AI Ownership, Capital Concentration, and the Redistribution of Economic Power: A Macroeconomic and Structural Analysis

Full report

On this page

The acceleration of artificial intelligence and advanced robotics represents a profound structural shift in the organizational mechanics of global production. As intelligent systems transition from routine task automation to cognitive labor, complex reasoning, and physical manipulation, the central macroeconomic question moves beyond aggregate wealth generation to wealth distribution. If capital assets—embodied in software algorithms, specialized silicon, data center infrastructure, and automated physical systems—produce an expanding share of macroeconomic output, the distribution of economic power will depend on who owns and controls those assets. This research report investigates whether the widespread adoption of AI and robotics is likely to produce an unprecedented concentration of income, productive assets, corporate control, and economic influence. It examines historical distribution trends, maps structural monopolies across the AI value chain, evaluates macroeconomic transmission channels using task-based economic frameworks, analyzes historical technological shocks, presents opposing economic perspectives, and models alternative income-distribution scenarios through 2040\.

1. Historical Macroeconomic Foundations: Distribution of Income, Wealth, and Factor Shares (1980–Present)

1.1 Distributional Macroeconomic Dynamics

Since 1980, economic inequality across advanced economies—and within the United States in particular—has widened substantially. Data integrated from the Federal Reserve’s Distributional Financial Accounts (DFAs), the Bureau of Economic Analysis (BEA), and the World Inequality Database (WID) show a persistent concentration of total net worth and national income at the upper end of the wealth distribution1. In 1989, the top 1 percent of U.S. households held 22.8 percent of total national net worth3. By the mid-2020s, this figure had expanded to 32.5 percent4. Over the same timeframe, the share of total wealth held by the top 0.1 percent nearly doubled, rising from 8.6 percent to 15.0 percent4. Conversely, the bottom 50 percent of the U.S. population experienced a structural contraction in its net worth share, falling from 3.5 percent in 1989 to 2.3 percent4. According to recent Federal Reserve distributional metrics, the top 50 percent of American households hold 97.7 percent of all household net worth, leaving the lower half of the population with 2.3 percent3. This divergence in net worth is directly tied to asset portfolio composition across wealth percentiles3. Lower- and middle-income households hold the vast majority of their gross assets in primary residences, which carry high mortgage leverage and modest long-term real appreciation rates3. Financial assets—specifically corporate equities, mutual fund shares, and private business equity—are concentrated in the upper decile3. The top 10 percent of households own between 81 and 89 percent of total corporate stock market value, while the top 1 percent alone holds more equity value than the bottom 90 percent combined3. The bottom 50 percent holds 2.2 percent of total national financial assets5.

1.2 Secular Contraction of Labor's Share of Income

For much of the post-WWII era, macroeconomic analysis operated under the empirical baseline formulated by Paul Douglas and Arthur Bowley, which posited that the division of national income between labor ([Figure omitted from source export]) and capital ([Figure omitted from source export]) remained relatively stable over time, oscillating around 65 percent for labor compensation and 35 percent for capital returns. However, since the early 1980s, this structural balance has broken down across advanced industrial economies6. Data from the OECD, BEA, and McKinsey Global Institute reveal that labor's share of national income across advanced economies fell from an average of 54 percent in 1980 to 50.5 percent by 20146. In the United States, the drop was more acute: the labor share of nonfarm business value-added fell from roughly 64 percent in the immediate postwar decades to 56.7 percent by 20166. Three-quarters of this post-1947 contraction occurred after 20006. The decline in the labor share reflects a widening divergence between labor productivity growth and median wage growth. Between 1946 and 1980, per-adult real national income grew by 2.0 percent annually in the United States, with income gains distributed broadly across all income percentiles2. Post-1980 productivity continued to expand, but median real compensation growth lagged behind6. Empirical decomposition indicates that the fall in labor's share of value-added accounts for approximately 18 percent of the gap between median real wage growth and productivity growth, while rising compensation inequality among workers accounts for 19 percent, and sluggish TFP growth accounts for 50 percent6.

1.3 Structural Drivers of Labor Bargaining Power Erosion

The secular erosion of labor's share of income stems from four mutually reinforcing macroeconomic structural shifts: First, the relative price of capital investment goods—particularly information technologies, software automation, and industrial robotics—declined rapidly after 1980\. This relative price change induced firms to substitute cheap automated capital for routine manual and cognitive human labor. Second, globalization and the integration of international labor markets weakened domestic labor's bargaining leverage, particularly in trade-exposed manufacturing sectors8. Third, institutional changes in corporate governance prioritized shareholder value maximization9. The decline of labor union density, the expansion of non-compete agreements, and the eclipse of corporate rent-sharing mechanisms reduced the proportion of firm profits allocated to non-executive compensation9. Fourth, product markets experienced rising concentration, giving rise to high-margin "superstar firms" characterized by low labor-to-revenue ratios. As these lead firms captured expanded market shares, the aggregate labor share across industries contracted.

