U.S. States and Cities Are Shaping the Future of AI Infrastructure

The U.S. AI data center boom is increasingly colliding with local opposition.

Summary

• What’s happening: As demand for data centers skyrockets, AI infrastructure decisions in the U.S. are increasingly being made by state and local governments, not just in Washington.

• What’s the problem? Key actors are moving in different directions. States are competing for investment, utility companies are warning of looming grid limits, and local governments are imposing restrictions amid community concerns.

• So what? Local pushback has already delayed or blocked $64 billion in U.S. data center projects. Without better alignment across levels of government, more major AI projects could face delay or cancellation.

• Better policy: Smarter alignment across all levels of government – through partnership agreements, grid standards, and guaranteed community benefits – could accelerate deployment while protecting communities from higher costs and strained electricity grids.

‍

Building AI’s backbone

The growth of AI is hitting America’s infrastructure limits. In Oklahoma, Google is spending $9 billion on new data center infrastructure; while across the U.S., OpenAI’s Stargate project is building more than 5 gigawatts of data center capacity.

What’s notable is that the decisive actors for these eye-watering investments aren’t CEOs in San Francisco – they’re local planners figuring out where in their jurisdictions are the power lines that can carry that load.

The American experience offers lessons for other federal systems. From Canada to India, countries are facing a tension between the AI ambitions of national governments and local control. Because these large AI projects demand fundamentally different infrastructure than traditional computing, policymakers will need to coordinate across jurisdictions in a way that hasn’t been necessary for other technology policies.

Unlike conventional data centers, AI facilities need far more power, high-performance GPU clusters, and advanced cooling systems. In 2018, a year before GPT-2 was released, overall power consumption of U.S. data centers was an estimated 76 TWh. By 2024, this number had more than doubled to 180 TWh.

Silicon Valley and Northern Virginia still lead in total installed AI data center power capacity, but Kearney’s 2025 AI Data Center Location Attractiveness Index shows momentum shifting to newer hubs like Austin, San Antonio, and Iowa. These regions boast abundant renewable power, fewer land constraints, and strong incentives. Oracle’s post‑deal site list reflects this shift: with new capacity planned in states such as Michigan, Wisconsin, Wyoming, New Mexico, Georgia, Ohio, and Pennsylvania, it’s clear the next wave of AI build‑out is moving beyond the traditional coastal corridors.

States themselves are fueling this shift through an incentives race. More than 40 U.S. states now offer tax breaks, fast-track permitting, or multi-decade exemptions to attract AI campuses – large sites that cluster multiple AI data centers with shared infrastructure.

As one example, in late 2024, Michigan extended its program to exempt large data center investments from sales and use taxes through 2050. Microsoft responded by quickly acquiring major parcels of land in the state, in turn prompting utilities to draft new pricing models. After all, a single hyperscale campus can consume as much electricity as an entire city.

‍Kansas, Kentucky, Arkansas, and New Mexico have followed with their own long-term packages, and Pennsylvania is preparing similar measures. The result is intensifying competition among states to capture investment, even as the concentration of very large loads is creating new pressures on local grids.

When AI infrastructure meets the ground

U.S. federal strategy is increasingly running into local resistance. In Virginia, the PW Digital Gateway – planned as the world’s largest data center corridor – suffered a major setback in August. A Circuit Court judge voided the rezoning approval after residents challenged the county’s public notice process, effectively blocking the project for now.

Across Northern Virginia, local governments are tightening their rules: enacting zoning restrictions in response to complaints about noise and proximity to homes; requiring public board review or Provisional Use Permits for new data centers; or mandating impact studies and special-use permits before construction.

Similar tensions are surfacing elsewhere. In Georgia, Atlanta’s city council has restricted new data center construction in several neighborhoods, while the South Fulton community has raised concerns about water use and rising electricity bills as dozens of projects advance.

In West Virginia, the dynamic played out differently. As residents in Tucker County mobilized against a proposed data center complex, state lawmakers passed the Power Generation and Consumption Act. The law stripped counties of zoning authority over data centers and microgrids, diverted most tax revenue to the state, and left communities with little say over projects in their backyard.

