Skip to main content
v2026.11,986 entries · CC-BY 4.0

Editorial · CASRAI · AI and ML research outputs

Federal Reserve Working Paper Tracks Surging AI Data-Center Investment Against Modest Productivity Gains

Federal Reserve working paper FEDS 2026-074 documents surging AI data-center investment alongside muted aggregate productivity and employment effects.

Published 15 Nov 2026· 7 minute read

Ask CASRAI · free to try

Ask about this story

Ask your first 2 questions free below. Subscribers get 150 a day for $29 a month.

An AI assistant specialized in research administration. It cites the sources behind every answer, labels web answers and says when it can't answer.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.

CASRAI is the reference for research administration — bookmark it for the next question.

On November 12, 2026, the Federal Reserve Board published a comprehensive Finance and Economics Discussion Series working paper documenting that while U.S. capital commitments for artificial intelligence data centers have surged past $1 trillion, aggregate macroeconomic productivity gains and labor market disruptions remain muted. For university research leaders, economics scholars, and facility directors, the empirical findings highlight severe upstream hardware inflation, substantial import leakage, and mounting compute procurement pressures across higher education.

FEDS 2026-074: Quantifying the Capital Boom in Artificial Intelligence

Authored by Federal Reserve Board economists Nick Carollo, Leland D. Crane, Daniel Garcia Molina, and Joseph Nichols, working paper FEDS 2026-074—titled Artificial Intelligence and the U.S. Economy: Current Evidence (DOI: 10.17016/FEDS.2026.074)—evaluates how generative AI infrastructure is rippling through national accounts. Crucially, this working paper investigates real macroeconomic variables, capital accumulation, and industry output, keeping its analysis distinct from the central bank’s semi-annual Federal Reserve Financial Stability Report, which focused on debt liquidity and bank credit exposures.

The paper documents a massive escalation in planned domestic data-center construction, which climbed from roughly $300 billion in 2024 to more than $1 trillion in announced pipelines by 2025. Applying the project forecasting methodology from Brandsaas et al. (2025), the authors estimate that baseline annual data-center investment will step up from $100 billion in 2024 to over $880 billion by the end of 2027, exceeding 2.0% of U.S. GDP. Approximately one-third of this outlay funds physical structures, while two-thirds is absorbed by high-tech equipment including GPUs, TPUs, high-density memory, and server cooling systems.

Data-Center Capex Projections, Import Leakage, and Component Inflation

The pace of data-center capital deployment has generated distinct supply-chain pressures and macroeconomic offsets. Key benchmarks established in FEDS 2026-074 include:

Economic Dimension Empirical Benchmark / Metric Analytical Context
Annual Data-Center Capex $100B (2024) to $880B+ (2027 Projected) Baseline model projects annual outlays to exceed 2% of U.S. GDP.
Capex Cost Distribution 33% Structures / 67% High-Tech Equipment Two-thirds of capital outlays concentrate in chips and server hardware.
Import Leakage Share Approximately 50% of IT Equipment Heavy foreign hardware sourcing dampens net domestic GDP multipliers.
Upstream Price Pressures 7.8% Annualized Hardware Price Inflation Memory chip and server board spikes halted long-term price deflation.
Frontier AI Lab Revenues ~$40B (OpenAI) / ~$65B (Anthropic) Run Rates Foundation model providers capture ~19% of U.S. software gross output.
Workplace Daily AI Adoption 10% of Surveyed U.S. Workforce Broad experimentation contrasts with limited intensive daily integration.

Crucially, Carollo and colleagues calculate an import share of roughly 50% for high-tech compute hardware, with components primarily sourced from East Asian fabrication and packaging corridors. Because imported equipment enters national accounts as a deduction, this substantial leakage moderates the direct domestic output multiplier.

Concurrently, direct inputs like high-bandwidth memory (HBM) and specialized printed circuit boards experienced annualized price increases of nearly 8% through early 2026, reversing a multi-decade trend of technological price deflation. Surging electricity requirements also prompted a sharp rise in power-generation interconnection requests in late 2025. These local infrastructure dynamics align with NBER research by Alvarez et al. (2026), which shows that data centers generate brief construction bursts but minimal long-term local employment.

