Candice Horbacz
US companies are on the cusp of investing trillions of dollars in artificial intelligence (AI) over the coming years, a monumental shift poised to reshape global economies and technological landscapes. This burgeoning investment is driving an unprecedented demand for robust infrastructure, primarily in the form of advanced data centers—the physical backbone that stores, processes, and transmits the vast quantities of information essential for AI operations. Financial institutions like Goldman Sachs underscore the scale of this commitment, estimating that capital expenditures from just a handful of major US AI "hyperscalers" could reach a staggering $581 billion in 2026 alone, as part of a global AI-related investment projected to exceed $1 trillion in the same year. While a significant portion of this capital will fuel the development of cutting-edge AI models and applications, a substantial allocation is earmarked for the construction and upgrade of these indispensable data centers.
However, this rapid expansion has not gone unnoticed by policymakers. Members of Congress, increasingly concerned about the multifaceted impacts of these energy-intensive and land-hungry facilities, as well as the broader labor market implications of widespread AI adoption, have begun to introduce legislative proposals aimed at taxing new data centers. These concerns span environmental footprints, including significant land and water consumption, substantial energy demands, and potential disruptions to local communities. Furthermore, the specter of AI-driven automation affecting employment patterns adds another layer of urgency to these legislative efforts. Among the various ideas floated, two recent and prominent proposals—one from Senator Mark Warner (D-VA) and another from Senate Finance Ranking Member Ron Wyden (D-OR)—seek to deny certain AI data centers the use of full expensing for machinery and equipment, a critical tax provision designed to incentivize capital investment. Critics argue that such measures could introduce undue complexity into the tax code, create arbitrary distinctions between investments, and potentially divert crucial AI investment overseas, thereby undermining US competitiveness and denying local communities the economic benefits of these high-tech developments.
The Accelerating Pace of AI Investment and Its Infrastructural Demands
The figures projected by Goldman Sachs highlight a transformative period for the US economy, driven by an estimated $581 billion in AI-related investment from American hyperscalers in 2026 alone. This projection signals not just a technological boom but a massive infrastructural undertaking. Data centers are the physical manifestations of this digital revolution, housing the servers, networking equipment, and power infrastructure necessary to train complex AI models, run sophisticated algorithms, and deliver AI-powered services to users worldwide. These facilities are not merely warehouses for data; they are highly specialized, secure, and energy-intensive environments designed for continuous operation and massive computational loads.
The relentless pursuit of greater computational power for AI development necessitates increasingly larger and more sophisticated data centers. This demand is fueled by advancements in machine learning, deep learning, and generative AI, which require immense processing capabilities to handle petabytes of data for training and inference. The race among tech giants to develop superior AI capabilities directly translates into a race to build and operate more powerful and efficient data centers. This dynamic creates both immense economic opportunities and significant challenges, particularly concerning resource allocation and environmental sustainability.
Growing Congressional Scrutiny: Environmental and Socioeconomic Impacts
The rapid proliferation of data centers, often concentrated in specific regions, has ignited a wave of congressional concern. Lawmakers are scrutinizing the environmental footprint, which includes substantial land acquisition, significant water usage for cooling systems, and massive electricity consumption. For instance, a single hyperscale data center can consume as much electricity as a medium-sized town, drawing heavily on local power grids and contributing to carbon emissions if powered by fossil fuels. The demand for land, particularly in rural areas where data centers are increasingly located dueates to cheaper land and access to power lines, can strain local resources and alter landscapes. A Pew Research report highlighted that most new data centers in the US are indeed being built in rural areas, raising questions about equitable resource distribution and community impact.
Beyond environmental considerations, the economic and social implications are also under review. While data centers bring construction jobs and some highly skilled technical roles, they are not typically large employers once operational, leading to questions about the return on investment for communities that offer tax incentives. Furthermore, Senator Wyden’s white paper specifically links data center expansion to the broader labor market effects of AI, suggesting a need for mechanisms to aid workers potentially displaced by AI adoption. The concern is that while AI promises immense productivity gains, its rapid deployment could exacerbate existing economic inequalities or create new challenges for certain segments of the workforce. These complex issues form the backdrop against which the proposed tax changes are being considered.
