AI Boom Should Prompt States to Rein in Data Center Tax Losses

Data center tax breaks are exploding in cost nationwide. If states insist on subsidizing these data centers, which power artificial intelligence models, they should tie every dollar to how responsibly these facilities use power. That means capping the duration and size of exemptions, requiring environmental concessions, and prioritizing placement of centers that align with grid and climate interests.

In Texas alone, projected losses from data center tax exemptions ballooned from $157 million to over $1 billion in less than two years. Virginia now devotes nearly half of its economic development incentive spending to data centers. In state after state, the advent of AI has turned what began as modest carveouts for tech infrastructure into open-ended giveaways.

The rush has seemingly been treated as a contest of which jurisdiction can give away the most money with the least oversight. But to keep investment of public funds into data centers sustainable, the question should be broader than how many total data centers a state can attract.

These subsidies are so potentially problematic because of how little transparency is available regarding how they’re being handed out, not just their cost. In many states, data center tax breaks are virtually automatic once a facility qualifies. And with that qualification comes exemptions from state sales tax, and frequently from local taxes as well—municipalities get their right to tax from the states in which they sit.

The results are self-perpetuating, opaque credit programs that cost more each year—with no check mechanism to break the cycle. Because most states don’t even require basic disclosure of how much tax revenue is being left on the table, the public doesn’t always have a full view of the cost even at the highest level of abstraction.

The policy theory behind these exemptions is that they reward companies for investing in high-tech, job-creating infrastructure. In practice, they often reward scale and repetition. Data centers need constant server upgrades and replacements, so exemptions apply repeatedly, and with more total savings for larger facilities. This transforms a policy carrot into a bottomless tax credit buffet for big tech.

For what they lack in job creation, data centers more than make up for in electricity and water demand. AI data centers particularly require far more power than traditional ones, thanks to the use of energy-intensive graphical processing units that require more aggressive cooling systems.

Perhaps more importantly, they don’t just use more energy overall—they often do so at peak demand hours. Other energy-intensive applications, such as charging electric vehicles, can be time-shifted to overnight off-peak hours. But generating a picture of a monkey in sunglasses to attach to a work email can’t wait until 3 a.m.

If data centers are inevitable, fail to deliver jobs, and undercut grid resilience and state climate commitments, what can they do to make them worth their share of lost tax revenue?

Photographer: Nathan Howard/Bloomberg via Getty Images

When data centers are being indiscriminately promoted across the country, many will necessarily rely on electricity generated by fossil fuels in states where clean power isn’t abundant and accessible. This creates a policy push-pull, where state legislatures pass climate goals and net-zero targets while shoveling cash into fossil-powered server farms.

This raises a relevant policy question: If data centers are inevitable, fail to deliver jobs, and undercut grid resilience and state climate commitments, what can they do to make them worth their share of lost tax revenue?

Policymakers could require that any facility receiving a tax break be sited near stranded renewable energy—that is, wind or solar power that otherwise goes uncollected because of limits of the local transmission system rather than a lack of turbines or solar panels.

Building data centers near sites that often have excess energy makes those centers about as low impact as possible, at least in terms of the grid. In some cases, these centers could act like virtual batteries for renewable power, reducing load when demand spikes and ramping it up when the grid has capacity to spare.

States can shape how the industry develops. Tax credits tied to demand flexibility, the use of dry cooling systems, or the capacity for energy storage could reduce stress on water systems and grids.

Demand flexibility would require data centers to scale their use back based on grid conditions—reining in workloads when demand spikes and making use of off-peak hours or renewable overproduction for high-energy tasks such as training.

Dry cooling systems, meanwhile, make use of air or minimal water to cool servers, reducing stress on local aquifers. Similarly, requiring on-site energy storage through battery banks allow centers to store solar or wind power generated during low-demand hours for use when grid systems are stressed.

These requirements for credit eligibility would help align the interests of electricity ratepayers and taxpayers, while still encouraging investment in technology infrastructure.

The AI boom clearly isn’t slowing down any time soon, and the infrastructure it runs on must live somewhere relatively close to the users it serves. That doesn’t mean states must continue to hand over taxpayer funds to big tech. With data center tax breaks spiraling out of control this early into AI development, policymakers must start asking what more data center owners can be doing to support public interests and earn public funds.

Absent intervention, we aren’t building an AI-powered future—we’re running yesterday’s policy inefficiencies on faster and hotter servers. States must begin prioritizing the energy cost, location, and environmental footprint of centers they incentivize and demand more than just fiber hookups in return.

Andrew Leahey is a tax and technology attorney, principal at Hunter Creek Consulting, and practice professor at Drexel Kline School of Law. Follow him on Mastodon at @andrew@esq.social

Read More Technically Speaking