AI, Work and the Paths to Wealth
How technology could reshape employment, ownership and household prosperity.
AI could create substantial new wealth. The question this paper examines is who accumulates it, and through which channels it reaches households.
A richer economy does not tell us who becomes wealthier
AI could create substantial new wealth by raising productivity and enabling valuable businesses. The question is who will accumulate it: the top 1%, the next 9%, or the remaining 90%? The answer depends on how gains and losses reach households through earnings, asset ownership and the costs of living.
We began investigating that question before Anthropic's October 26 S-1 filing. Its estimate of more than $40 trillion in annual global knowledge-work value indicates the scale of activity AI could transform, not the amount of new wealth it will create.1 An indicative US allocation is about $8 trillion.2 New wealth can emerge through productivity gains, business formation, saving and asset appreciation.
Dot-com expansion
Real median income: +9.8% (1995-2000)
Financial crisis
Real median income: -4.0% (2007-2009)
Pandemic-era asset boom
Real median income: -2.4% (2019-2021)
Sources: Federal Reserve DFA; BLS CPI and unemployment; Census income. Different windows; changes are not annualized. Income is annual, wealth is quarterly.
The historical contrasts are instructive. During the dot-com expansion, the remaining 90% gained real wealth but lost share. During the financial crisis, their share rose while their wealth fell. In the pandemic-era asset boom, both increased. None of those outcomes can be inferred from the national total alone.3
Income tells another part of the story. Real median household money income fell about 2.4% from 2019 to 2021 even as real aggregate wealth increased in the corresponding wealth window.4 Asset ownership, earnings and policy responses can pull households in different directions.
Our working view: early gains may favor existing owners; wider participation depends on earnings and access to ownership.
AI changes industries at different speeds
A technology changes an industry only when firms can deploy it reliably, economically and within the responsibilities of the business. Digital work can be reorganized without replacing a factory. Physical automation must also clear the cost, safety and integration hurdles of robotics. Historical diffusion and productivity research explains why organizational investment matters as much as access to a tool.5
The evidence describes different stages. Census found AI in at least one business function at 18% of US firms in late 2025 and early 2026.6 McKinsey found widespread use but enterprise earnings impact at only 37% of surveyed organizations. BCG classified roughly half as scaling or more advanced; MIT researchers found deep integration disclosed by 11% of S&P 500 firms. These are different measures, not one success funnel.7
| Industry setting | Work likely to change first | What slows wider change | Wealth channel to examine |
|---|---|---|---|
| Digital services | Coding, analysis and customer support | Data integration; quality; demand for additional output | Wages, hiring, scalable businesses and employee equity |
| Finance and healthcare | Documentation and administration before high-stakes decisions | Accountability, validation and legacy systems | Compensation, professional careers and institutional profits |
| Manufacturing and logistics | Planning and selected physical tasks | Robotics economics; equipment; capital-replacement cycles | Skilled work, industrial ownership and regional housing |
| Public and private education | Preparation, assessment and administration, K-12 through PhD | Labor agreements, funding, responsibility and institutional choices | Income continuity, pension claims and ownership of research |
| Retail and hospitality | Digital service, scheduling and routine operations | Physical tasks; customer preferences; thin margins | Hours, earnings and the household saving margin |
Jacobs & Company analytical synthesis. The map identifies mechanisms and constraints; it does not assign measured adoption rates or job-loss probabilities.
Once deployment works, a firm can produce the same output with less labor, produce more with its existing workforce, or develop a new business. Those responses can coexist. Whether lower costs expand demand enough to sustain employment is a question to examine, not an assumption to impose.8
Our working expectation is earlier adjustment in digital tasks and hiring, with more uneven physical transformation. The relevant comparison is what adopting firms do with additional capacity, not simply whether they use AI.
Exposure is a technical property. The economic outcome depends on deployment, demand and business choices.
Industry shifts reshape incomes, homes and household wealth
Productivity can rise before pay or employment responds. Customer-support research found a 15% productivity gain; early Danish evidence found no detectable average earnings or recorded-hours effect.9 As industries change, the consequences reach household balance sheets at different speeds.
Pittsburgh lost more than 120,000 manufacturing jobs in the 1980s. Regional employment regained its 1979 peak by 1990, yet the city's 1995 housing plan reported that declining incomes left almost a third of households unable to afford their housing. New jobs did not necessarily restore the position of displaced families.10
Historical examples and analytical channels, not causal AI estimates. Different periods and measures; the wage premium also reflects occupational composition.