Legislative Proposals Emerge: Targeting Data Center Taxation
In response to these escalating concerns, Congress has begun to explore legislative avenues to address the perceived externalities of the AI boom and its infrastructural requirements. Two notable proposals have emerged from influential senators, both targeting the tax treatment of new data centers and specifically the provision of full expensing for machinery and equipment.
Senator Warner’s Data Center Tax Accountability and Disclosure Act (July 2026)
Senator Mark Warner (D-VA) introduced the Data Center Tax Accountability and Disclosure Act in July 2026. This bill proposes a significant alteration to the existing tax code by preventing businesses from claiming full expensing under Section 168(k) for "any property used in an AI data center." The crux of Warner’s proposal lies in its definition of an "AI data center," which is key to determining eligibility for this tax treatment. While the original content does not provide the specific criteria for this definition, it implies a threshold based on the percentage of a data center’s operations dedicated to AI.
Crucially, Warner’s bill includes an incentive for environmental sustainability. Businesses would still be able to claim full expensing for their AI data centers if these structures achieve Leadership in Energy and Environmental Design (LEED) certification at the Platinum or Gold levels. LEED certification, administered by the U.S. Green Building Council (USGBC), is a globally recognized standard for designing, constructing, operating, and maintaining green buildings. By tying tax benefits to specific LEED levels, the proposal aims to encourage data center operators to adopt more energy-efficient and environmentally responsible practices, thereby mitigating some of the ecological concerns associated with their expansion. This aspect of the bill represents a strategic effort to leverage tax policy for environmental outcomes.
Senator Wyden’s Framework for Data Center Taxation (August 2026)
Following Warner’s initiative, Senate Finance Ranking Member Ron Wyden (D-OR) released a white paper in August 2026 outlining his own comprehensive proposal for data center taxation. Wyden’s framework also seeks to deny full expensing to new data centers but goes further by introducing a novel gross receipts tax. While the specific details of the gross receipts tax rate are described as being in the "low single digits," this type of tax can have a disproportionately high impact on a company’s net income, as it applies to total revenue before expenses are deducted.
Wyden’s proposal includes certain exemptions, notably for "internet infrastructure" and "small local data center operators," though these terms remain undefined in the white paper, posing potential challenges for implementation. Furthermore, the framework introduces a "deemed minimum regime" for the gross receipts tax. This regime would compel data center users exceeding specified asset and expense thresholds to pay a minimum tax, which Wyden frames as a measure to combat tax avoidance. This component suggests a broader aim to ensure that even highly complex financial arrangements involving data centers contribute their fair share to the tax base. The combination of denying full expensing and imposing a gross receipts tax signifies a more aggressive approach to extracting revenue from the burgeoning data center industry.
Unanswered Questions and Administrative Hurdles
Both the Warner and Wyden proposals, despite their detailed objectives, currently suffer from significant definitional ambiguities that could complicate their administration and compliance. For instance, Senator Warner’s bill does not explicitly detail how the "20 percent AI threshold" for determining an AI data center would be calculated or verified. This lack of clarity could lead to disputes, require extensive guidance from the Treasury Department and the IRS, and potentially create loopholes or unintended consequences. How would a data center with fluctuating AI workloads be categorized? What metrics would be used to assess the 20 percent threshold—power consumption, server allocation, or data throughput? These practical questions are critical for effective implementation.
Similarly, Senator Wyden’s white paper leaves crucial terms undefined. The exemptions for "internet infrastructure" and "small local data center operators" lack clear parameters, which could lead to confusion and inconsistencies in application. Without precise definitions, businesses may struggle to determine their tax obligations, and the IRS could face significant challenges in enforcement. Moreover, the "deemed minimum regime" for the gross receipts tax, targeting "complex financial arrangements [with] a data center," is broad and could cast a wide net, potentially subjecting a diverse array of businesses to unexpected tax liabilities. These definitional gaps underscore the need for substantial regulatory clarification before such proposals could be effectively integrated into the tax code.
