Skip to map

The map couldn't load right now.

United States

Total workers --
Wage bill --

Region

AI-exposed workforce
--
AI-exposed wage bill
--
/
/

Top 5 industries · Size: # of Workers · Color: AI Exposure

Select a geography to view its top industries.

Home

Settings

Select a region

Headcount AI Exposure

--

Exposed headcount: --

Total headcount: --

Wage bill AI Exposure

--

Exposed wagebill: --

Total wagebill: --

/
/

5 largest industries | Size: # of Workers | Color: AI Exposure

Select a geography to view the top industries by employment.

Methodology & data

AI Labor Exposure Map Quantifies $1.4 Trillion in U.S. Wages Substitution Potential

Interactive tool reveals which regions and jobs face the greatest exposure to artificial intelligence adoption.

  • Current scenario · Anthropic
    $1.4T/yr
    ≈ 18M FTE workers in wage-bill equivalent
  • Theoretical scenario · OpenAI
    $2.9T/yr
    ≈ 36M FTE workers exposed at the 50% threshold

Cambridge, MA — The MIT Center for Transportation and Logistics (CTL) has launched an interactive mapping tool that quantifies exposure to artificial intelligence across U.S. regions, industries, and job types. The new AI Labor Exposure Map, developed by PhD student Pierre Bouquet and visiting student Luca Mouchel, shows that under a full-adoption, substitutive-use scenario based on current AI capabilities, AI could currently perform work equivalent to approximately $1.4 trillion per year in U.S. wage-bill equivalent.

Data and Methodology

The analysis integrates three key data sources: employment and wage data sourced from the Bureau of Labor Statistics May 2025 Occupational Employment and Wage Statistics (OEWS) at state and metropolitan statistical area levels; job-to-task mapping constructed using the MIT Work Analytics Lab methodology from “Estimating the Task Content of Work” (Bouquet et al., 2026); and AI capability assessments derived from both the Anthropic AI Economic Index for current capability and the GPTs are GPTs framework (Eloundou et al., 2023) for theoretical capability.

Under the current Anthropic-based scenario, if reported AI task capabilities were fully adopted and substituted at the levels reported by Anthropic, Claude could perform work equivalent to ≈ 18 million FTE workers — about $1.4 trillion per year in wage-bill equivalent. The broader theoretical OpenAI-based scenario, which considers tasks AI could accelerate by 50% or more, rises to ≈ 36 million FTE workers — about $2.9 trillion per year.

These figures are not predictions of layoffs, impacted workers, or unemployment. They are scenario-based estimates of labor input that would be exposed under full-adoption and substitutive-use assumptions.

Key Findings

This research cuts through the hype around AI by providing concrete data about where economic exposure actually exists. The key point is that exposure is not evenly distributed geographically nor across the workforce. While overall workforce exposure appears at 13%, these jobs represent 16% of the total wage bill, meaning high-paying white-collar positions are most exposed to AI impact.

Pierre Bouquet

The interactive platform reveals stark regional disparities. Washington, D.C., for example, faces twice the workforce exposure of Wyoming — 20% versus 10%. Even within the same region, variations are significant: Boston's metropolitan area shows substantially higher AI exposure than Bangor, Maine, reflecting starkly different workforce compositions.

The tool distinguishes between current AI capabilities — measured through adoption patterns of Anthropic's Claude AI system — and theoretical capabilities based on OpenAI capability-frontier research. This dual-measurement approach provides both a current-use-based assessment and a broader scenario-based projection.

‘Exposure’ does not necessarily mean job elimination. It means that some tasks within jobs can be accelerated or automated, potentially augmenting rather than replacing workers. A financial manager, for example, might see loan application evaluation become highly automatable, while training, oversight, and judgment-intensive responsibilities remain human-dependent.

Pierre Bouquet

Granular Insights Across Sectors

The map lets users narrow down by state, metropolitan area, industry, and specific job tasks. Service-industry positions such as food preparation and serving remain largely AI-resilient. On the other hand, computer and mathematical fields show among the highest exposure, with programmers particularly exposed to current and theoretical AI capabilities. Some local areas may also show unexpected exposure patterns due to their specific workforce composition.

Applications for Policymakers, Business, and Media

Policymakers can identify communities facing the greatest AI-related impact and plan workforce transitions accordingly. Companies can evaluate the implications of facility placement and workforce strategy. Local news outlets can report on region-specific exposure patterns, while national media can contextualize AI's economic scale using localized, task-level labor-market data.

Someone in Wyoming sees very different exposure than someone in D.C., and that is now visible and actionable. Regions with higher AI exposure may face greater labor-market pressure if displaced workers compete in a smaller pool of available roles. That could affect job competitiveness, entry-level opportunities, and wage dynamics.

Pierre Bouquet

While the map does not forecast those outcomes, it helps identify where such pressures could emerge.

About MIT CTL

The MIT Center for Transportation & Logistics (MIT CTL) is a world leader in supply chain management research and education, with over 50 years of expertise. Its work spans industry partnerships, cutting-edge research, and driving supply chain innovation into practice through three pillars: research, outreach, and education. ctl.mit.edu

Media Contact
Mackenzie Berry
Communications Writer, MIT CTL
berrym@mit.edu
617-258-6083

Sources

Core inputs used to build the US and Canada AI exposure map.

Employment and Wage Data

US

Sourced from the Bureau of Labor Statistics (BLS) May 2025 Occupational Employment and Wage Statistics (OEWS) (released 2026) at the state and MSA levels.

Canada

Statistics Canada 2021 Census of Population employment and average employment income by NOC 2021 (provinces, CMAs, and CAs), brought forward to 2025 with Labour Force Survey growth factors and converted to USD at the 2025 Bank of Canada average rate. 2026 Census occupation tables (expected 2027) will replace the survey-based rescaling once released.

Canadian Occupation Mapping

Canadian NOC 2021 occupations are mapped to the same O*NET-SOC profiles used for the US with The Dais NOC–O*NET crosswalk (The Dais at Toronto Metropolitan University, formerly the Brookfield Institute), so both countries share one occupation-task exposure model. Occupations without a US counterpart (e.g. Canadian Armed Forces, legislators) are excluded to mirror OEWS coverage.

Job-to-Task Mapping

Constructed by the MIT Work Analytics Lab (WAL) using the methodology developed in “Estimating the task content of work” (Bouquet et al., 2026) to estimate occupation-task time allocation.

Adoption

What AI is doing today. Derived from the Anthropic AI Economic Index, where penetration scores represent loosely the share of a task currently capable of being automated or significantly assisted by Claude (Anthropic, 2026).

Current Capabilities

What AI could do today. Evidence-grounded estimate of present-day AI ability: multiple open-weight models judge each occupation–task pair using retrieved news and research abstracts, then the scores are calibrated to observed adoption. Designed to sit between pure theoretical capability and single-platform usage measures (Mouchel, Bouquet & Sheffi, 2026).

Theoretical

What AI might eventually do. Based on the GPTs are GPTs framework (Eloundou et al., 2023). Exposure is defined as a potential 50% reduction in task completion time while maintaining output quality. Tasks classified as E1 or E2 are treated as theoretically accelerable under this 50% threshold.

Contact

Reach out about the research or for media inquiries.

Research Contact
Pierre Bouquet
PhD student, MIT CTL
Research Lead, MIT Work Analytics Lab
pibou149@mit.edu
Media Contact
Mackenzie Berry
Communications Writer, MIT CTL
berrym@mit.edu
617-258-6083
MIT Center for Transportation & Logistics MIT Work Analytics Lab