The Disruption Dividend
Where the money actually is
Let's start with the numbers. AI-related technology names drove roughly 60% of the S&P 500's gains in 2025. OpenAI reached a $500 billion valuation. Anthropic hit $183 billion. IDC projects AI will add $19.9 trillion to global GDP by 2030. Goldman Sachs estimates generative AI will raise labor productivity by 15% when fully adopted, with specific task-completion improvements of 15–50% already documented across software development, legal work, customer support, and medical diagnostics.
This is real value creation. Not hype inflation — measurable output per unit of labor input increasing at a pace not seen since electrification. The question is not whether enormous value is being created. The question is where it is going.
The answer, so far, is: up. U.S. labor's share of national income fell to a record low in Q3 2025 — the lowest in a dataset stretching back nearly eighty years. A 2026 analysis across 238 regions in 21 European countries found a direct correlation between intensified AI adoption and declining labor income shares. The mechanism is simple: AI is capital-biased. It substitutes for labor at the margin, which means the productivity gains accrue to whoever owns the capital, not to whoever does the work. The worker who uses AI tools to do twice as much is rarely paid twice as much. The company deploying AI at scale captures the margin expansion.
The disruption is the funding mechanism. The same dynamics concentrating value at the top are generating the largest pool of investable capital in modern history. The question is not how to create new money for the transition — the money exists. The question is how to route it.
This is not a new problem. Every major technological transition in history has created this exact configuration: enormous new value, concentrated initially in the hands of those who owned the enabling capital, and then — over time, through a combination of political pressure, institutional innovation, and structural economic change — redistributed through mechanisms that didn't exist before the transition created them. The historical record on how this plays out is instructive. It is also sobering.
The industrial transition took 80 years
The first industrial revolution began in Britain around 1760. Real wages for factory workers remained largely flat for sixty years. The surplus from mechanized production went almost entirely to factory owners and the landed gentry who owned the land on which industrial towns were built. Child labor was normal. Twelve-hour workdays were standard. The new productivity gains were real; their distribution was catastrophic.
The redistribution mechanisms that eventually worked came from three directions, and none of them were fast:
Labor organization. British trade unions were illegal until the Combination Acts were repealed in 1824, and remained severely restricted for decades after. The Chartist movement of the 1830s and 40s — the first large-scale working-class political movement — fought for basic democratic rights as preconditions for economic rights. Real wage gains for industrial workers didn't become consistent until the 1850s and 1860s, roughly a century into industrialization. The mechanism was straightforward: concentration of labor in factories made collective action possible. Workers who couldn't bargain individually could withdraw labor collectively. The credible threat of shutdown created negotiating leverage that had not previously existed.
Progressive taxation and public infrastructure. The income tax in its modern form — graduated rates, applied to capital as well as wages — was a 20th century innovation. The UK introduced graduated income tax in 1909. The U.S. federal income tax was established by constitutional amendment in 1913. The New Deal programs of the 1930s used fiscal policy at scale: public works employment, agricultural price supports, banking regulation, and eventually Social Security. These were not charity — they were circuit breakers for demand collapse. Without consumer purchasing power, the industrial economy ate itself. The redistribution mechanisms weren't moral; they were structural maintenance.
Social insurance. Bismarck introduced the first social insurance programs in Germany in the 1880s — health insurance in 1883, accident insurance in 1884, old-age insurance in 1889. Not for humanitarian reasons: to forestall socialist revolution. The calculation was that a worker who couldn't be destroyed by illness or old age was less likely to throw bombs. The same calculation runs through every major welfare state expansion in history. The redistribution was paid for by the surplus the industrial economy had generated, routed back to prevent the conditions that would destroy the industrial economy.
The lag from disruption to redistribution was roughly 80 years in the first industrial revolution. We do not have 80 years. The pace of the current transition is measured in model generations, not parliamentary terms. The mechanisms need to be designed faster than they were discovered last time.
What the historical record tells us is which mechanisms actually moved money: collective bargaining, progressive taxation, mandatory social insurance, public ownership of infrastructure. Not charity. Not voluntary corporate responsibility. Not the goodwill of industrialists. Structural mechanisms with enforcement teeth.
Impact investing: what works, what is theater
The impact investing market is now estimated at over $1 trillion in assets under management globally. ESG funds have proliferated across every asset class. Green bonds, social bonds, sustainability-linked loans — the apparatus of mission-aligned capital is enormous and growing. Most of it is not doing what it claims to do.