Economic Implications and Policy Debates
The proposed changes carry substantial economic implications, prompting a robust debate among economists, industry leaders, and policymakers. The core of this debate revolves around the potential impact on investment incentives, US competitiveness, and the efficiency of the tax system.
The Peril of Gross Receipts Taxes
Wyden’s proposal for a gross receipts tax, even if pitched in the "low single digits," presents a significant concern for businesses. Unlike a net income tax, which applies to profits after expenses, a gross receipts tax applies to total revenue. As illustrated by economic analysis, a seemingly modest 4 percent gross receipts tax can translate into a dramatically higher effective tax rate on net income, potentially reaching 40 percent or even 200 percent for firms with thin profit margins. This is particularly problematic for capital-intensive industries like data centers, which often incur massive upfront costs and ongoing operational expenses. Such a tax can disproportionately burden new businesses or those undergoing significant expansion, potentially stifling growth and investment. It also creates a "tax on a tax" problem, as intermediate goods and services may be taxed multiple times along the supply chain.
Risk of Capital Flight and Eroding US Competitiveness
A major concern shared by critics of both proposals is the potential for driving AI investment overseas. Neither the Warner nor Wyden proposals capture non-US investment in data centers. By increasing the cost of capital and imposing new taxes on domestic AI infrastructure, these measures could make the United States a less attractive location for data center development and, by extension, broader AI innovation. The US currently holds a projected advantage in attracting global AI investment, a position that could be eroded if its tax policies become less favorable than those of other nations. This potential "capital flight" could lead to a loss of high-tech jobs, diminished economic growth, and reduced tax revenues for US towns and cities, directly counteracting some of the intended benefits of these legislative efforts. The global nature of AI development means that capital and talent are highly mobile, and tax policy plays a significant role in investment location decisions.
The Broader Tax Landscape and Economic Literature
A fundamental premise underlying these proposals is that the current US tax system is ill-equipped to capture the economic gains from AI investment and adoption. However, this premise has been widely scrutinized by tax experts across the ideological spectrum. Existing tax mechanisms—including corporate income taxes, capital gains taxes, and property taxes—are generally considered capable of capturing any "supernormal returns" generated by AI companies and the data centers they operate. As AI companies become more profitable, their corporate income tax liabilities increase. As their valuations rise, capital gains taxes are realized by investors. And as data centers are built, they contribute to local property tax bases. These existing taxes inherently collect revenue from economic activity, including that driven by AI.
Furthermore, a vast body of economic literature consistently finds that the benefits of new technologies are not primarily captured by the innovators themselves but rather dispersed broadly throughout the economy. This diffusion of benefits, often termed "spillovers," is a common justification for favorable tax treatment of research and development (R&D) activities, which are seen as foundational to technological advancement. The proposals from Wyden and Warner, by increasing taxes on AI infrastructure, represent a stark contrast to this long-standing policy approach of incentivizing technological innovation through the tax code.
The Role of Full Expensing
Denying full expensing, also known as bonus depreciation, for certain investments is primarily a timing shift rather than a long-term increase in tax revenue. Full expensing allows businesses to immediately deduct the full cost of certain capital investments, such as machinery and equipment, in the year they are made. This accelerates cost recovery, reducing the effective cost of investment and encouraging businesses to invest more in productive assets. While disallowing full expensing would bring some tax revenues sooner, it does not raise sustained long-run revenue for the federal government. Instead, it discourages new investment into revenue-generating economic activity, like the construction and equipping of data centers, by increasing the after-tax cost of capital. Over the long run, this can lead to a smaller capital stock, reduced productivity, lower wages, and a smaller overall economy.
Tax Foundation’s Revenue and Economic Impact Analysis
The Tax Foundation, utilizing its sophisticated Taxes and Growth (TAG) Model, has provided a central estimate for the revenue and economic impact of Senator Warner’s proposal to disallow bonus depreciation for qualified data centers. Their analysis projects that this measure would raise approximately $29.9 billion from 2027 to 2036 on a conventional basis. This revenue projection is based on the assumption that firms would shift from immediate deduction to depreciating these assets over their useful life, accelerating tax payments.
