Who Owns the Machines?

As machines do more of the producing, who is legally entitled to what they produce — and does the current answer still make sense?

In the third quarter of 2025, the share of American economic output paid to workers as wages and compensation fell to 53.8% — the lowest figure recorded since the Bureau of Labor Statistics began tracking it in 1947, when the number stood closer to 70%. Over the same stretch, the top 1% of U.S. households came to hold 31.7% of the nation's wealth, also a record, worth roughly as much as the entire bottom 90% of the country combined. These are not predictions. They are what has already happened, prior to any scenario in which automation displaces labour at scale. This paper asks a narrower, more mechanical question than "is this fair": given that a shrinking share of output is going to labour and a growing share of wealth is concentrating in a small ownership class, what alternative ownership architectures actually exist, and does any of them have a working precedent rather than just a proposal?

• U.S. labour share of GDP — the percentage of output paid to workers — has fallen from roughly 60–65% through most of the postwar period to 53.8% in Q3 2025, the lowest level on record.
• The top 1% of U.S. households held 31.7% of household wealth in Q3 2025, an all-time high in Federal Reserve records dating to 1989, equal to roughly $55 trillion — comparable to the combined wealth of the bottom 90% of the country.
• Equity ownership is substantially more concentrated than wealth generally: the wealthiest 1% of Americans hold more corporate stock than the bottom 90% combined, and the top 10% of households control the large majority of all corporate equity and mutual fund shares.
• A working precedent for broad-based capital ownership already exists in the United States: the Alaska Permanent Fund, a sovereign wealth fund built from a fixed share of oil royalties since 1976, held $85.1 billion in assets as of mid-2025 and has paid a direct, unconditional annual dividend to every eligible resident for over four decades — $1,000 in 2025, historically ranging as high as $3,284.
• No comparable fund currently exists that is capitalized by, or entitles citizens to, income generated by automation or AI specifically — every existing large-scale model is resource-based, not technology-based.

Private ownership (status quo). Capital returns accrue to whoever owns the equity — founders, investors, shareholders. This is the default architecture and the one under which the wealth concentration figures above have occurred. Its efficiency case is straightforward: capital flows to whoever can deploy it best, with no redistribution overhead. Its distribution case is what the data above describes.

Social wealth funds / universal capital. A public fund, capitalized by a share of resource royalties, tax revenue, or (in an automation scenario) a levy on machine-generated output, invests broadly and pays dividends to citizens as a matter of ownership right, not welfare. Alaska's Permanent Fund is the clearest working example in the U.S., though it is resource-funded rather than automation-funded, and its dividend — a few hundred to a few thousand dollars annually — is a modest supplement, not a replacement income. Norway's Government Pension Fund Global operates on a similar principle at a far larger scale ($1.7+ trillion), though it reinvests returns into the state budget rather than paying direct dividends. The architecture is proven. What has never been tried is capitalizing one specifically with automation- or AI-derived income.

Worker and employee ownership. Cooperatives and employee stock ownership plans (ESOPs) attach capital returns directly to the people doing the labour, rather than to a separate investor class. This directly counteracts the labour-share decline by keeping ownership and labour in the same hands, but it does not obviously scale to a labour market where automation reduces the number of workers needed in the first place — a smaller workforce owning a stake in a more automated firm still concentrates gains among fewer people, just a different fewer.

Public AI infrastructure. Rather than distributing dividends from privately-owned automation, the state owns or heavily regulates the automating infrastructure itself — compute, data, foundational models — as a public utility, similar to how some infrastructure (power grids, in some countries) is publicly owned. No functioning large-scale precedent exists yet in any economy; this remains the most theoretical of the four models, proposed in policy literature but untested.

The single most important fact this research turned up is a negative one: there is no existing example, anywhere, of a fund capitalized directly by automation- or AI-generated income and distributed to citizens as a matter of ownership right. Every "AI dividend" or "robot tax" proposal currently in public discussion is a proposal, not a case study. The closest working analogues — Alaska, Norway, and smaller resource funds elsewhere — prove that the mechanism (collect a share of an asset class's returns centrally, distribute or reinvest broadly) works and has worked for decades. They do not prove it would work at automation's scale, speed, or political economy, which differs from oil in at least one important way: oil revenue arrives to a state government by default, through royalties on extraction it already controls, while AI-generated income currently accrues directly to private firms with no equivalent public claim built in.

That gap — between a proven mechanism and an unproven funding source — is where the real policy argument sits, and where the Ownership control in the Labour Machine model on this site is deliberately built as a spectrum rather than a binary, since the honest answer to "which model wins" is that none of them currently has evidence at the scale the automation question requires.

This paper draws on published Federal Reserve and Bureau of Labor Statistics data on labour share and wealth distribution, and on public financial disclosures from the Alaska Permanent Fund Corporation, rather than original data collection or modelling.

• Labour share and wealth concentration trends observed through 2025–2026 are treated as a continuation of a multi-decade structural pattern rather than a cyclical, temporary fluctuation, consistent with BLS data showing the decline predates and persists across multiple economic cycles.
• The Alaska Permanent Fund and similar sovereign wealth funds are treated as valid structural precedents for a universal-capital model, despite being resource-funded rather than automation-funded, on the basis that the distribution mechanism is fund-source-agnostic.

• No direct empirical evidence exists yet for how an automation- or AI-funded dividend would behave at scale; the Alaska comparison is structurally informative but not a direct test.
• Wealth and labour-share figures are aggregate national statistics and do not capture significant variation by industry, region, or occupation, some of which may be more or less exposed to automation than the average.
• This paper does not model the political-economy question of how a public claim on private AI-generated income would be established or enforced — only whether comparable distribution mechanisms have precedent.

• U.S. Bureau of Labor Statistics, Labor Productivity and Costs report, Q3 2025 (labour share data).
• Federal Reserve Board, Distributional Financial Accounts, Q3 2025 (household wealth and equity concentration).
• Elsby, M., Hobijn, B., & Şahin, A. The Decline of the U.S. Labor Share. Brookings Papers on Economic Activity.
• Alaska Permanent Fund Corporation, 2025 Annual Report; Alaska Department of Revenue, Permanent Fund Dividend Division.