With the artificial intelligence (AI) boom escalating in the United States at a rapid pace, states have rushed to entice data centres with juicy tax breaks. Data centres not only provide cloud-based storage but also power the training of increasingly powerful AI models.
Over 30 states have carved out tax incentives for data centre companies, arguing that without them, the data centres wouldn’t come—and that their presence is essential toward growing property and income tax revenue and driving economic development.
However, a new study from nonprofit research group Good Jobs First, authored by Greg LeRoy and Kasia Tarczynska, found that data centre tax breaks have swelled to billions of dollars in lost revenue for states a year, and that those losses, for some states, actually outweigh the tax revenue that the data centres bring in.
“At least 10 states already lose more than $100 million per year in tax revenue to data centres, the cloud-computing warehouses that were proliferating before artificial intelligence greatly accelerated their growth. The industry’s high-velocity growth, combined with the virtually automatic structure of the state tax exemptions, is preventing states from making accurate cost projections. For example, in the space of just 23 months, Texas revised its FY 2025 cost projection from $130 million to $1 billion. Virginia, Texas, and Illinois have each recorded revenue-loss spikes of more than 1,000% in recent years,” the study stated, while adding, “The loss of state spending control is surely worse than we can yet document—of the 32 states with tax incentives to data centres, 12 fail to disclose even aggregate revenue losses.”
These 12 “Dark States” (as defined by LeRoy and Tarczynska), which failed to disclose even aggregate revenue losses, much less company-specific subsidies, as is common in economic development, include Indiana, North Carolina, and Utah, all of which have substantial and/or growing data centre investments.
Miscalculation on the horizon
LeRoy and Tarczynska stated that, as the end users of building materials, machinery, and equipment, data centre companies are mandated to pay sales and use taxes. States, however, are exempting those purchases, making these exemptions the costliest subsidies for data centres. Because server farms are extremely capital-intensive and require replacement of servers every two to five years when they wear out, these exemptions are lucrative for companies and costly for states and localities.
“Some states are apparently failing to comply with a Generally Accepted Accounting Practices standard that requires annual disclosure of revenues lost to tax abatement programmes such as data centre tax exemptions. We know of no other form of state spending that is so out of control. Therefore, we recommend that states cancel their data centre tax exemptions. Such subsidies are absolutely unnecessary for an extremely profitable industry dominated by some of the most valuable corporations on earth, such as Amazon, Microsoft, Apple, Meta (the owner of Facebook and Instagram), and Alphabet (owner of Google),” the duo noted.
The paper’s authors further contended that even if there are positives, the public doesn’t yet know the extent of their costs, which are often obscured by tax privacy laws, confidentiality agreements, or a sheer lack of research.
State subsidy programmes specifically created for the industry typically exempt projects from paying sales and/or use taxes on their largest start-up and maintenance expenses, including construction materials, servers and server racks, cabling for power, data transmission and monitoring, distributed data-storage systems, security, surveillance, firewall, encryption, and cyber-intrusion defence equipment and software, and many more crucial materials.
“These sales and use tax revenue losses are multi-layered. When a state certifies a data centre as eligible for these tax exemptions, that certification typically includes the local share of the tax as well as the state share. That is, states effectively pre-empt local sales tax authority on data centres,” the researchers noted.
The data centre changes associated with AI are raising complexity and costs, and therefore, lost tax revenues. For example, the more advanced and higher-priced graphics processing units (or GPUs) required for AI computing use several times more electricity than central processing units (CPUs). This means companies are not only spending more on equipment that generates more heat, but they also need more electricity to run and cool the equipment, and that electricity is typically exempt from utility taxes.
“This is an additional subsidy atop the power-rate discounts, often of undisclosed value, that data centre operators negotiate with utility companies. These electricity subsidies, as well as local subsidies such as property tax abatements and dedicated infrastructure, are beyond the scope of this study. Most states’ eligibility rules for the sales and use tax exemptions were written when most data centres were far smaller than today’s. So virtually every new data centre these days easily qualifies (usually based on what are today very small hiring and capital investment requirements). So, data centre tax exemptions are virtually automatic,” noted LeRoy and Tarczynska.
These exemptions are very long or permanent and uncapped. No state limits the amount of tax exemption any one facility or company can receive. Once approved, a data centre is exempt from paying taxes for decades, and in six states, indefinitely. And statewide, the exemptions are uncapped: there is no limit on how much revenue can be foregone each year.
In fact, a handful of states saw the emergence of debates and reform efforts due to the runaway revenue costs. In 2024, the Georgia legislature passed a bill to pause the state’s data centre incentive programme for review. However, Governor Brian Kemp vetoed this bill. In early 2025, Washington State Governor Bob Ferguson established a multi-department task force to study the impact of data centres on tax revenue. The task force will also examine how data centres affect the state’s carbon neutrality plan and the number of jobs created.
Virginia recently completed 2025 legislative session included debates on 30 bills intended to improve disclosure over water and energy use, create incentives for data centres to operate more efficiently, and protect ratepayers from subsidising data centres’ energy infrastructure. Almost all of the bills failed, with no substantial reform proposals being seen on the horizon.
