approves a permit, the conditions of that permit are public record and worth reading.

Share what you find: Debatable aggregates primary source documents and both-sides analysis. If you’re tracking a data center debate in your community and have access to documents or insights that aren’t in the knowledgebase, reach out. The goal is to build a resource that helps communities make better-informed decisions — not to tell them what to decide.