Silicon Valley illustrates expansion through new firms, specialized work and ownership. In 2025, San Jose's mean hourly pay was 71% above the national average. A boom can support saving and equity wealth, but rising housing costs can limit the gains for renters and new buyers. Research links local productivity to wages, housing costs and purchasing power, not equal benefits for every resident.11
Neither history is an AI forecast. Together they show why employment, housing, retirement saving and business ownership must be considered jointly. Employee equity, entrepreneurship, new occupations, bargaining and licensing income can broaden participation. Layoffs, lower-paid work and additional debt can erode it. The balance will differ by industry, location, career stage and ownership.12
The same wealth total can describe very different economies
The model combines Federal Reserve history with assumptions about industry adoption, labor income, saving and asset returns. Starting at $172.5 trillion in 2025, its existing scenarios reach $176-214 trillion in 2028 and $187-246 trillion in 2030; 2035 extends the illustration to $244-340 trillion. These remain conditional paths, not validated forecasts.13
Nominal $ trillions. Historical series to 2025; legacy reference path thereafter. Whiskers: scenario span.
Returns accrue mainly to existing owners; wage and ownership gains lag.
Each wealth group retains its 2025 share.
Earnings, new businesses and wider ownership spread the gains.
same in every case
Left: historical data and unchanged legacy scenarios. Right: new, deliberately wide allocation stress tests, not calibrated probabilities. Census-based household estimates; see notes 13-14.
The allocation test is deliberately more demanding. All three cases use $229.4 trillion in 2030. Concentrating gains raises the top 1% share to 45%; wider participation raises the remaining 90% share to 45%. Unchanged shares preserves the 2025 distribution. These are explicit stress assumptions, not estimates of how likely either outcome is.14
The contrast is material. National wealth rises 33% from 2025, yet under concentration the remaining 90%'s nominal wealth falls 16%, to $45.9 trillion. Under wider participation they hold $103.2 trillion; all three groups gain nominal wealth. Inflation and debt still determine real financial strength.
The population assumption yields 139.1 million households in 2030. Each case therefore contains the same 1.39 million top-1%, 12.52 million next-9% and 125.18 million remaining-90% households. Wealth per group changes; percentile size does not. These counts do not estimate how many households cross an absolute asset threshold.
The same growing economy can leave households with very different financial futures.
Use 2028 and 2030 to test the assumptions behind strategy
Leaders need to know which assumptions about work, ownership and customer prosperity their decisions require, and what would challenge them. Our contribution is to make the AI-wealth mechanisms usable for corporate decisions, not to claim the question is new.15
By 2028, we expect hiring and work allocation to provide earlier evidence than accumulated wealth. By 2030, the test is whether these changes create durable earning and ownership opportunities or persistent financial strain. Measure every channel throughout; the dates change the emphasis, not the economic sequence.
| Economic question | 2028 | Early adjustment | 2030 | Durable effects | Corporate decision |
|---|---|---|---|
| Capacity or substitution? | Hiring and task allocation diverge as workflows change. | Does expanding output create enough new work to offset displacement? | Redeploy, develop or replace? |
| Who participates? | Watch real pay, employee equity and viable new firms. | Do those channels generate durable income and ownership? | Which customer base supports growth? |
| Can gains endure? | Track saving, mortgage stress and retirement contributions. | Is adjustment strengthening or depleting household balance sheets? | Which demand assumptions still hold? |
Jacobs & Company working expectations, conditional on continued diffusion. The 2035 horizon illustrates cumulative outcomes; it is not a point forecast for capital budgeting.
Three choices follow. Invest in opportunities that do not require one favorable distribution of prosperity. Decide where automation should release capacity and which human capabilities must still be developed. Test growth plans against the industries, careers and ownership structures that create customer wealth.
AI can expand national wealth without preserving the paths through which households build it. Neither concentration nor broader participation is predetermined. Current leaders and those who will lead in 2035 need a shared account of these mechanisms, even when their investment horizons differ.
For asset and wealth managers, this raises the next strategic question: if wealth emerges in different places and through different pathways, who will need advice, who will control allocation, and how must distribution adapt? A subsequent paper will examine those choices.