Things are not right
Many data centre agreements are shrouded in secrecy, thereby forcing LeRoy and Tarczynska to conduct their research based on limited data. However, the result still looks damning. According to LeRoy, states share no data on tax expenditures for data centres at all.
By looking through state budgets and other fiscal reports, the duo found that at least 10 states are missing out on $100 million in yearly tax revenue. Much of this comes from subsidies for equipment like expensive servers, which need to be swapped out every few years. Many of these exemptions are uncapped in terms of their dollar amounts or time limit. This means that as data centres grow in size and power, so do the subsidies, forcing states to drastically recalculate their budget projections. In Texas, the cost projection for its data centre sales tax exemption programme increased from $157 million in 2023 to $1 billion this year.
Data centre proponents offer several rebuttals, like most states already offering tax exemptions for manufacturing equipment, and these new carveouts are simply bringing the data centre industry in line with those rules. They also argue that without subsidies, data centre companies would set up shop elsewhere.
In 2022, a tax incentive evaluation study from the Carl Vinson Institute of Government at the University of Georgia found that 90% of data centre activity in Georgia was due to the incentive, suggesting that without those tax breaks, all of that business would have vanished.
However, there is debate among economists around how much incentives actually cause activities. A 2018 paper by economist Timothy Bartik found that the percentage of firms making location decisions based on incentives was significantly lower than the Vinson number, ranging from 2% to 25%.
Even if we consider data centres central to the digital infrastructure powering the 21st century, they may not be robust job creation engines. Instead, in the words of Andrew Leahey, an attorney and tax professor, “they resemble traditional infrastructure projects like highways or bridges. The primary value lies in their utility rather than the jobs they create long term.”
“While there may be compelling reasons to support data centre construction and the enhancement of other related internet infrastructure, providing subsidies for ownership and operation of data centres provides a limited return on investment for local communities,” he stated during his article for Forbes last year.
Citing the ProPublica report that highlighted how Washington State’s tax breaks for data centres spiralled into one of the state’s largest corporate giveaways, Leahey stated that the breaks, which were intended to spur job creation in rural areas and have cost more than $474 million in taxpayer funds since 2018, only ended up with the bulk of these benefits being accrued to Microsoft—not local communities.
“In Washington, the exact number of jobs created is not a matter of public record, but, according to ProPublica, an array of data centre projects with a total taxpayer cost of $53.3 million would only need to collectively hire 260 people to meet the required threshold. That is an average cost of $205,000 per job or about three years’ salary at the median income in the state,” he remarked.
Data centres not only fail to create jobs in significant numbers, but they are huge drains on local infrastructure. Everyone is aware that they use a lot of electricity, but few realise they also need tremendous amounts of water to operate. This makes data centres located in water-scarce environments particularly expensive from a resource perspective and puts further emphasis on their ineffectiveness as job creation mechanisms.
Data centres do not serve as permanent and ongoing job creation engines any more than the construction of a highway or a bridge does. According to Leahey, it is more appropriate for public subsidies to focus on the construction and development of these facilities and the infrastructure required to make use of them, rather than their ongoing operation and ownership. Investing in the construction and laying out of telecommunications infrastructure can be justified in the broader context of development in underserved areas.
Subsidising ownership of data centres, through property and sales tax breaks, for example, is less defensible. The tech companies that dominate the market for these centres are among the most valuable companies in the world, with market caps that regularly dwarf the GDP of the states they are asking to foot the bill. Leahey believes these corporations have ample resources to manage their own operational costs without public support.
“State governments should therefore reconsider their approach to supporting data centres, focusing on subsidies for the construction phase and overall improvement of internet infrastructure. Simultaneously, states must ensure that these investments are tied to clear public benefits such as job creation in the construction industry, environmental sustainability, and enhanced connectivity for under-connected communities,” he opined.
Incentives vs impact
Stephen Hartka, vice president of research at the Virginia Economic Development Partnership, says that the “overwhelming majority” of data centres in Virginia were brought to the state by incentives. He argues that economists who only look at state tax revenues miss the fact that Virginia’s tax structure sends more benefits to smaller localities, which especially benefit from the property taxes that data centres pay.
He also says that data centres can be particularly valuable in rural areas that have struggled to attract businesses.
“Data centres are a very good fit for these rural areas because they don’t require a lot of people, and they generate a ton of revenue for local governments. They are financial boons to these communities that might not have many other options,” he noted.
However, there has been overwhelming activism against data centres in many Virginia communities. In 2024, Ian Lovejoy, a Republican state delegate in Virginia, told TIME that data centres were the “number one issue we hear about from our constituents.” Residents expressed concerns about data centres threatening electricity and water access, and worried that taxpayers may have to foot the bill for future power lines. Across the country, communities have complained about data centres emitting pollutants and harming the quality of life.
The direction of data centres (in terms of making the US an AI powerhouse) and their national expansion is still hazy. However, policy experts wish that more data were provided about the massive trade-offs that states are making and whether they are worth it. And the demand is not a wrong one either.