2. Industrial Organization and Structural Monopoly Across the AI Value Chain

2.1 Monopolistic Bottlenecks in Semiconductor Hardware

Analyzing whether AI will concentrate economic power requires examining the hardware value chain that underpins modern artificial intelligence. The hardware architecture of advanced AI is characterized by technological specialization and extreme capital intensity, leading to structural monopolies and tight oligopolies across vertical production stages10. At the foundation of semiconductor fabrication is photolithography equipment. ASML, a Dutch technology firm, holds an absolute monopoly (100 percent market share) in Extreme Ultraviolet (EUV) photolithography systems10. EUV systems use light at a wavelength of 13.5 nanometers to etch circuit patterns smaller than 5 nanometers onto silicon wafers10. Standard low-NA EUV machines cost approximately \$150 million to \$200 million, while next-generation High-NA EUV machines cost upwards of \$380 million to \$400 million per unit10. Each machine comprises over 100,000 components sourced from a global network of over 2,000 specialized suppliers10. Because High-NA EUV technology is required to print sub-2nm logic nodes, ASML functions as an irreplaceable bottleneck for the semiconductor industry10. In chip fabrication, Taiwan Semiconductor Manufacturing Company (TSMC) maintains a global foundry market share of roughly 62 percent, which rises to over 90 percent for the advanced logic nodes (5nm, 3nm, and sub-2nm) required for frontier AI accelerators10. Constructing a modern leading-edge semiconductor fab requires over \$20 billion in capital expenditure, creating high barriers to market entry14. Competitors such as Intel Foundry Services and Samsung Foundry face technical yield challenges, leaving TSMC as the primary manufacturer for top-tier chip designers11. At the accelerator design layer, Nvidia holds between 80 and 95 percent of the market for data center AI chips10. Nvidia's market position is reinforced by hardware performance (such as Tensor Cores and NVLink high-speed interconnects) integrated with its proprietary Compute Unified Device Architecture (CUDA) software platform10. Developed over two decades, CUDA functions as the default environment for deep learning frameworks (e.g., PyTorch, TensorFlow). Because software libraries and operators are natively optimized for CUDA, migrating software workloads to alternative hardware architectures requires extensive re-engineering, creating strong platform lock-in16.

2.2 Hyperscaler Cloud Platforms and Infrastructure Economics

The operational layer of advanced AI requires physical hosting infrastructure, including high-density data centers, liquid cooling systems, optical interconnects, and gigawatt-scale electrical grid access. This infrastructure is dominated by tech hyperscalers: Amazon (AWS), Alphabet (Google Cloud), Microsoft (Azure), Meta, and Oracle10. Capital expenditure trajectories highlight the widening resource gap between hyperscalers and potential market entrants. Aggregate hyperscaler CapEx dedicated to AI infrastructure is projected to reach \$760 billion annually, with individual annual CapEx budgets scaling to unprecedented levels18:

• Alphabet: $180–$190 billion annually18.
• Amazon / AWS: Upwards of \$150 billion annually18.
• Microsoft & Meta: Exceeding \$100 billion annually each18.

These investments create substantial barriers to entry. Establishing a modern AI data center facility requires between \$18 million and \$54 million per permanent job created, making advanced AI deployment one of the most capital-intensive industries in corporate history18. Furthermore, securing base-load electric power capacity—measured in multi-gigawatt blocks—has emerged as a major physical bottleneck, favoring incumbent firms capable of executing multi-decade power purchase agreements for nuclear, geothermal, and utility-scale solar generation18.

2.3 Foundational Models, Proprietary Datasets, and Physical Robotics

The software layer displays similar structural dynamics. Training frontier foundational models requires non-linear scaling of compute clusters, costing hundreds of millions of dollars in direct compute per training run. While open-weight models provide public baseline capabilities, frontier models depend on proprietary, non-public data pipelines, human-feedback loops (RLHF), and specialized infrastructure optimizations. In physical robotics, concentration is driven by manufacturing scale and supply chain integration. Actuator design, precision gearboxes (e.g., strain wave gearing), tactile sensors, and high-energy-density battery cells require high capital outlays and complex manufacturing facilities. Companies with established manufacturing capabilities, software integration capabilities, and access to capital are positioned to lead the physical automation of warehousing, logistics, manufacturing, and domestic service tasks.

