The "tax-free dividend" for data centers changes! Over ten US states reassess tax incentives, AI computing power expansion faces new obstacles

According to Zhitong Finance APP, tax incentives that are crucial to the pace of data center construction across the United States are now under renewed scrutiny, with more than ten states reassessing or reducing subsidies, while more than 35 states still offer relevant benefits. Last year, Ohio’s data center tax exemption exceeded $1.5 billion, more than ten times the original forecast, prompting some legislators to call for the repeal of these incentives and the renegotiation of long-term agreements.

Amazon, the e-commerce and cloud computing giant leading the unprecedented wave of AI data center construction in the US, emphasizes that it has invested nearly $40 billion and created thousands of jobs in Ohio since 2015. The core dispute between US technology giants and state governments centers on how governments further weigh investment, jobs, and tax revenue against the public utility costs brought by data center expansion. For the tech giants, uncertainty in long-term tax arrangements is becoming a new variable in returns on capital expenditures.

For Amazon, Meta, and Google, the direct impact of shrinking tax incentives is higher construction costs and pressured after-tax project returns. Once data center server clusters and related equipment lose sales tax exemptions, cash outlays for equipment procurement and upgrades rise; if firms also need to bear more expenses for electricity and grid construction, the payback period for data center investments may further extend. However, Ohio's suspension of new applications is a different stage of weighing than the legislators' proposals to repeal incentives and renegotiate existing contracts—it does not currently mean technology giants have lost all long-term tax advantages.

The release of Astra has fueled market expectations for Artificial General Intelligence (AGI), bringing even stronger computing demand trajectories, especially as "outcome-based charging" emerges as a new growth model that could drive overall demand for computing power. More directly, recent evidence of heightened demand comes from the AI development process itself—namely, the "Recursive Self-Improvement (RSI)" trajectory where AI starts "building AI."

With Astra driving demand for advanced, cutting-edge large models and growing AI compute requirements, Morgan Stanley forecasts that combined capital expenditures for North America's four super cloud computing and AI application vendors will rise from $917 billion in 2026 to $1.47 trillion in 2027 and $1.64 trillion in 2028, with deployed capacity expected to expand from 35GW in 2025 to 145GW in 2028 over the same period.

As states re-examine tax incentive agreements, Amazon, Meta, and Google face new risks

More than ten states are reassessing or reducing tax subsidies for data centers. These facilities, once welcomed as key engines driving the accelerated expansion of the AI economy, are increasingly seen as political risks by local leadership. Consequently, as opposition against these centers grows, America’s largest tech giants—Amazon (AMZN.US), Meta (META.US), the parent company of Facebook, and Google (GOOG.US) (GOOGL.US)—face the risk of losing decades-long tax exemptions.

Over a decade ago, Ohio legislators exempted tech companies from paying sales tax on computer server purchases and other equipment for data centers, betting that these multimillion-dollar tax adjustments would draw critically needed investment. The initiative paid off, turning Ohio into a major hub for data centers. However, reports indicate that the recent AI boom has dramatically inflated the scale of exemptions, which last year topped $1.5 billion—over ten times the state's initial estimate.

Now, some state lawmakers, including Democratic state representative Tristan Reit, are pushing to scrap sales tax exemptions and renegotiate prior agreements with tech firms such as Amazon, Meta Platforms, and Alphabet’s Google. These companies secured decades-long tax breaks by signing contracts with state governments. According to reports, in several states, major tech companies enjoy more than $1 billion in annual tax exemptions.

"They seem wealthier than God and are capable of constructing these facilities without such incentives," Reit said. He is proposing new taxes on data centers and wants developers to shoulder more of the costs for power and infrastructure. His district covers parts of Cleveland, where residents have voiced opposition to these centers. He hopes these new pressure tactics will force companies to the negotiating table.

In a statement, Amazon noted that since 2015 the company has invested nearly $40 billion in Ohio data centers, creating thousands of jobs, and paid nearly $11 million in property taxes and related fees in the state last year. Meta and Google declined to comment, as did the governor’s office.

The attacks on tax incentive provisions are the latest example of mounting opposition to the AI boom across the US. Media reports say that John Perkins, a Kansas City, Missouri councilman who previously voted to approve billions in tax breaks for a data center, was recently defeated in an election.

Although recent polls show that the vast majority of Americans do not want data centers built near them, President Trump continues to urge voters to welcome such facilities. "If you want your state or town to get richer, pay less tax, generate enormous wealth, and see your home and other assets become more valuable, you'll want a large AI data center," he said Friday. "If you want to live in poverty, crime, and filth, I'd say don't approve the construction of data centers."

Media reports cite Ian Boccaccio, a veteran researcher at tax services firm Ryan, as saying that Ohio, Arizona, and Illinois have become less attractive for investment after signaling intentions to pause or eliminate some exemptions. He says he is urging data center clients to consider other tax incentives offered by states. Industry observers broadly say that, ultimately, the practical and future benefits of these facilities are so significant that governments find it difficult to turn them away.

How does this impact the global AI compute investment theme?

Resistance to data centers is shifting AI investment headwinds from chips and power supply to construction permits and community acceptance. A report by Annenberg in August, based on actual surveys from June to July, found that the proportion of adults opposed to new local data centers rose from 49% to 61%, though public attitudes toward AI's overall impact were largely unchanged. Meanwhile, Data Center Watch confirms that in Q1, at least 75 projects valued at approximately $130 billion were stalled or delayed. For the AI industry, the direct result is greater uncertainty around the commissioning of new computing power, which could delay equipment delivery, cloud service expansion, and revenue recognition for AI—a key market focus—potentially leading to a sector correction as compute-related valuation compresses.

The economic essence of the debate is who ultimately bears the costs of new power generation, transmission, and water supply. Although tech companies have signed the “Electricity Payer Protection Commitment,” cost-sharing disputes persist. September reports say Microsoft is appealing against a new rule in Virginia requiring data center developers to pay upfront for transmission infrastructure.

This case illustrates that making companies pay for their own power involves timing of payments, scope of infrastructure, and risk allocation. If more expenses must be prepaid by developers, the initial capital outlay and financing needs for projects increase; if costs are passed to residents, political pushback may persist. How commitments are written into power agreements and construction terms will directly affect data center investment returns.

For the AI compute investment theme, policy changes first impact where new centers are built, at what cost, and how investment returns are realized. For example, Ohio’s incentives cover computing equipment, cooling systems, electrical equipment, and some construction materials, so any adjustment might affect the entire data center campus. Based on this cost transmission mechanism, Amazon AWS and Google Cloud may need to reassess the costs of expanding available compute, while Meta faces shifts in infrastructure costs that support advertising recommendations and AI products. However, capital expenditures might not decline, but instead shift regionally, and it’s crucial to note that higher tax burdens do not in themselves prove weaker end-user AI demand.

The potential for a medium- and long-term positive push depends on whether increased costs can deliver clearer construction and access terms—if cost sharing becomes clear and projects secure reliable power, companies can more easily plan commissioning and strong AI-related revenue timelines. AEP Ohio has already implemented dedicated data center tariffs and behind-the-meter grid systems, advancing new projects through phased studies, contract signings, and well-defined expected power delivery dates. This demonstrates, to some extent, that having tech firms take on more infrastructure responsibility can occur alongside continued project expansion.