These episodes underscore the challenges in aligning federal, state, and local interests when it comes to AI infrastructure.

So what might better policy look like across levels of government?

Federal level: partnership, not preemption

The White House’s July 23 Executive Order already fast-tracks federal permits for data center projects and opens the door to using federal or brownfield land. But as recent cases show, local approval still hinges on concrete answers about power, water, noise, traffic, and tangible community gains. And the stakes are high: at least $64 billion in data center projects has been blocked or delayed in the past two years amid organized resistance across 24 states, including opposition from both Republican and Democrat district officials.

‍Recommendation: Federal policymakers should create voluntary partnership agreements for data center projects exceeding 100 MW that make use of the federal fast-track or federal/brownfield land.

Under this framework, the developer, local government, and relevant utility would jointly commit at the outset to a set of basic provisions, including:

• procedures to verify power capacity;

• a water-use plan;

• a defined package of local infrastructure improvements (such as roads, distribution upgrades, or sound mitigation);

• participation in local workforce pipelines.

To ensure accountability, the agreement could require developers to provide financial guarantees, with the terms publicly posted on the federal Permitting Dashboard.

With this approach, the federal government could help keep zoning and land-use decisions local while making the federal fast-track more workable for communities. Developers would benefit from greater predictability, utilities would gain a clearer sense of the expected load, and residents would be more likely to see tangible benefits rather than vague promises.

State level: managing large loads before they stall

U.S. states generally want the high-wage jobs and tax base that AI data centers bring, but rapid growth is colliding with grid limits. Utility company Dominion Energy has already said it can no longer guarantee service dates for new Virginia data centers without multi-year upgrades. Meanwhile, Texas grid operator ERCOT has received requests equal to 572 GW of new large-load connections, far more than the grid can deliver.

These strains show that permits are not enough; without a clear process for managing very large loads, states risk approving projects that cannot be energized on time.

‍Recommendation: States should require data centers over a certain threshold (e.g., 75-100 MW) to demonstrate grid readiness and operational flexibility before approval.

Texas offers a useful precedent here. Its SB-6 legislation requires large-load customers to pay for grid studies upfront, agree to reduce power during grid stress, and disclose if they're pursuing multiple grid connections.

Other states could follow suit by requiring developers to coordinate with utilities early, verify site control, and submit a flexibility plan before incentives or permits are granted. States might also offer priority incentives for projects paired with firm or on-site power.

Ultimately, states – not the federal government – get to decide who plugs into their power grids. Clear state rules would give utilities and regulators a consistent playbook, reduce speculative requests, and reassure communities that new projects will be managed responsibly.

Local level: balancing development and community concerns

Local governments sit at the front line of AI infrastructure growth. Even when federal and state approvals are in place, projects often hinge on whether towns and counties feel their concerns about power, water, noise, and community benefits are addressed. In many relevant regions there is local push back on data center growth, and without credible ways to reconcile those concerns, projects risk delay, litigation, or reversal.

‍Recommendation: Local governments could require formal community benefit and impact agreements for large data centers.

Such agreements could set out:

• transparent studies on electricity, water, and land use;

• commitments from developers to fund necessary grid upgrades so ratepayers are not left with higher bills;

• community benefits, such as workforce training, energy-efficiency programs for households, or road improvements.

Formalizing these agreements would not end opposition, but it could reduce litigation and give developers greater certainty before committing large investments. Residents would see their concerns about power, water, and local infrastructure addressed up front. Utilities would also gain a clearer path to recover the costs of serving very large loads.

Together, these steps could help local governments capture investment while limiting the risks that have fueled community resistance elsewhere.

Local choices, global lessons

The U.S. experience shows that AI infrastructure isn’t just about federal policy – state and local decisions determine whether and when projects actually get built. Other democracies building their own AI capacity can learn from America’s mix of rapid investment and local push back. Getting alignment across federal, state, and local levels, even incrementally, will help determine whether the U.S. can build its computing base fast enough to stay competitive.

‍This article reflects the authors’ perspectives and does not necessarily represent the views of any institution with which they are affiliated.