2.4 Corporate Case Studies in Value Capture

Case Study 1: Nvidia and the CUDA Platform Moat

Nvidia’s evolution from a graphics hardware vendor into a \$3 trillion+ enterprise illustrates the power of vertical hardware-software integration16. By bundling its proprietary CUDA software layer with its graphics processing units (GPUs) beginning in 2006, Nvidia created a developer ecosystem that locked academic researchers and enterprise developers into its hardware architecture10. As deep learning workloads expanded post-2012, CUDA became the default standard10. Competitors offering equivalent hardware specifications struggle to gain market share because migrating existing software stacks away from CUDA incurs high switching costs16. This platform lock-in has enabled Nvidia to command gross margins near 75 percent on its data center accelerators, capturing a significant portion of total AI hardware spending16.

Case Study 2: ASML, High-NA EUV, and Extreme Lithography Bottlenecks

ASML’s position in the semiconductor ecosystem demonstrates how extreme technological complexity can generate absolute corporate monopolies10. Developed over three decades in partnership with research institutes and optics manufacturers (such as Carl Zeiss), ASML’s Extreme Ultraviolet lithography technology uses a optical assembly to bounce 13.5nm light off mirrors polished to atomic tolerances10. Because alternative lithography methods (such as direct-write electron beam or nanoimprint lithography) cannot match EUV throughput at advanced nodes, ASML holds a 100 percent market share in EUV systems10. This absolute monopoly allows ASML to generate gross margins exceeding 54 percent while dictating delivery allocations to major chipmakers13.

Case Study 3: TSMC and Advanced Node Manufacturing Hegemony

TSMC’s dedicated pure-play foundry model enabled it to achieve manufacturing scale advantages over integrated device manufacturers (IDMs) like Intel11. By consolidating manufacturing demand from fabless chip designers worldwide (including Nvidia, Apple, AMD, Qualcomm, and Broadcom), TSMC achieved high production volumes that accelerated its learning-curve progression11. This allowed TSMC to optimize yield rates and node transitions faster than competitors11. Today, with over 60 percent overall foundry market share and over 90 percent market share in sub-5nm logic fabrication, TSMC acts as an essential manufacturer for advanced computing silicon, creating significant geopolitical and supply chain concentration10.

Case Study 4: Hyperscaler-Startup Strategic Alliances

The structural relationship between AI startups and cloud hyperscalers is illustrated by multi-billion-dollar partnership arrangements, such as Microsoft’s alliance with OpenAI and Amazon’s investments in Anthropic. Rather than operating as independent disruptors, leading AI startups rely on hyperscalers for capital, computing capacity, and distribution channels18. In these strategic alliances, hyperscalers provide compute credits funded by corporate equity stakes18. This allows hyperscalers to convert venture capital investments back into cloud infrastructure revenue while securing long-term customer lock-in for their compute platforms18. These arrangements allow incumbent tech platforms to maintain control over frontier AI research without facing direct market disruption.

3. Scale Dynamics, Intellectual Property, and Platform Market Competition

3.1 Economic Moats and Scale Feedback Loops

The market structure of the AI ecosystem is shaped by positive feedback loops that reinforce incumbent platform advantages: First, the Data Flywheel Effect creates a self-reinforcing advantage. Deploying AI models at scale generates continuous streams of real-world user interaction data. This data is fed back into model fine-tuning and safety alignment, which improves model performance and attracts additional users, further expanding data access. Second, the Fixed-Cost Amortization of Compute creates scale economies. The marginal cost of serving an AI inference query is low relative to the high fixed cost of training frontier models (\$100M+ per training run). Large-scale operators spread these fixed R\&D and training expenditures across billions of queries, lowering unit costs below levels achievable by smaller competitors. Third, Intellectual Property Aggregation protects incumbent market share. Incumbent platform companies accumulate broad patent portfolios spanning software architecture, neural network optimizations, custom silicon, and hardware interconnects, creating legal and financial exposure for new market entrants.

3.2 Market Entry Dynamics: Incumbent Consolidation vs. Disruptive Entrants

A central debate in industrial organization is whether AI will follow a disruptive entrant model (where agile startups displace legacy market leaders) or an incumbent consolidation model (where established platforms capture structural gains). Current market trajectories favor the incumbent consolidation model. Unlike the early commercial Internet era, which was characterized by low initial capital costs, open network protocols, and distributed server infrastructure, the modern AI paradigm requires capital-intensive physical infrastructure18. Startups that develop innovative model architectures routinely depend on hyperscalers for compute resources and distribution. This dependency frequently leads to strategic acquisition, minority equity investments, or talent acqui-hires by incumbent platforms, preserving established corporate market positions.

