A Federal Reserve staff research paper finds that demand for artificial intelligence (AI) has fueled a surge in U.S. data center investment and pushed up prices for some tech inputs, yet its effects on overall productivity and employment remain limited. That means the current AI boom's boost to the U.S. economy mainly shows up as investment growth from data center construction and purchases of chips and related equipment, while whether AI technology can truly translate into broad-based corporate efficiency gains and a productivity dividend still needs time to be tested.
The paper, titled "Artificial Intelligence and the U.S. Economy: Current Evidence," was written by Nick Carollo, Leland Crane, Daniel Garcia Molina, and Joseph Nichols and published in November. The authors note that while AI is very likely to generate large and lasting productivity effects over the long run, only a few years have passed since ChatGPT's launch, and technology diffusion, corporate organizational adjustments, and the realization of economic effects remain at an early stage. It should be noted that this is a discussion paper by Fed staff, and its analysis and conclusions represent the authors' views and do not necessarily reflect agreement by the Federal Reserve Board or other staff.
Data center investment plans top $1 trillion, but import spending offsets part of the economic boost
The paper finds that planned investment in new data centers in 2025 exceeded $1 trillion, up from nearly $300 billion in 2024. Based on projects announced as of May 2026, under a central scenario, the annualized pace of data center-related investment is expected to exceed $880 billion by the end of 2027, equivalent to more than 2% of GDP. Estimates from Morgan Stanley show that cash capital expenditures by hyperscalers and neoclouds are expected to keep growing, rising from $946 billion in 2026 to $1.66 trillion in 2027 and further to $1.77 trillion in 2028. Alphabet (GOOGL.US) and Amazon (AMZN.US) account for the largest shares of projected spending. Combined cash capital expenditures for the two companies are expected to reach $730 billion in 2027 and about $765 billion in 2028.
However, the scale of data center investment does not mean an equivalent increase in domestic economic output. The authors estimate that about half of related spending flows to imported goods, especially high-tech equipment, so higher imports will offset part of the boost to U.S. domestic output. At the same time, data center construction may also drive supporting investment such as power infrastructure, but rising prices for high-tech equipment could squeeze other companies' investment budgets. There is also uncertainty around whether projects will actually be completed. Data center construction requires land, power, water, labor, and large amounts of equipment, and shortfalls in any one of these resources could raise construction costs and delay project timelines. Some local governments have already restricted data center projects because of electricity and water demand, while grid connection queues, shortages of natural gas turbines, and chip supply constraints also pose potential bottlenecks. The paper notes that delivery times for some natural gas turbines have been extended from two years to more than five years.
AI construction boom pushes up prices of some tech products, with inflation effects notable but limited
The data center construction boom has also driven up prices for products such as memory chips, data storage devices, bare circuit boards, and assembled circuit boards. In the 12 months through March 2026, prices for a category of personal consumption expenditures (PCE) covering video, audio, photographic, and information processing equipment and media rose nearly 8%, whereas in the prior three years prices for these products fell by about 3% per year on average. The paper notes that higher software and accessory prices were also an important factor in the category's price increase, and some software firms may be passing on the cost of new AI features to users. Higher prices for data storage devices also contributed to related consumer prices. However, the authors believe that so far the impact of these price increases on the overall PCE price index remains clear but limited, while future effects could expand further. Related research also found that since January 2025, the core PCE price categories most exposed to data centers have had inflation about 5 percentage points higher than categories at the median level of exposure. This means AI infrastructure buildout may, in the short term, push up some costs through chips, electronic devices, software, and electricity, while the cost-reducing effects of productivity gains have not yet fully appeared. Fed researchers believe that if AI can significantly raise productivity in the future, it could indeed ease inflation by lowering corporate production costs, but that effect cannot yet be observed.
AI use keeps expanding, but depth of corporate adoption remains insufficient
AI use among consumers and businesses continues to expand, but adoption remains uneven. A 2025 survey found that about 50% of adults use AI outside of work, mainly for information gathering and writing; multiple surveys as of early 2026 showed that more than 40% of workers use AI on the job. However, only about 10% of workers use AI daily at work, indicating the technology has not yet become a core part of most employees' daily workflows. Business-level survey results vary widely. The U.S. Census Bureau's Business Trends and Outlook Survey shows that about 20% of firms adopt AI, while other studies put the figure at 58% to 78%. This variation is related to the specific wording of survey questions, sample scope, and statistical weighting. Overall, larger firms and younger firms have higher AI adoption rates, and sectors such as information services, financial activities, and professional and business services are relatively ahead in application. Still, firms often use AI only in a few business functions, such as IT, human resources, marketing, finance, and strategic management, rather than fully integrating it across the organization. Notably, AI adoption in research and development remains relatively low, even though R&D may be an important channel through which AI drives long-term innovation and productivity growth. Fed researchers believe that to achieve sustained and broad-based productivity gains, firms may need to further incorporate AI into core production processes, redesign organizational structures, automate more work tasks, and invest in supporting resources such as training and management. Simply buying AI tools or giving employees access is not enough to guarantee a large productivity increase.
AI's impact on the job market is emerging, but the overall effect remains hard to quantify
AI's impact on the labor market remains mixed. One study cited in the paper found that after the launch of large language models, employment among workers aged 22 to 25 in occupations highly exposed to AI fell 16% compared with same-occupation workers in other age groups and workers in occupations with lower AI exposure. Other research has reached similar conclusions. The authors note that this effect mainly reflects reduced hiring demand rather than mass layoffs. Workers early in their careers may be more vulnerable because their tasks often include more simple, standardized, and easily codified work, while experienced employees often possess expertise and firm-specific experience that is harder for AI to replace. At the same time, AI may also lead to job reallocation. Some firms prefer to retrain affected employees rather than lay them off directly. A New York Fed survey found that when facing roles potentially affected by AI, firms reported retraining workers more often than layoffs. Therefore, AI's impact on overall employment and unemployment remains difficult to identify. The occupational groups currently most affected account for a relatively limited share of total U.S. employment, but as the scope and depth of corporate AI adoption continue to expand, this impact could become more evident in the future.
Comments