However, the economic repercussions are also significant. The proposal is estimated to reduce the long-run size of the US economy by less than 0.05 percent. This slight contraction is attributed to an increase in the cost of capital for businesses investing in affected data centers, as they can no longer fully recover their investment costs immediately. On a dynamic basis, which accounts for changes in economic behavior due to the tax policy, the proposal is estimated to raise $18.2 billion over the same 10-year period. This lower dynamic revenue reflects the impact of a smaller economy, which generates less taxable income and payroll tax revenue.
The accuracy of these estimates is subject to several variables, including the precise trajectory of future data center investment, the proportion of data centers that will meet the 20 percent AI usage criteria, and the adoption rate of LEED certification. To account for this inherent uncertainty, the Tax Foundation modeled three distinct scenarios: low, central, and high, reflecting varying amounts of investment basis affected by the proposal.
In the low scenario, where $37.9 billion in investment is initially bonus-ineligible in 2027, the conventional revenue raised over 10 years is projected at $17.5 billion. Conversely, a high scenario, beginning with $101.1 billion in affected investment in 2027, would raise $46.5 billion in conventional revenue over the same period. While the economic harm increases with the higher scenarios, it consistently remains below 0.05 percent of GDP in the long run.
A crucial finding across all scenarios is that none result in sustained increases in federal tax revenues over the long term. This reinforces the understanding that denying bonus expensing primarily functions as a timing shift—it accelerates the collection of existing tax revenues rather than creating new, enduring revenue streams. The analysis underscores the trade-off between immediate revenue gains and potential long-term economic dampening effects.
Modeling Notes and Assumptions
The Tax Foundation’s model for Warner’s proposal assumes that a share of economy-wide information and communication technology (ICT) investments would become ineligible for bonus depreciation. Based on BEA detailed fixed-asset data, approximately 80 percent of data center equipment investment consists of ICT. The model projects a near-term AI data center investment boom, with a 20.5 percent annual growth rate in bonus-ineligible investment through 2030, followed by a gradual decline back toward the 2027 share of ICT made bonus-ineligible by 2036. These figures are sensitive to the share of data center investment qualifying for LEED certification and the interpretation of the 20 percent AI threshold under the bill.
Industry and Expert Reactions
While specific statements from industry groups or academic experts are not provided in the original text, it can be logically inferred that such proposals would elicit strong reactions. Tech industry associations, representing the major AI hyperscalers and data center operators, would likely express concerns about the increased cost of doing business in the US, the potential for stifling innovation, and the administrative burdens of complex new tax rules. They might argue that punitive taxes could disadvantage American companies in the global AI race and lead to job losses in related sectors. Environmental groups, on the other hand, might laud the proposals, particularly the LEED certification incentive, as a step toward more sustainable technological growth, while potentially pushing for even stricter environmental mandates. Economic policy experts, as highlighted in the article, would continue to debate the efficacy of targeted taxes versus the existing, broader tax framework in capturing AI-driven wealth, emphasizing the importance of tax neutrality to avoid market distortions.
Looking Ahead: The Future of AI Taxation
The legislative proposals from Senators Warner and Wyden mark the beginning of a critical debate on how the US tax code should adapt to the rapid emergence of artificial intelligence and its demanding infrastructure. Policymakers face a delicate balancing act: addressing legitimate concerns about environmental impact, community burden, and labor market shifts, while simultaneously fostering an environment conducive to technological innovation and economic growth.
Using the tax code to selectively target returns from AI in a way that discourages investment in data centers and AI development carries significant risks. Denying full cost recovery for capital investments and imposing new excise taxes, such as a gross receipts tax, could introduce substantial complexity, distort crucial investment decisions, and ultimately risk pushing valuable economic activity and innovation abroad. A neutral tax code that allows businesses to fully recover their investment costs, thereby encouraging capital formation, while taxing the resulting profits through established mechanisms, is generally considered better suited to capturing the economic gains from AI without inadvertently undermining US competitiveness and leadership in this transformative field. The ongoing discussions will shape not only the future of AI infrastructure but also the broader economic landscape for decades to come.