The Solow Productivity Paradox in the Generative Era

Despite heavy capital commitments, FEDS 2026-074 finds that generative AI has yielded negligible measurable impact on aggregate U.S. labor productivity and Total Factor Productivity (TFP). Addressing this gap on November 13, 2026, at the UC San Diego Economics Roundtable, Federal Reserve Board Vice Chair Philip N. Jefferson delivered a keynote address titled New Technology, Old Questions: Productivity Puzzles and the Macroeconomy. Jefferson invoked Robert Solow’s famous 1987 paradox that ‘you can see the computer age everywhere but in the productivity statistics.’

Jefferson noted that early personal computer investment in the late 1980s required more than a decade of organizational adaptation before aggregate productivity accelerated. Econometric decomposition in FEDS 2026-074 indicates that recent post-2023 productivity pickups primarily reflect cyclical supply-chain normalization rather than AI diffusion. Furthermore, the authors review growth literature: Daron Acemoglu (2025) estimates cumulative TFP gains of only 0.7% over the next decade, while optimistic projections by Filippucci et al. (2025) reach 0.3 to 0.9 percentage points annually. Task-level studies by Jones and Tonetti (2026) and Demirer, Musolff, and Yang (2026) confirm that coding agents accelerate syntax generation, but software deployment remains tightly bounded by human review and integration testing.

Federal Reserve Leadership Perspectives: Barr and Waller on Labor and Inequality

Central bank governors expanded on these macroeconomic themes during November 2026 addresses. On November 12 at Yale University, Governor Michael S. Barr presented remarks on American Inequality and the Future of AI, analyzing findings from the Federal Reserve’s 2025 Survey of Consumer Finances. Barr warned that because AI infrastructure requires massive capital reserves, financial gains are heavily concentrated among top software firms and high-wealth equity holders, raising distributional risks if routine entry-level cognitive tasks face wage erosion.

Conversely, Governor Christopher J. Waller emphasized labor resilience in his November 10 speech at the Bank of England Watchers Conference on The Economic Outlook and Some Thoughts on Labor Market Dynamism. Waller noted that aggregate claims, vacancies, and payrolls show no sign of widespread displacement. This stability is corroborated by New York Fed research (Abel et al., 2025) and independent market analysis from TradingView, which report that only ~10% of workers use AI daily, with firms primarily augmenting existing roles rather than reducing headcounts.

Implications for University Research Offices and Compute Budgeting

The findings of FEDS 2026-074 carry immediate operational lessons for university research offices, campus HPC directors, and sponsored programs teams navigating hardware scarcity and escalating compute costs:

First, memory chip inflation directly impacts sponsored project proposals. With server component prices rising nearly 8% annually, campus computing centers must replace historical hardware estimates with real-time vendor quotes. Under Uniform Guidance (2 CFR 200), research administrators must carefully distinguish between capital equipment expenditures and operational cloud charges, as detailed in CASRAI’s comparative analysis of direct cost vs indirect cost accounting.

Second, as commercial cloud providers lift API rates to recoup data-center buildouts, academic researchers risk being priced out of foundation model fine-tuning. Campus research offices must actively direct faculty toward subsidized national initiatives, particularly the NAIRR pilot, to secure allocated GPU time for public-interest research and algorithmic auditing.

Third, institutional compliance teams must address municipal power restrictions, sustainability goals, and evolving federal governance frameworks, cross-referencing state and federal mandates mapped in CASRAI’s guide to AI regulation in the US. Finally, the working paper offers academic economists an empirical template for combining Census Bureau survey supplements with BEA gross output data to evaluate technology diffusion objectively.

What Happens Next: National Accounts Revisions and Regulatory Milestones

Federal agencies and central bank policymakers have outlined key milestones to monitor AI’s macroeconomic footprint over the next 12 to 18 months:

  • BEA NIPA Revisions: Through mid-2027, the Bureau of Economic Analysis will update national accounts accounting for ‘own-account’ software and cloud operators’ in-house server assembly.
  • Utility Interconnection Hearings: Regional transmission operators and state public utility commissions will conduct evidentiary hearings in early 2027 on grid reliability and cost allocations for large-scale data centers.
  • FOMC Inflation and Productivity Monitoring: The Federal Open Market Committee will assess whether hardware component inflation influences non-housing core services prices or whether downstream organizational efficiency gains deliver disinflationary benefits.
  • Federal Compute Allocation Rounds: The National Science Foundation will launch expanded resource allocation cycles for the NAIRR initiative in early 2027, expanding subsidized compute access for academic researchers.

Sources

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →