Problem

The paper investigates how the burgeoning artificial intelligence (AI) sector, particularly its capital investment and production, is being quantified within US national accounts. It acknowledges that AI-related investments, especially in data centers, have been cited as a primary driver of US economic growth in 2025. However, accurately isolating AI's specific contribution is challenging due to its integration into broader technology aggregates and the difficulty in obtaining granular firm-level data. The study aims to provide a clear assessment of AI's economic footprint and the specific channels through which it influences measured US Gross Domestic Product (GDP), emphasizing the distinction between domestic value added and foreign content in AI supply chains.

Method

The authors employ a macro-accounting framework combined with a detailed description of the AI production ecosystem to analyze its impact on the US economy. This ecosystem is broken down into three main categories of actors: hardware vendors (responsible for designing and manufacturing AI chips and servers), cloud infrastructure providers (who own and operate the extensive data centers), and AI labs (which train frontier models and offer derived services). A key aspect of the methodology involves examining the geographical distribution of these production stages to discern the proportion of value creation that accrues domestically versus internationally. The paper then quantifies the effect of IT-related capital expenditure on US demand and output, carefully adjusting for imported components and sector-specific value added. Finally, it details how the output of AI services, largely facilitated by data centers, contributes to GDP either as final consumption/investment/exports or as intermediate inputs for other industries.

Results

The paper identifies three main findings regarding AI's macroeconomic footprint. First, while IT and AI-related capital expenditure significantly boosted aggregate demand in the first three quarters of 2025 (with IT-related capex growing 36%, 20%, and 5% annually), its net contribution to US GDP growth was more limited. This is largely because a substantial portion of AI hardware is imported, particularly for chip manufacturing and server assembly, diminishing the domestic value added. After accounting for imports, AI-related investment contributed approximately 0.3 to 0.6 percentage points to GDP growth across quarters, averaging about 20% of overall GDP growth in the first nine months. Second, data centers are central to the AI ecosystem, serving as the nexus for computational power for AI model training, inference, and cloud services. Their construction and equipment procurement generate significant aggregate demand, while the computational services they provide contribute to GDP as either final consumption or intermediate inputs. Third, new AI data centers, operating at high utilization rates and current GPU-based service pricing, are projected to generate revenues that could support GDP increases of a magnitude comparable to the initial capital expenditures. Calculations suggest a payback period of around one year for AI data center investments, implying substantial GDP contributions from service revenues within months. This is supported by sectoral value-added data showing a 0.5 to 0.7 percentage point contribution from tech-intensive sectors to GDP growth in the first half of 2025, and growing exports of computer services.

Implications

The findings have several implications for policymakers. For monetary policy, they suggest that while AI is a significant factor, the resilient US consumer remains a central driver of economic dynamics, cautioning against solely attributing US economic strength to AI. For fiscal and industrial policy, the analysis underscores the importance of fostering high-value segments of the AI value chain domestically, highlighting the role of infrastructure, skills, and regulatory frameworks. The paper also discusses macroeconomic risks: while concerns about rapid hardware depreciation due to obsolescence are mitigated by short payback periods and chip redeployment, the frequent reinvestment cycles could still constrain free cash flow over the long term. Furthermore, significant uncertainty surrounding future AI service demand poses risks of both overinvestment (potentially leading to market corrections and financial stress) and underinvestment (resulting in lower service quality and competitive disadvantages). The authors conclude by stressing the need for improved statistical data and more comprehensive modeling of AI diffusion to facilitate timely and informed policy decisions in response to AI's accelerating macroeconomic impact.