The evidence on ESG is pointed. A major 2026 study on Europe's Sustainable Finance Disclosure Regulation — the most ambitious attempt to standardize green investment disclosure in the world — found that the regulation had no meaningful effect on mutual fund flows. Investors already knew which funds were "light green" and "dark green" from marketing materials; the disclosure requirements added no new information and changed no behavior. The signal was already priced in, such as it was.
The structural problem with ESG is that it is a negative screen, not a positive allocation mechanism. You exclude the worst actors. You do not route capital toward the best outcomes. A portfolio that doesn't own coal companies does not, by itself, fund the energy transition. The capital that would have gone into coal goes into the next-best risk-adjusted return, which is usually not wind turbines in underserved communities — it is large-cap tech.
Social impact bonds are more interesting and more limited. The model: private investors fund a social intervention (a job training program, a reentry housing program, an early childhood education initiative), the government pays back the investors if pre-agreed outcomes are achieved, and the investor earns a return tied to the outcome rather than the activity. As of 2023, 276 SIB projects had launched across 23 countries, raising $745 million in capital. That is a proof of concept, not a movement. The limitation is definitional: SIBs require outcomes that are quantifiable in advance, attributable to the intervention, and verifiable by an independent party. Real social problems — chronic poverty, structural racism, climate-driven displacement — resist these requirements. You can build an SIB around recidivism rates in a county jail. You cannot easily build one around the long-run earnings of a generation.
Impact investing works at the margin. It does not work as the primary financing mechanism for a paradigm transition. The mechanisms that do that kind of work are not voluntary — they are structural, political, and built into the operating architecture of the economy itself.
What does work in impact investing, at smaller scale: community development finance institutions (CDFIs), which use patient capital to underwrite loans in markets that commercial banks won't touch. Development finance institutions like the IFC, which can take first-loss positions that crowd in private capital in emerging markets. Catalytic grants that de-risk early-stage interventions until they reach commercial viability. These are real mechanisms with real track records. They share a common feature: they operate at the edges of commercial markets, doing deals that market-rate capital won't do, to prove concepts that market-rate capital will later fund. They are bridge mechanisms, not end states.
Public capture: the sovereign wealth playbook
The cleanest model for capturing resource-extraction value and routing it to broad benefit is the sovereign wealth fund. Norway's Government Pension Fund Global is the template. Established in 1990, funded by petroleum revenues, and now sitting at $1.9 trillion in assets — roughly 1.5% of all listed equities on earth — it is the largest sovereign wealth fund in the world. The operating principle is elegant: only the return is spent, never the principal. Parliament can allocate up to 4% of the fund annually. The underlying capital grows, and future generations are funded by the surplus of the current one.
Alaska's Permanent Fund is the distribution variant. Rather than using fund returns to pay for government services, Alaska distributes a direct annual dividend to every resident — typically $1,000–$2,000 per person per year. It is, functionally, a universal basic income funded by oil extraction. No application required. No work requirement. No means test. Every Alaskan gets a check because Alaskans collectively own the oil beneath their land.
The case for treating AI productivity as an extractive resource — the way Norway treats petroleum — is not frivolous. The productive inputs to AI are largely commons: human-generated data scraped from the public internet, publicly funded university research, decades of publicly subsidized semiconductor development, the infrastructure of the internet itself built on public spectrum and public right-of-way. The value created by AI is not purely private invention. It is private extraction from a public commons. The same argument that justified public capture of oil revenues applies.
Norway did not nationalize its oil industry. It taxed it at 78% on upstream profits and required state participation in all production licenses. The private sector did the extraction work. The public captured the surplus. That architecture is available for AI.
Several proposals are being actively developed. Andrew Yang's value-added tax on AI and robotics output, routed to a UBI — the version he ran on in 2020 — has resurfaced in updated form: in March 2026, he proposed stopping taxes on labor entirely and replacing the revenue with taxes on AI. The IMF has separately proposed excess profits taxes on AI-generated returns and usage levies tied to AI deployment. The Tax Foundation and Brookings have both published frameworks for digital services taxes that follow sound consumption-tax principles without deterring investment.
The mechanism design question is real: you want to tax the surplus, not the activity. A compute tax discourages infrastructure investment in the same way a steel tax would have discouraged the rail network. A consumption tax on AI services, or a progressive profits tax on AI-derived earnings above a threshold, captures the surplus without choking the engine. The Norway model — high rates, applied to profits not capital, with a public vehicle for the proceeds — is the template that has actually worked at scale.
New mechanisms: the toolkit being assembled now
Several financial mechanisms are being tested at the edges of the current economy that could scale into structural tools for the transition. None are complete. All are worth understanding.
