Inside the AI boom: Why massive investment doesn't fully translate into GDP growth

AI’s most immediate macroeconomic impact has come through spending, not productivity. In 2025, U.S. private investment surged, led by extraordinary growth in IT and AI-related capital expenditure. Spending on servers, computing equipment, and data-center infrastructure expanded at a pace not seen in decades, giving a substantial boost to aggregate demand at a time when other investment categories were weakening.

Artificial intelligence (AI) has rapidly become one of the most powerful forces shaping the U.S. economy, driving record levels of corporate investment and reshaping expectations about growth, productivity, and competitiveness. Yet behind the surge in headlines and capital spending lies a more nuanced economic reality, one that is only partly visible in conventional growth statistics.

In a study “Artificial Intelligence and the US Economy: An Accounting Perspective on Investment and Production,” released as an arXiv working paper, researchers take a macro-accounting approach to assess how the current wave of AI investment and production is actually reflected in U.S. GDP. The study focuses on what can already be measured in national accounts and what that measurement reveals about the true scale and structure of AI’s economic footprint.

AI spending is lifting demand, but imports dilute GDP gains

AI’s most immediate macroeconomic impact has come through spending, not productivity. In 2025, U.S. private investment surged, led by extraordinary growth in IT and AI-related capital expenditure. Spending on servers, computing equipment, and data-center infrastructure expanded at a pace not seen in decades, giving a substantial boost to aggregate demand at a time when other investment categories were weakening.

This investment surge has fueled narratives that AI is single-handedly propping up U.S. growth. However, the study shows that such claims overstate the domestic economic impact once accounting realities are taken into consideration. A large share of AI-related hardware, including servers and accelerators, is imported. While U.S. firms dominate chip design, cloud platforms, and AI software, much of the physical manufacturing and assembly of AI hardware occurs abroad, particularly in Asia and neighboring economies.

From a GDP accounting perspective, imports subtract from domestic output. As a result, the contribution of AI-related investment to U.S. GDP growth is significantly smaller than headline capital expenditure figures suggest. When the import content of AI hardware is netted out, the study finds that AI-driven investment still makes a meaningful contribution to growth, but not at the scale implied by raw spending data.

This distinction matters for policymakers and investors alike. High investment volumes can coexist with more modest GDP gains if much of the value added accrues to foreign suppliers. The study emphasizes that AI investment has supported demand and economic activity, but it has not replaced other drivers of growth, particularly household consumption, which remained a major contributor to U.S. GDP expansion in 2025.

At the same time, the authors note that official statistics may understate certain aspects of AI investment. Some advanced servers may be misclassified, and rapid price-performance changes in AI hardware complicate measurement. Even with these caveats, the overall conclusion remains that AI investment’s direct contribution to GDP is substantial but more limited than popular narratives imply.

Data centers sit at the core of the AI economy

While AI-related hardware imports dilute some of the GDP impact of investment, the study identifies data centers as the central mechanism through which AI spending is transformed into domestic economic output. Data centers are described as the backbone of the AI ecosystem, serving as the physical infrastructure that enables model training, inference, cloud services, and the delivery of AI-enabled products.

The construction and equipping of data centers generate immediate economic activity through demand for construction services, electrical systems, cooling infrastructure, and networking equipment. These activities are recorded as investment in national accounts and contribute directly to GDP through domestic value added.

More importantly, once data centers become operational, they generate ongoing flows of computational and AI services. These services enter GDP through multiple channels. When sold directly to consumers, businesses, governments, or foreign users, they contribute to final demand through consumption, investment, government spending, or exports. When used as intermediate inputs by other industries, they reshape cost structures and production processes, even if they do not immediately appear as final GDP components.

The study argues that, at present, a large share of AI services is sold directly to final users rather than being embedded deeply across all sectors of the economy. This means that AI services surface in GDP statistics relatively quickly compared to past technological revolutions, where productivity gains took years to materialize.

Given current utilization rates and pricing, revenues from AI services produced by new data centers could soon match the scale of the capital expenditures required to build them. In practical terms, this implies that AI infrastructure investments may generate substantial GDP contributions within a short time frame, not just through construction but through service production and sales.

This revenue channel helps explain why value-added measures of GDP show a strong contribution from technology-intensive sectors, even as import-adjusted investment figures appear more modest. Growth in sectors related to computing services and digital infrastructure reflects the income generated by operating AI systems, not just the cost of building them.

Short payback periods and medium-term risks

While the study highlights the economic momentum created by AI investment and production, it also flags potential risks that could shape the medium-term outlook. One concern relates to the rapid pace of reinvestment required in AI infrastructure. AI hardware evolves quickly, and servers and accelerators may become technologically obsolete within a few years. This raises questions about depreciation, capital lifetimes, and the sustainability of returns on investment.

The authors argue that fears of widespread overinvestment may be overstated in the short term. Current evidence suggests that AI data centers are operating at high utilization rates, with demand for computational capacity still exceeding supply. At prevailing prices, the payback period for new AI infrastructure appears relatively short, often on the order of a year, which mitigates some concerns about stranded capital.

However, frequent replacement cycles could have important implications for corporate cash flows. Even if gross revenues rise rapidly, the need for continuous reinvestment may limit growth in net income and free cash flow over time. This dynamic could affect valuations, investor expectations, and the distribution of gains from the AI boom.

Another source of uncertainty is future demand for AI services. Adoption has occurred at an unprecedented pace, with generative AI reaching a global user base in just a few years. Such rapid growth makes forecasting difficult. Overinvestment could lead to excess capacity and price pressure, while underinvestment could constrain service quality and limit economic benefits.

From a macroeconomic perspective, these uncertainties highlight the importance of distinguishing between short-term demand effects and longer-term structural change. The study deliberately sets aside productivity and labor market impacts, noting that these second-order effects will take time to emerge and will ultimately determine whether AI delivers sustained growth rather than a temporary spending-driven boost.

Monetary authorities should be cautious about attributing economic resilience solely to AI, while recognizing that AI-related investment and production can meaningfully affect demand and inflation dynamics, the study notes. Fiscal and industrial policymakers, meanwhile, face choices about how to encourage domestic value creation within global AI supply chains, particularly in areas such as data-center infrastructure, skills, and high-value services.

The study also sheds light on the importance of better data. AI activity is currently embedded within broad technology categories, making it difficult to isolate its specific contribution to growth, trade, and productivity. Improved statistical classification and greater use of firm-level and sectoral data will be essential for tracking AI’s economic impact as adoption deepens.