3.3 Antitrust Metrics and Multi-Sided Platform Concentration

Antitrust research indicates that traditional market concentration metrics, such as the Herfindahl-Hirschman Index (HHI), systematically understate platform power in digital and AI ecosystems. Conventional HHI calculations rely on narrowly defined product markets (e.g., "consumer search" or "cloud database management"). However, AI platform operators manage multi-sided ecosystems that span multiple layers of the technology stack:

• Layer 4: Consumer and Enterprise Application Interfaces
• Layer 3: Foundational AI Models, Synthetic Data Pipelines, and API Networks
• Layer 2: Cloud Compute Hosting, Data Center Networks, and Power Infrastructure
• Layer 1: Custom Accelerator Silicon, High-Speed Interconnects, and Semiconductor IP

Vertical integration across custom silicon, cloud infrastructure, foundational model development, and consumer-facing applications enables platform companies to leverage dominance in one layer to capture adjacent value pools. Regulatory enforcement faces structural hurdles: traditional antitrust frameworks focus primarily on short-term consumer price harms, whereas AI platforms frequently offer subsidized or low-cost access during early market capture phases to establish ecosystem dependencies.

4. Macroeconomic Transmission Channels: Task Dynamics, Factor Shares, and Academic Debates

4.1 The Task-Based Production Framework

To model the macroeconomic impact of AI on factor shares, contemporary literature relies on the task-based production framework developed by Daron Acemoglu and Pascual Restrepo19. Aggregate economic output [Figure omitted from source export] is produced by combining a continuum of tasks normalized on the unit interval [Figure omitted from source export]: [Figure omitted from source export] Where [Figure omitted from source export] represents the elasticity of substitution among tasks. Tasks [Figure omitted from source export] are automated (produced by capital [Figure omitted from source export]), while tasks [Figure omitted from source export] are performed by human labor [Figure omitted from source export]20. Technological progress influences factor shares through three primary mechanisms19:

1. The Displacement Effect: As AI and robotics advance, the automation threshold shifts rightward from [Figure omitted from source export] to [Figure omitted from source export]. Capital replaces labor across tasks [Figure omitted from source export], directly contracting labor's share of value-added and exerting downward pressure on employment and wages in those tasks19.

2. The Productivity Effect: Automating tasks reduces production costs and increases Total Factor Productivity (TFP)20. Cheaper output lowers product prices, expanding total aggregate demand and increasing labor demand in remaining non-automated tasks [Figure omitted from source export]20.

3. The Reinstatement Effect: Technological innovation creates new, complex tasks where human labor holds a comparative advantage, extending the task horizon from [Figure omitted from source export] to [Figure omitted from source export]20. The reinstatement effect creates labor-intensive occupations, supporting labor's share of national income20.

When AI progress concentrates primarily on replacing labor without generating complex new human tasks—a dynamic termed "so-so automation" or "the wrong kind of AI"—the displacement effect dominates the productivity and reinstatement effects20. Under these conditions, capital's share of national income expands ([Figure omitted from source export]), labor's share contracts ([Figure omitted from source export]), and real wages for displaced workers stagnate or fall20.

4.2 Cognitive Automation vs. Physical Mechanization

A structural difference exists between 20th-century mechanization and modern cognitive automation. Mid-20th-century automation primarily replaced physical human labor in agricultural and assembly-line manufacturing tasks20. Displaced workers transitioned into growing administrative, professional, and service occupations20. In contrast, 21st-century generative AI and LLMs directly automate information processing, language synthesis, software design, and analytical decision-making—the cognitive tasks that absorbed labor displaced from physical production23. If AI simultaneously automates physical tasks through robotics and cognitive tasks through software, the range of remaining human-dominated tasks narrows, dampening the reinstatement channel20.

4.3 Impacts on Profitability, Wage Polarization, Asset Prices, and Purchasing Power

The macroeconomic transmission of AI adoption operates through four financial channels:

Corporate Profitability

Substituting scalable capital assets for human labor lowers operating expenses. Because software and automated capital scale at low marginal costs, corporate profit margins expand, particularly for capital-intensive lead firms.

Wage Growth Polarization

AI adoption impacts worker earnings non-uniformly23. Labor markets experience wage polarization: highly specialized technical workers who complement AI platforms see real wage gains, while workers performing routine or standardized cognitive tasks face increased displacement and wage pressure20.

Asset Price Dynamics

Expanding corporate profit margins increase the net present value of tradeable financial assets (equities, platform IP, and data infrastructure). Because equity ownership is concentrated in the top decile of households, stock market appreciation automatically expands top-tail wealth inequality3.

Household Purchasing Power

The net impact on household purchasing power depends on whether productivity gains lower consumer prices faster than nominal wages adjust. If market power enables platform owners to capture productivity gains as higher profit margins rather than passing them on as price reductions, purchasing power for wage-dependent households declines.