Sources, calculation methods and limitations
- Anthropic, PBC, Form S-1, filed October 26, 2026, "A Near-Term Opportunity to Transform Cognitive Work." The company estimates more than $40 trillion of annual global knowledge-work value in 2025. This is a company estimate of existing labor-related economic activity, not AI revenue or additional wealth. The inquiry in this publication predates the filing. Source: Anthropic filing
- Jacobs & Company indicative US allocation, not an occupational estimate: $40T x [US wages and salaries of approximately $13.0T / (global GDP of approximately $118.4T x 52.6% global labor-income share)] is approximately $8.3T, rounded to $8T. US wages and global labor income differ in coverage; neither is a matched measure of knowledge-worker earnings. The estimate assumes a similar knowledge-work share and is not used to calibrate our household-wealth scenarios. Sources: BEA, September 2026 annual update; World Bank, GDP (current US$); ILO, 2025 labor-income share. Sources: BEA annual update | World Bank GDP | ILO labor-income share
- Federal Reserve Board, Distributional Financial Accounts, accessed November 2, 2026. Top 1%, next 9%, next 40% and bottom 50% levels: FRED WFRBLT01026, WFRBLN09053, WFRBLN40080 and WFRBLB50107. Remaining 90% combines the last two groups. Stocks are quarter-end; displayed real changes use CPI-U, seasonally adjusted, in each quarter's last month. Windows: 1995 Q4-2000 Q1; 2007 Q3-2009 Q1; 2019 Q4-2021 Q4. They differ in duration and are not annualized. Percentile groups are not fixed household panels. Sources: Federal Reserve DFA | BLS consumer price index
- US Census Bureau, real median household money income, FRED MEHOINUSA672N (2025 dollars); BLS unemployment rate, FRED UNRATE. Income comparisons use calendar years 1995-2000, 2007-2009 and 2019-2021, not the wealth quarters; Census applies its own inflation methodology. Unemployment is the final month of the corresponding wealth quarter. Money income excludes capital gains and is not comprehensive disposable income. The historical examples are descriptive, not estimates of technology's causal effect. Sources: Census household income | BLS unemployment rate
- Diego Comin and Bart Hobijn, "An Exploration of Technology Diffusion," American Economic Review 100(5), 2010, 2031-2059. Erik Brynjolfsson, Daniel Rock and Chad Syverson, "The Productivity J-Curve: How Intangibles Complement General Purpose Technologies," American Economic Journal: Macroeconomics 13(1), 2021, 333-372. These peer-reviewed foundations support heterogeneous diffusion and complementary investment, not a fixed adoption timetable for AI. Sources: Technology diffusion | Productivity J-curve
- Kathryn Bonney and coauthors, "The Microstructure of AI Diffusion," US Census Bureau, CES Working Paper 26-25, April 2026. BTOS evidence for November 2025-January 2026: 18% of firms used AI in a business function; 32% employment-weighted. Firm use is distinct from individual worker use, deep integration and a causal employment effect. Source: Census working paper
- McKinsey & Company, "The State of AI in 2026: On the Road to ROI," August 25, 2026: nearly nine in ten respondents reported use, 37% enterprise EBIT impact and 6% high performance. BCG, "The Formula for Agentic AI Value," September 30, 2026: 7.5% future-built and 41% scaling with some value. MIT IDE/FutureTech, "AI Adoption in S&P 500 Firms," July 2026: 11% classified as deeply integrated by end-2025 using SEC filings. These different samples, dates and definitions cannot be joined into a conversion funnel or representative failure rate. Sources: McKinsey study | BCG study | MIT adoption research
- Daron Acemoglu and Pascual Restrepo, "Automation and New Tasks: How Technology Displaces and Reinstates Labor," Journal of Economic Perspectives 33(2), 2019, 3-30. The task framework distinguishes displacement, productivity and the reinstatement of labor through new tasks. Our industry map is an analytical application, not an estimated industry-specific effect. Source: Automation and new tasks
- Erik Brynjolfsson, Danielle Li and Lindsey Raymond, "Generative AI at Work," Quarterly Journal of Economics 140(2), 2025, 889-942: 5,172 customer-support agents; average productivity increase of 15%. Anders Humlum and Emilie Vestergaard, "Still Waters, Rapid Currents," revised March 2026: linked Danish survey and administrative records; no detectable average earnings or recorded-hours effect over the early observation period. Their designs, populations and outcomes differ. Neither identifies a 2035 US labor coefficient. Sources: Generative AI at work | Danish labor-market study
- City of Pittsburgh, 1995 Consolidated Plan (HUD archive): over 120,000 manufacturing jobs lost regionally in the 1980s; the four-county labor market regained its 1979 job total by 1990, while lower incomes left almost one-third of city households unable to afford their housing. Guhan Venkatu, Cleveland Fed, Rust and Renewal (2018), documents both severe adjustment and subsequent renewal. Jacobson, LaLonde and Sullivan, AER 83(4), 1993, establish persistent earnings losses for displaced workers in a historical sample. These examples do not estimate AI effects. Sources: Pittsburgh housing plan | Cleveland Fed retrospective | Displaced-worker research
