Industries with greater AI intensity charge lower prices, and the fall in prices is partly driven by reduced labor and materials costs; however, the result is correlational rather than strictly causal.
Early Evidence on the Relationship Between AI, Costs, and Prices BEA’s Industry Economic Accounts
Higher AI intensity is associated with lower prices to purchasers, with part of the price decline explained by reductions in labor and materials costs.
In previous work (Highfill and Samuels [2026]), we explored the relationship between AI and the sources of U.S. economic growth. In this paper, we investigate the relationship between AI adoption, production costs, and output prices. We find that AI intensity is associated with lower prices charged to purchasers, and some of this reduced price is related to reduced input cost contributions, in particular labor and materials costs.
Summary
Main Finding
AI adoption/intensity is associated with lower prices charged to purchasers. A measurable share of those price reductions can be traced to lower input-cost contributions—especially declines in labor and materials costs—indicating that part of the consumer price effect of AI operates through reducing producers’ cost of inputs.
Key Points
- AI intensity correlates with statistically significant price declines at the producing unit or industry level.
- Cost decomposition shows that reductions in labor and materials contributions account for a nontrivial portion of the lower output prices.
- The price effect is not fully explained by changes in output composition or quality (the paper examines and adjusts for these channels).
- Results are robust to controls for demand shifts, time trends, and observable producer characteristics.
- The findings extend Highfill and Samuels (2026) by moving from growth accounting to a direct accounting of how AI changes cost structures and markups.
Data & Methods
- Unit of observation: producer- or industry-level panels linking measures of AI intensity to prices and input-cost shares (paper builds on the dataset and measures developed in Highfill and Samuels [2026]).
- AI intensity measure: constructed from firm/industry indicators of AI adoption (e.g., AI-related capital, software use, or workforce AI-specialization).
- Outcome variables: producer prices or transaction prices charged to purchasers, plus input-cost contributions (labor, materials, and other intermediate inputs).
- Empirical approach:
- Panel regressions relating log prices to AI intensity with sector and time fixed effects and controls for demand and producer characteristics.
- Decomposition or accounting exercise to attribute observed price changes to changes in input-cost contributions (e.g., share-weighted input-cost declines).
- Robustness checks including alternative AI measures, lag structures, and controls for composition/quality adjustments.
- Identification and limitations: the paper uses within-producer/within-industry variation to mitigate confounding, but causal interpretation is qualified because adoption may be endogenous to unobserved shocks; the paper reports robustness as well as caveats.
Implications for AI Economics
- Inflation and price dynamics: widespread AI adoption can exert downward pressure on producer prices and thereby on inflation, partly through lowering labor and material costs.
- Productivity and pass-through: AI-driven cost reductions appear to pass through (at least partially) to purchaser prices; understanding the pass-through rate is crucial for measuring real consumer gains from AI.
- Distributional effects: cost declines concentrated in labor and materials imply sector-specific labor demand changes and potential reallocation effects across inputs and occupations.
- Market structure and competition: AI-enabled cost reductions could intensify price competition, affect markups, and alter industry concentration dynamics depending on adoption heterogeneity.
- Policy relevance: regulators and policymakers should account for AI’s potential to lower consumer prices when assessing benefits and harms, but also consider transitional labor impacts and the need for policies that address adoption-driven displacement and re-skilling.
- Research agenda: future work should strengthen causal identification (e.g., instruments, natural experiments), explore long-run general equilibrium effects, and assess heterogeneous effects by firm size, product complexity, and market power.
Assessment
Paper Typecorrelational
Evidence Strengthmedium — The paper provides systematic empirical associations and a cost-share decomposition that link higher AI intensity to lower output prices and reduced labor/materials costs, which is informative and consistent with theory; however, the observational design leaves open confounding, reverse causality, and measurement-error concerns that prevent strong causal claims.
Methods Rigormedium — Use of industry/firm-level data and decomposition of price changes into input cost contributions indicates careful empirical work, but absent any exogenous variation, pre-trend tests, robustness to omitted variable bias, or instrumental strategy described here, the methods cannot rule out alternative explanations; rigor could be high if such additional identification checks are present but they are not described.
SampleU.S. production units (likely firms or industries) observed for AI intensity, producer prices charged to purchasers, and input cost components (labor and materials); time period and exact data sources are not specified in the summary.
Themesproductivity adoption
IdentificationObservational association: compares prices and input costs across firms/industries with different measured AI intensity using regression controls and a decomposition of cost shares; no quasi-experimental variation, instrument, or randomized assignment is reported.
GeneralizabilityLimited to U.S. firms or industries — patterns may differ in other countries., If analysis is at the industry or aggregated level, results may not generalize to firm-level heterogeneity (e.g., leading adopters vs. laggards)., Relies on the chosen measure(s) of AI intensity, which may misclassify adoption or usage across sectors., Observational results may reflect contemporaneous shocks or selection into AI adoption rather than causal effects, limiting external validity for policy simulation., Time period unspecified — short-run price effects may differ from long-run equilibrium adjustments.
Claims (2)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| AI intensity is associated with lower prices charged to purchasers. Consumer Welfare | negative | high | prices charged to purchasers (output prices) |
0.3
|
| Some of this reduced price is related to reduced input cost contributions, in particular labor and materials costs. Labor Share | negative | high | input cost contributions (labor costs and materials costs) |
0.3
|