4.4 Opposing Economic Interpretations

Perspective A: Capital-Biased Disruption and "So-So Automation"

Economists such as Daron Acemoglu, Pascual Restrepo, Joseph Stiglitz, and Thomas Piketty argue that market-driven AI adoption is structurally biased toward labor displacement20. They contend that tax policies favoring capital investments, coupled with corporate strategies focused on cost-cutting, incentivize "so-so automation"—deploying automation that yields modest TFP gains while significantly displacing labor20. This framework posits that without institutional intervention, AI will reduce labor shares, accelerate wealth concentration, and depress aggregate consumer demand20.

Perspective B: Techno-Optimism, TFP Expansion, and Consumer Surplus

Conversely, economists and theorists such as William Nordhaus, Philippe Aghion, Jason Furman, and Erik Brynjolfsson maintain that AI represents a general-purpose technology (GPT) capable of elevating long-term Total Factor Productivity growth. They argue that market competition will compress corporate margins over time, passing cost savings to consumers as lower prices for AI-enabled goods and services. In this view, expanded consumer surplus, aggregate economic growth, and the creation of entirely new industries will raise real living standards and generate new employment opportunities, offsetting initial task displacement.

5. Historical Parallels: Productivity Acceleration and Wage Decoupling

5.1 The British Industrial Revolution and Engels' Pause

The most direct historical parallel to potential AI-driven capital concentration is "Engels' Pause," a term coined by economic historian Robert C. Allen to describe the prolonged decoupling of real wages from productivity growth during the British Industrial Revolution (1780–1840)25. During this sixty-year period, major technological innovations—including steam engines, mechanized cotton spinning, and advanced metallurgy—transformed British industry25. Economic data compiled by Crafts, Harley, and Feinstein reveal a distinct macro-structural pattern25:

• Between 1780 and 1840, real output per worker in Britain increased by approximately 46 percent25.
• Over the same six-decade period, real wages for industrial workers grew by only 12 percent25.
• Profit shares of national income surged, while labor shares fell significantly25. Capital owners captured the vast majority of productivity gains and reinvested them into expanded industrial capacity25.

It was only after 1840 that "Engels' Pause" concluded25. Between 1840 and 1900, real output per worker increased by 90 percent, and real wages grew by 123 percent, realigning wage growth with productivity growth and restoring labor's share of national income25. Engels' Pause demonstrates that technological revolutions can generate multi-decade periods of real wage stagnation, falling labor income shares, and high capital accumulation before market adjustment mechanisms or policy interventions realign compensation with productivity growth25.

5.2 The Information and Communications Technology (ICT) Epoch

A more recent parallel is the Information and Communications Technology (ICT) expansion from 1980 to 2010\. The adoption of enterprise software, personal computers, and telecommunications network infrastructure automated routine administrative tasks and clerical operations. While ICT adoption raised overall productivity, it contributed to labor market polarization, displacing routine middle-skill office jobs while raising demand for highly educated analytical workers8.

5.3 Comparative Analysis for the AI Era

Modern AI cognitive automation differs from historical precedents in its speed, scope, and capital intensity. While previous technological shifts allowed displaced workers to migrate from physical manual labor to cognitive administrative roles, AI directly targets cognitive tasks across multiple domains simultaneously23. If automated capital can be deployed at near-zero marginal cost, the economic time horizon required for market-driven labor reinstatement may exceed the workforce transition capacity of displaced populations20.

6. Institutional and Policy Mechanisms for Broadening Capital Ownership

6.1 Tax System Non-Neutrality and the "Robot Tax" Debate

A driver of capital substitution is the tax code's non-neutrality between human labor and capital investments24. In the United States and many OECD nations, corporate payroll and labor income are subject to personal income taxes and payroll contributions (e.g., Social Security, Medicare), generating effective marginal labor tax burdens ranging from 25 to over 50 percent24. Conversely, capital investments in automation software and robotics equipment benefit from preferential tax provisions, including accelerated depreciation allowances, Section 179 expensing, and lower effective capital gains rates24. This tax differential creates a policy bias: corporations face financial incentives to substitute automated capital for human workers, even in cases where human labor is pre-tax cost-effective21. To address this tax distortion, public economists debate automation levies or "robot taxes"24:

• Proponents (e.g., Bill Gates, Daron Acemoglu): Argue that levying taxes on automation equipment neutralizes fiscal biases against labor, moderates the pace of workforce displacement, and generates dedicated public revenue to finance retraining initiatives and social safety nets24.
• Opponents (e.g., Lawrence Summers): Contend that taxing capital equipment depresses Total Factor Productivity growth, stifles technological innovation, reduces international industrial competitiveness, and presents administrative challenges in defining what constitutes an automated system versus standard capital equipment24.