- BLS, San Jose-Sunnyvale-Santa Clara Occupational Employment and Wages, May 2025: mean hourly pay $57.32 versus $33.54 nationally, a 70.9% premium. This is descriptive and partly reflects occupational composition, not a causal technology estimate. Hornbeck and Moretti, Review of Economics and Statistics 106(3), 2024, 587-607, examine local and distant effects of productivity on wages, rents and purchasing power; their evidence is not specific to Silicon Valley. Sources: BLS San Jose wage data | Local productivity research
- Benjamin Moll, Lukasz Rachel and Pascual Restrepo, "Uneven Growth: Automation's Impact on Income and Wealth Inequality," Econometrica 90(6), 2022, 2645-2683; Emma Rockall, Marina Mendes Tavares and Carlo Pizzinelli, "AI Adoption and Inequality," IMF Working Paper 2025/068. Both connect technology to income and wealth distributions. The distinction between labor's share and the distribution of capital ownership is central to our analysis, not a claim of priority. Sources: Uneven Growth | IMF AI adoption study
- Jacobs & Company scenario model, November 2026 publication extract. The 2025 Q4 DFA baseline totals $172.466T; the top 1%, next 9% and remaining 90% hold 31.77%, 36.49% and 31.75%. Four legacy annual scenarios imply nominal totals of $175.7-214.4T in 2028, $187.5-245.8T in 2030 and $244.1-339.9T in 2035. Historical data and scenario calculations are supplied separately. The ranges are not confidence intervals. Earlier rolling-origin tests benchmarked historical extrapolation, not the integrated AI model. No new causal calibration is claimed. The later 2026 Q2 total is $185.651T; the legacy year-end scenarios have not been rebased to it. Source: Federal Reserve DFA | accompanying data and calculation notes
- New matched-total stress test, not a forecast recalibration. Concentrating 2030 shares: 45/35/20%; wider participation: 25/30/45%; unchanged: exact 2025 shares. Shares interpolate to 2030 and are then held to 2035. At $229.373T in 2030, remaining-90% wealth is $45.875T, $72.816T or $103.218T. Counts use Census CPS 2025 households (134.790m) plus 0.860m annually, rebasing JCHS's projected 8.6m decade increment; this is not JCHS's level forecast. The 2030 total is 139.090m; 1/9/90% imply 1.3909/12.5181/125.181m in every case. CPS households and DFA/SCF family units differ. Counts are illustrative, not fixed-asset-threshold or household-mobility forecasts; group means are not medians or investable assets. Sources: Census households | Harvard household projections | accompanying calculations
- Related work includes Allianz Global Wealth Report 2026, especially "Who will capture the AI wealth dividend?" (p. 17), and the IMF and Econometrica studies in note 12. This paper does not claim to originate the AI-wealth question. Its contribution is an executive synthesis of industry adjustment, household balance sheets and matched-distribution scenarios. The 2028/2030 expectations are conditional analytical judgments, not causally estimated probabilities. Source: Allianz Global Wealth Report
Research, responsibility and permissions
Copyright and use
Copyright 2026 Jacobs & Company. All rights reserved. The original text, frameworks and exhibits in this publication are the intellectual property of Jacobs & Company. Third-party research and data remain the property of their respective owners and are cited with attribution.
Brief quotations for commentary, review or scholarship may be used with full attribution. Permission is required to reproduce or adapt substantial portions of this publication. For permissions and research inquiries, contact info@jacobsandcompany.biz.
Use of AI
Jacobs & Company set the research questions, shaped the analysis and modeling, and directed the drafting and editorial work. AI tools assisted with research support, analysis and drafting. Jacobs & Company is responsible for verifying sources, evaluating the analysis and approving the publication.
Scope and interpretation
This publication provides economic analysis for discussion. It is not investment, legal, tax or other professional advice. Historical observations, interpretations, hypothetical mechanisms and conditional model outcomes are distinguished in the exhibits and endnotes. Scenarios are not promises of future performance or statistically calibrated confidence intervals.
The analysis centers on the United States. International and firm-level studies illuminate mechanisms but do not establish economy-wide US effects. The existing model is a decision-support framework: its computational consistency does not establish causal identification or predictive superiority. Important assumptions remain uncertain.
Suggested citation
Jacobs & Company, AI, Work and the Paths to Wealth: How Technology Could Reshape Employment, Ownership and Household Prosperity, November 2026.
About Jacobs & Company
Jacobs & Company helps growth-focused asset and wealth managers turn go-to-market strategy into a working system. Our work connects distribution strategy, segmentation, data, CRM, analytics and AI with the decisions people make every day.
The value of a forecast is not only its endpoint. It is the assumptions it makes visible and the decisions it helps leaders examine.
jacobsandcompany.biz | info@jacobsandcompany.biz