Data dividends
The productive input to most AI systems is human-generated data — text, images, behavioral signals, health records, conversations. This data was generated by individuals who received no compensation for its use as training material. A data dividend is a payment to individuals or communities for the commercial value extracted from their data. California has explored legislation. The European Data Governance Act creates frameworks for data intermediaries who can negotiate collective data licensing on behalf of individuals. The mechanism is nascent, but the principle — you own something of value that others are extracting for commercial gain, you should receive consideration — is sound and increasingly legally defensible.
Digital public infrastructure
Open source software is already a form of digital public infrastructure that distributes value broadly without requiring redistribution from private actors. Linux runs most of the internet and costs nothing to copy. The value it creates is enormous; the capture is decentralized. Red Hat, before its acquisition by IBM, built a $1 billion business providing enterprise support and certification on top of free software — paying developers, funding contributions back upstream, monetizing the service layer while the underlying resource remained common. Automattic does the same with WordPress. The model — open core plus commercial services — is a proven architecture for sustainable open source that distributes economic value through the ecosystem rather than concentrating it in a single owner.
UBI and guaranteed income
The experimental evidence on unconditional cash transfers is now substantial. Finland's two-year randomized controlled trial found improvements in health, wellbeing, and cognitive function among recipients. Stockton's SEED program found that $500/month unconditional payments led to significantly higher full-time employment (not lower), with recipients using the stability to find better jobs rather than exiting the workforce. Across every major pilot — Finland, Kenya, India, Stockton, multiple U.S. city programs — the fears that underpin opposition to cash transfers (people will stop working, spend money on vices) are empirically false. People buy food, pay rent, and take better jobs.
The constraint on UBI is not evidence — it is funding. Stockton's program was philanthropically funded. Finland's was time-limited and small. The missing piece is a durable public funding mechanism: a sovereign AI fund, a profits levy, a data dividend pooled into a common vehicle. The transfer mechanism works. The capture mechanism is what needs building.
The VC problem and how to route around it
Venture capital is the dominant financing mechanism for technology companies. It is also structurally designed to concentrate returns. A standard VC fund raises capital from limited partners — endowments, pension funds, family offices — deploys it into a portfolio of startups, and returns proceeds when those startups exit via acquisition or IPO. The return to LPs depends on a small number of portfolio companies achieving extreme multiples. The model requires power-law outcomes: a few companies that return 100x, and the rest that lose money or return 1x. This is not a design flaw. It is the design.
The consequence is that VC systematically selects for companies with extreme return potential — which means companies building toward monopoly or near-monopoly positions, in markets large enough to support multibillion-dollar exits. Companies structured to share returns broadly do not produce the multiples the model requires. A worker-owned platform cooperative that generates $50 million in annual surplus and distributes it to its worker-members is a spectacular outcome for everyone involved — and an unacceptable outcome for a venture investor who needed a 50x return.
The problem is not that VCs are bad actors. The problem is that the incentive architecture of the asset class is incompatible with broad distribution of returns. You cannot fix this through better values. You fix it by building alternative capital structures that are compatible with broad distribution by design.
Several alternatives exist and are scaling:
Revenue-based financing. Investors receive a percentage of revenue until they've received a capped multiple of their investment — typically 1.5x to 3x. No equity, no control transfer, no exit required. The company and investor interests align around sustainable revenue growth rather than explosive scale and exit. Clearco, Capchase, and Pipe have built this into a multi-billion-dollar asset class for software companies. It is compatible with cooperative or employee-owned structures in a way that equity VC is not.
Steward ownership. The Patagonia model — transferred in 2022 to a purpose trust and a nonprofit — is the highest-profile example of a growing movement. In steward-owned structures, voting control is held by people connected to the company's operations and values, and cannot be sold or inherited for financial gain. Economic rights are separated from control rights, and profits are either reinvested in the mission or distributed to a broader set of stakeholders. The Purpose Foundation and European impact networks have built frameworks for this — it is a real, legally operable ownership structure, not a theoretical model. Zeiss, Bosch, and a number of European companies have operated under functionally similar structures for over a century.
Platform cooperativism. Worker- or user-owned platforms organized as cooperatives. Stocksy, the artist-owned photography cooperative, generated $3.7 million in revenue in its first full year and distributed a meaningful share to contributing artists — something no Getty or Shutterstock model will ever do. Mondragon, the Basque cooperative federation founded in 1956, now employs over 80,000 people across finance, manufacturing, and education, with a pay ratio between the highest and lowest earners that has historically been capped at 6:1. These are not utopian experiments. They are operating businesses with decades of performance data.