6.2 Employee Stock Ownership Plans (ESOPs) and Democratic Cooperatives

A policy mechanism for broadening capital gains capture is expanding employee ownership structures26. When workers hold equity stakes in their employers, productivity gains and capital dividends generated by AI adoption accrue to employees alongside wages26. Empirical research by Ethan Rouen and Thomas Dudley (Harvard Business School) models the distributional impact of broad-based corporate equity distribution26:

• If all private corporations in the United States were structured with 30 percent employee equity ownership, average American household net worth would nearly double26.
• For lower-wealth demographic groups, the impact is significant: median net worth for Black households would increase more than fourfold, rising from approximately \$24,000 to over \$106,00026.
• Quantitative data compiled by the National Center for Employee Ownership (NCEO) shows that employee-owners in ESOP firms possess 92 percent higher median net household wealth, earn 33 percent higher wages, and experience lower layoff rates during economic downturns relative to non-employee-owners in conventional firm structures27.

Worker cooperatives integrate equity ownership with democratic governance28. By providing workers with voting rights over firm capital allocations, technology adoption decisions can be evaluated for labor augmentation rather than workforce elimination28.

6.3 Sovereign Investment Funds and Public Compute Infrastructure

Where private equity ownership remains concentrated, state institutional mechanisms can socialize capital returns:

Sovereign Wealth Funds (SWFs)

Governments can establish public investment funds that acquire equity stakes in technology platforms, semiconductor foundries, and infrastructure developments. Modeled after the Alaska Permanent Fund or Norway’s Government Pension Fund Global, a sovereign wealth fund can capture capital returns and distribute them directly to citizens as an annual Universal Capital Dividend.

Public Compute Infrastructure Commons

To prevent cloud hyperscalers from controlling foundational AI development, governments can construct publicly funded compute infrastructure networks. Providing affordable compute capacity and open datasets to academic researchers, startups, and civic organizations lowers entry barriers and counteracts market concentration18.

7. Strategic Macroeconomic Scenarios through 2040

To evaluate long-term outcomes, four macroeconomic scenarios are modeled through 2040\. These scenarios vary based on institutional policy choices, tax structures, intellectual property regimes, and the concentration of AI infrastructure ownership.

7.1 Scenario A: Neofeudal Technopoly (High Concentration / Closed Proprietary IP)

In this scenario, regulatory oversight remains minimal, tax structures continue to favor capital investments over human labor, and intellectual property enforcement secures incumbent platform dominance24.

• Macroeconomic Outcomes: Cloud hyperscalers, chip designers, and semiconductor manufacturers capture the vast majority of TFP gains10. Capital's share of national income expands past 55 percent, while labor's share contracts below 45 percent6.
• Labor Market: AI software and physical robotics displace cognitive and manual tasks faster than new human tasks are created20. Real median wages stagnate, while executive and specialized technical compensation diverges6.
• Wealth Distribution: Stock market valuations appreciate, concentrating wealth in the top 1 percent (holding over 42 percent of net worth)3. The bottom 50 percent wealth share falls below 1.0 percent, creating persistent consumer demand shortfalls4.

7.2 Scenario B: Bounded Oligopoly with Universal Redistribution (High Concentration / Tax & Dividend Policy)

In this scenario, technology platforms maintain high market concentration due to scale advantages and capital intensity, but state intervention mitigates extreme wealth divergence.

• Macroeconomic Outcomes: Platform profits are taxed through targeted corporate capital levies, automation taxes, and digital resource extraction fees24. Capital shares remain elevated, but tax revenues fund social transfers.
• Labor Market: Private-sector wage growth remains subdued due to persistent task displacement20. State redistribution via Universal Basic Income (UBI) or Universal Capital Dividends buffers household disposable income.
• Wealth Distribution: Financial asset ownership remains concentrated, but state transfers support lower- and middle-income household consumption. The public sector acts as a central distributor of capital returns generated by platform oligopolies.

7.3 Scenario C: Open-Source Democratization (Low Concentration / Open Infrastructure)

In this scenario, breakthroughs in algorithmic efficiency, open-source foundational models, and decentralized compute networks lower entry barriers across the AI ecosystem.

• Macroeconomic Outcomes: Open-source models match proprietary performance benchmarks. Hardware costs fall due to specialized chip design and expanded global foundry capacity, diluting platform market power11.
• Labor Market: Startups and small-to-medium enterprises (SMEs) leverage open-source AI capabilities to launch products and enter markets, driving task reinstatement20. Labor's share of income stabilizes near historical averages as technology functions primarily as a labor-augmenting tool20.
• Wealth Distribution: Economic gains are distributed broadly across entrepreneurs, skilled workers, and new market entrants, suppressing top-tail wealth concentration.

