The paper builds a new measure of AI use from a large proprietary OpenRouter dataset, then links it to stock returns, countries, occupations, and skills. The measure could become influential in work on the economic effects of AI. The main questions are whether OpenRouter users are representative, whether the factor construction is stable, and how much can be inferred from market covariance about firms or workers.
These are provisional prioritization judgments. Each explanation links the score to paper-specific evidence and, where relevant, the public-use search.
The new AI-use factor could affect how labor-market agencies, financial institutions, and AI-policy groups measure exposure across firms, countries, and occupations.
The score reflects the need to test a prominent new measure built from proprietary data. Public attention is high, but the search found no independent replication or detailed methodological critique.
This is a July 2026 working paper. Early evaluation could affect revisions before its measure becomes a standard citation in AI-economics and policy discussions.
The data are unusually large and granular, but the design links AI consumption to stock-return covariance. Evaluation should test factor construction, sample representativeness, omitted technology shocks, and the interpretation of market exposure as economic exposure.
NBER distribution, established authors, and rapid coverage in major media and policy-facing economics outlets support a high prominence score.
The paper is already reaching economics and general audiences, so the measure may travel quickly. The score is tempered because no regulator, investor, firm, or policy institution was found using it in a decision framework.
Using 380 trillion tokens of realized AI consumption across more than four hundred large language models from the licensed proprietary OpenRouter dataset covering approximately 2 percent of current global monthly AI token consumption, we analyze how AI affects firms, markets, and workers. Leveraging the unprecedented size, scope and granularity of this data, we construct the AI Factor from growth in tokens, dollars, and users, estimate firm-level AI Betas from stock return comovement, and characterize the AI Premium. First, we build a high-frequency AI factor and decompose it into salient components. Second, we show that firms whose returns covary more positively with the AI factor—high AI beta firms—earn higher subsequent returns, and the AI premium is large and heterogeneous. A value-weighted longshort strategy earns 64.1 basis points per week, and the premium is large for loadings on the intensive, frontier-oriented margin of AI consumption—closed-source models, paying and seasoned users, and long prompts—but not on casual or open-weight use. Third, the premium reaches beyond technology firms into consumer-facing and capital-heavy parts of the economy, but is absent in emerging markets, including China. Fourth, the AI exposure is more positive in nonroutine interactive work and more negative in analytical, scientific, and operations-control skills—an occupation one standard deviation higher in interaction-and-communication content has 0.36-standard-deviation higher market-implied AI exposure. Additionally, we provide early evidence of the rise of the agentic economy.
The main challenge is data access: the core OpenRouter data are proprietary, licensed, aggregated, and not redistributed, which may limit replication unless evaluators can access code, derived data, or secure summaries. The asset-pricing interpretation is also hard to validate because high AI betas may reflect correlated investor beliefs, hype, sectoral shocks, or omitted technology-risk factors rather than causal effects of AI adoption. Evaluation should focus on robustness of factor construction, representativeness of OpenRouter's roughly 2 percent usage sample, sensitivity to market controls and event windows, and whether occupational exposure mappings are appropriate for welfare and labor-policy inference.
This is a high-value Unjournal candidate: it is an NBER working paper using unusually granular proprietary data on realized AI consumption to estimate how AI demand is priced across firms, sectors, countries, and occupations. The findings could inform decisions by AI governance teams, labor-market agencies, financial regulators, industrial-policy groups, and major funders trying to understand which workers, sectors, and countries are gaining or losing from frontier AI diffusion. It is very recent and apparently not externally peer reviewed; because the claims are likely to be cited in debates about AI exposure, labor-market transition, and the distribution of AI gains, independent evaluation would add clear value. Main concerns are that the evidence is based on a proprietary OpenRouter sample that may overrepresent developers and sophisticated users, and the asset-pricing design identifies market-implied exposure rather than causal welfare effects.
The paper informs live decisions about AI labor-market adaptation, AI industrial policy, competition and market-concentration monitoring, financial-risk exposure to AI shocks, and prioritization of retraining or adjustment support across occupations. It is especially relevant for organizations deciding how to measure AI diffusion and economic exposure using realized usage rather than surveys or capability benchmarks.
This is a July 2026 NBER working paper, also recently posted on arXiv/SSRN, so feedback is early enough to affect revisions and downstream interpretation before the paper becomes a standard citation in AI-economics and policy discussions.
Economic growth under transformative AI (Trammell & Korinek) · 48% match
Firm-level AI consumption evidence could update whether AI substitutes for workers enough to affect labor demand and labor share.
The paper received unusually wide attention soon after release, including major media coverage, a CEPR/VoxEU column, university coverage, and social-media discussion. We did not find evidence that firms, regulators, investors, or policy institutions are using the measure in decisions.
The author’s research page links dedicated coverage in The New York Times and NPR Planet Money. The coverage is independent and appeared soon after release. It does not assess the methods in depth or show policy use.
The authors published a detailed CEPR/VoxEU column on the AI factor, the premium, country differences, and occupation mapping. This helps the paper reach policy and economics audiences. It is the authors’ account of their own work.
Yale News ran a detailed feature. It usefully notes that the exposure measure reflects market expectations and does not show that companies already benefit from AI. Yale is a coauthor’s institution.
Public posts discuss model-release event returns and the use of realized OpenRouter consumption instead of surveys. We also found several reposts and summaries. The paper is circulating, but we found no sign that it has changed investment or public policy.
Targeted search checked 2026-09-01. Search scope: Targeted exact-title searches across the open web, institutional and author pages, news and public social-media results, EA Forum/LessWrong, and policy/white-paper contexts. Evidence records distinguish commissioning or report use from independent discussion, media attention, indexing, and post-publication policy use. A search miss is reported as uncertainty, not proof of absence.
The privacy-safe aggregate contains 3 current ratings: 0 team and 3 public. The human mean is 76.7/100. The team has not made a final prioritization decision.
No written discussion is public. Private and team-only text is never copied to this page.