ESOPs. Employee Stock Ownership Plans are the most widely used broad-ownership mechanism in the United States. Over 14 million Americans hold equity in their companies through ESOPs. They are not a transformative model — they still require an exit-oriented capital structure — but they are a proven mechanism for distributing ownership more broadly than standard equity allows, and they are compatible with businesses that would not otherwise attract VC.
What builders should actually do
Here is the practical question: you are building something. You want the value to distribute broadly. What does the financial architecture actually look like?
The answer depends on where you are in the stack and what kind of value you are creating. But several principles hold across most configurations.
Design the ownership structure before you take the first dollar
The moment you take a standard equity investment, you have committed to an exit-oriented capital structure. The path dependency is set. Everything downstream — hiring decisions, pricing decisions, growth decisions — gets filtered through "does this maximize our exit valuation." If your goal is broad distribution of value, the time to make that structural commitment is before investors arrive, not after. This means understanding the full menu: cooperative incorporation, LLC with profit-sharing provisions, purpose trust overlay, B-corp certification with genuine operational teeth, ESOP provision in your cap table from day one.
None of these preclude raising capital. They change the terms on which you raise it. Revenue-based financing is available for software companies with recurring revenue. Mission-aligned investors — impact funds, family offices with long time horizons, development finance institutions for specific problem domains — will accept lower returns in exchange for mission alignment. The capital exists. The structures to receive it correctly need to be in place.
Open source the infrastructure, charge for the service layer
The most durable model for distributing technological value broadly is the Red Hat / WordPress / Linux architecture: build the infrastructure in the open, fund development through a commercial services layer, contribute improvements back upstream. This is not altruism — it is a competitive moat. An open-source core that the community contributes to is harder to replicate than a proprietary one. The community becomes a distributed R&D function. The commercial services layer captures value from organizations that need reliability, support, and customization — while the underlying capability remains freely available to everyone who can use it without support.
For AI specifically: models trained on public data and released openly, fine-tuning and deployment services sold commercially. The capability distributes. The return flows to the entity that provides reliable productization of it. This is not a perfect model — open-source AI has its own complications around safety and dual-use risk — but as a financial architecture for broad distribution of capability, it has the strongest track record of any mechanism we have.
Build the data relationship correctly from the start
The data generated by users of your platform has value. Most companies treat this as unilateral extraction — users generate data, company monetizes it, users receive nothing. The alternative — data dividends, data cooperatives, collective licensing structures — is not yet legally standardized but is moving in that direction. Companies that build the data relationship as a genuine exchange from the start are better positioned when regulation arrives (and it will), have better trust relationships with their user base, and are building toward a world where the value of data flows back to the people who generated it.
The builders who will look right in ten years are the ones designing toward the regulatory and social environment that is obviously coming, not the ones maximizing extraction in the window before it arrives. The ones who extracted before Napster are not celebrated. The ones who built Spotify are.
Treat your productivity gains as a surplus to be distributed, not a margin to be captured
If your company uses AI to do twice as much work with the same headcount, you have a choice: capture that as profit, or distribute it as wages, reduced prices, or investment in expanded capability. The profit-capture option is rational at the individual company level and destructive at the system level — it is exactly the dynamic producing record-low labor income share. The distribution option is harder to execute, requires investor alignment, and is better long-run business strategy. Companies with distributed ownership outperform conventionally owned ones on most long-run performance metrics. Worker-owners maintain equipment better, take fewer sick days, and build toward the long term in ways that hired labor with no stake does not.
None of this is about virtue. The distributional model is more resilient, more aligned with the direction of policy and social expectation, and — at the timescales that matter for building durable institutions — more profitable. The question is whether you are optimizing for the next funding round or for the next generation.
The transition is happening. The value creation is real and accelerating. The concentration is real and accelerating faster. The mechanisms for routing that value toward broad benefit exist — sovereign wealth models, progressive profits taxation, open-source infrastructure, cooperative ownership, revenue-based financing, data dividends, guaranteed income. None of them are new inventions. All of them have working precedents. The gap is not invention; it is deployment at the speed and scale the transition requires.
The last industrial transition took eighty years to produce its redistribution mechanisms. The window for this one is shorter. The good news is that the mechanisms don't need to be invented — they need to be chosen, built into the architecture of the things being built right now, and operated at scale by the people who understand how they work. That is a builder problem, not a policy problem. Policy will follow the structural facts on the ground. The structural facts on the ground are being set today, in cap tables and ownership agreements and data licensing terms and open-source licenses. The architects of the next paradigm are not in legislatures. They are reading this.