7.4 Scenario D: Shared Capital Ownership & Equitable Distribution (Low Concentration / Policy-Engineered Shared Ownership)

In this scenario, structural reforms are enacted to broaden capital ownership, equalize capital-labor tax treatment, and mandate worker equity participation24.

• Macroeconomic Outcomes: Tax reforms establish neutrality between capital investments and labor compensation24. Federal policies incentivize the adoption of Employee Stock Ownership Plans (ESOPs) and worker co-ops across private enterprises26.
• Labor Market: Because workers hold equity stakes in their employers, productivity gains from AI adoption generate proportional increases in both labor earnings and capital dividend payouts26.
• Wealth Distribution: Average household wealth doubles, and wealth gaps contract significantly as equity ownership expands across the lower and middle deciles26. Rising TFP supports proportional expansions in aggregate consumer demand.

8. Evidence-Based Assessment and Policy Imperatives

The central research question of this investigation asks: If intelligent machines produce an increasing share of economic output, who captures the resulting wealth, and what happens to people whose primary economic asset is their ability to work? The economic evidence synthesized across historical trends, industrial organization, task-based modeling, and policy analysis leads to a definitive conclusion: Unmanaged market forces and current institutional frameworks will cause the widespread adoption of AI and robotics to concentrate economic power, corporate control, and income. This outcome is not driven by deliberate political intent, but by the structural characteristics of the modern AI platform paradigm:

1. Industrial Chokepoints and Capital Intensity: The advanced AI value chain is constrained by physical and technological bottlenecks—including ASML's monopoly in EUV lithography10, TSMC's dominance in advanced foundries10, Nvidia's accelerator position10, and hyperscaler cloud CapEx requirements18. These capital requirements prevent standard market self-correction through low-cost entry.

2. Asymmetric Substitution of Cognitive Labor: Unlike historical technological shifts that automated physical tasks while creating cognitive administrative roles20, advanced AI directly automates cognitive information-processing tasks23. If task displacement consistently outpaces task reinstatement, labor's share of national income will continue to decline6.

3. Tax Code Non-Neutrality: Current tax systems penalize human employment through heavy payroll and income taxes while subsidizing corporate capital investments in automated equipment24. This tax differential creates a policy bias toward replacing human workers with capital assets, even in cases where human labor is pre-tax cost-effective21.

Capital concentration is not an inevitable outcome of technological progress. Technology defines productive potential; institutional policy determines economic distribution. To prevent wealth concentration and ensure that artificial intelligence democratizes productive capacity, four policy interventions are required: First, Equalize Tax Treatment Between Labor and Capital: Tax codes should be reformed to establish neutrality between human labor compensation and capital investments in automation, eliminating policy biases favoring labor substitution24. Second, Expand Employee Equity Ownership: Governments should enact tax incentives and procurement preferences to expand Employee Stock Ownership Plans (ESOPs) and worker cooperatives26. Linking worker compensation directly to corporate capital returns creates a structural mechanism to prevent wealth divergence as technology automates production26. Third, Fund Public Compute Infrastructure: Public investments in open data center infrastructure, academic compute clusters, and open-source foundation models reduce entry costs for non-incumbent firms and dilute cloud hyperscaler market power18. Fourth, Establish Public Sovereign Wealth Dividend Funds: Socializing a portion of capital returns generated by AI platforms through public sovereign investment funds enables governments to distribute automated capital yields directly to citizens via Universal Capital Dividends. By transitioning from institutional frameworks that treat human labor solely as an operational cost to policies that provide workers with equity ownership in capital assets, modern economies can align technological progress with broad-based economic prosperity.

Works cited

1. The True State of the U.S. Economy \- The Century Foundation, https\://tcf.org/content/report/true-state-u-s-economy/

2. The Rise of Income and Wealth Inequality in America, https\://wid.world/document/the-rise-of-income-and-wealth-inequality-in-america-evidence-from-distributional-macroeconomic-accounts-world-inequality-lab-wp-2020-20/

3. Wealth of the American 1% and the rest of the US | USAFacts, https\://usafacts.org/articles/who-owns-american-wealth/

4. Wealth inequality in the United States \- Wikipedia, https\://en.wikipedia.org/wiki/Wealth\_inequality\_in\_the\_United\_States

5. Share of Financial Assets Held by the Bottom 50% (1st to 50th, https\://fred.stlouisfed.org/series/WFRBSB50193

6. A new look at the declining labor share of income in the United States, https\://www\.mckinsey.com/\~/media/mckinsey/featured%20insights/employment%20and%20growth/a%20new%20look%20at%20the%20declining%20labor%20share%20of%20income%20in%20the%20united%20states/mgi-a-new-look-at-the-declining-labor-share-of-income-in-the-united-states.pdf

7. Democratic Capitalism at the Crossroads: Technological Change, https\://dokumen.pub/democratic-capitalism-at-the-crossroads-technological-change-and-the-future-of-politics-9780691191843.html

8. Assessing the Impact of New Technologies on the Labor Market, https\://www\.bls.gov/bls/congressional-reports/assessing-the-impact-of-new-technologies-on-the-labor-market.htm

9. Daron Acemoglu \- EconPapers, https\://econpapers.repec.org/RAS/pac16.htm

10. 4 Major Companies Monopolizing the AI Chip Supply Chain | TestFlow, https\://testflowinc.com/blog/ai-chip-supply-chain-monopoly-nvidia-asml-tsmc-aws

11. Why AI Chips Cannot Exist Without TSMC and ASML, https\://note.com/nasdaq\_kun/n/n64e818fed287?hl=en

12. Nvidia's AI Boom Has a Hidden Dependency: One Dutch Company, https\://www\.tradingview.com/news/benzinga:d0d56c9c5094b:0-nvidia-s-ai-boom-has-a-hidden-dependency-one-dutch-company/

13. ASML vs. Taiwan Semiconductor \- The Motley Fool, https\://www\.fool.com/investing/2026/09/05/asml-vs-taiwan-semiconductor-which-semiconductor/

14. Im trying to write about the impact of ASML's monopoly on the EUV, https\://www\.reddit.com/r/Semiconductors/comments/1de3oqa/im\_trying\_to\_write\_about\_the\_impact\_of\_asmls/

15. ASML Holds a Monopoly on the Machines That Build AI Chips, https\://www\.barchart.com/story/news/5014126/asml-holds-a-monopoly-on-the-machines-that-build-ai-chips-trade-earnings-volatility-with-2x-leverage

16. The Strategic "Shovel Seller" Game in the AI Era \- 36氪, https\://eu.36kr.com/en/p/4025195546660994

17. How are Monopolies formed in the Generative AI Hardware, https\://www\.eetimes.com/how-are-monopolies-formed-in-the-generative-ai-hardware-ecosystem/

18. \$690B in AI Data Centers — US Buildout Map \- Value Add VC, https\://valueaddvc.com/ai-buildout-tracker

19. Automation, AI & Work | American Academy of Arts and Sciences, https\://www\.amacad.org/publication/daedalus/automation-ai-work

20. The wrong kind of AI? Artificial intelligence and the future of labour, https\://economics.mit.edu/sites/default/files/publications/The%20Wrong%20Kind%20of%20AI%20-%20Artificial%20Intelligence%20and.pdf

21. The Wrong Kind of AI? Artificial Intelligence and the Future of Labor, https\://ftp.iza.org/dp12292.pdf

22. The Simple Macroeconomics of AI, https\://shapingwork.mit.edu/wp-content/uploads/2024/05/Acemoglu\_Macroeconomics-of-AI\_May-2024.pdf

23. The wrong kind of AI? Artificial intelligence and the future of labour, https\://www\.researchgate.net/publication/345360530\_The\_wrong\_kind\_of\_AI\_Artificial\_intelligence\_and\_the\_future\_of\_labour\_demand

24. Should Robots Pay Taxes? Tax Policy in the Age of Automation, https\://conferences.law.stanford.edu/werobot/wp-content/uploads/sites/47/2018/02/Should-Robots-Pay-Taxes.pdf

25. Is the US Economy Having an Engels\\' Pause?, https\://conversableeconomist.com/2019/09/13/is-the-us-economy-having-an-engels-pause/

26. Employee ownership can be a 'powerful' tool to close the wealth gap, https\://www\.hbs.edu/bigs/employee-ownership-can-close-the-wealth-gap

27. Employee Ownership & \- Economic Well-Being, https\://ownershipeconomy.org/wp-content/uploads/2017/05/employee\_ownership\_and\_economic\_wellbeing\_2017.pdf

28. Reducing Economic Inequality through Democratic Worker-Ownership, https\://tcf.org/content/report/reducing-economic-inequality-democratic-worker-ownership/

29. Does employee ownership improve performance? \- IZA World of Labor, https\://wol.iza.org/articles/does-employee-ownership-improve-performance/long

30. Toward Shared Prosperity: The Promise of Employee Stock, https\://www\.aspeninstitute.org/blog-posts/toward-shared-prosperity-the-promise-of-employee-stock-ownership-plans/