Who Owns the Machines?

Distribution becomes secondary once ownership of productive intelligence concentrates.

A pay stub and a brokerage statement can fit on the same phone screen. One says somebody needed your time. The other says somebody owes you money because you own a small piece of something.
We have spent two centuries getting used to the first relationship. You work. A company pays you. Part of the money disappears into rent, groceries, taxes, a car payment, perhaps children with the irritating habit of needing shoes again. Some of what survives may reach a pension, retirement account or a few shares.
Labor produces the cash that allows a worker to become, slowly and usually modestly, an owner of capital.

If machines produce an increasing share of economic value, arguments about wages eventually run into another question.
Who owns the machines?
The question sounds more literal than it is. Ownership does not require a humanoid robot with somebody's name engraved on the back. Productive intelligence can sit across semiconductor companies, data centers, power contracts, cloud providers, model developers, patents, software platforms and the businesses using all of them.
Somebody will own the claims on those assets.

The machine economy will have a cap table. That cap table may matter more than the personality of the machine.

Before the money exists

Take two people receiving $1,000. The first receives it from a public transfer financed by taxes on highly automated companies. The second receives it as a return from a fund that owns shares in those companies.

The grocery store does not care. One thousand dollars buys the same milk.

Their relationship with the productive system is different.
The first person's claim begins after production. A company earns money, the state taxes part of it, politicians approve spending and a transfer reaches the household.
The second person's claim sits inside production. The company earns money and part of the return belongs to the fund because the fund owns an asset.

Property is not immune to politics. Taxes can change. Shareholders can be diluted. Governments can regulate, nationalize or expropriate. Funds can fail. A legal claim is not a force field.

Still, sequence matters. One claim begins after production. The other sits inside production.
That distinction gets lost when arguments about AI and inequality move immediately to universal basic income.
UBI can solve a cash problem. It does not by itself answer a property question.
A person receiving a transfer can be materially secure and own almost none of the productive economy. A person receiving capital income may do little work and retain a legal claim on production.
Assets also behave differently from monthly payments. They can compound. Depending on their structure, they can be sold, inherited, borrowed against or carry voting rights. They can survive this month's political argument.
This does not automatically make an owner powerful. My pension may contain shares in hundreds of companies and still leave me unable to influence any of them.

Ownership, income and control are not the same thing.

There is another problem before we even reach that one. We are not entering the machine economy with ownership evenly distributed.
Federal Reserve data for the second quarter of 2026 show that the wealthiest 10 percent of U.S. households hold 88.1 percent of corporate equities and mutual-fund shares. The bottom half holds 0.6 percent. These numbers do not tell us who will own future AI systems. The United States is not the world. They tell us where much of the existing financial ownership in the world's largest economy already sits.

Machine capital would be landing on top of that structure.
The easiest future to build is usually the one that requires nobody to change the paperwork.
A company buys chips, it rents computing power, or it builds a data center or signs a contract with somebody who already owns one.
This buys access to a model. It replaces part of a department with software.
Costs fall, output rises or some combination of the two happens. If the investment works, profits belong to the company. The residual claim belongs to its owners.
Nothing strange happened. Nobody needed to decide that ordinary people should own less of the machine economy. Every actor followed an ordinary incentive.

That is precisely the problem.

The corporation already knows the answer
A corporation already has legal identity, property, contracts, bank accounts and persistence beyond the lives of the people working inside it. Artificial intelligence can add more cognition to an entity that already knows how to own things. The corporation can own the machine.

Who owns the corporation?

The question becomes harder once the machine stops being a machine in the familiar sense.
A manufacturer may own physical robots and rent intelligence from a model provider running on a cloud provider using chips produced by another company in a data center financed by an infrastructure fund and powered under a twenty-year electricity contract.
You can point at the robot, but... you cannot point at the ownership structure.
The money can. Every layer has contracts, creditors, shareholders and beneficiaries. And the cap table is not the entire document.

In September 2026, SoftBank raised $11.1 billion through a high-yield bond sale as it financed large AI investments, including its OpenAI commitment and other technology acquisitions. The people lending that money do not own OpenAI because they bought a SoftBank bond. They still hold a claim that has to be serviced before SoftBank's shareholders enjoy whatever remains.

AI-related borrowing has also been spreading through leveraged finance and private credit as companies search for ways to finance increasingly expensive infrastructure.

The machine economy will have owners. It will also have lenders.
They are not interchangeable. Equity participates in the upside and usually absorbs losses first. Debt receives contractual payments and stands elsewhere in the queue if things go badly.
Who owns the machines is still the right question. It just has a second question hiding underneath it.
Who gets paid first?

The bridge

Pension funds make this more interesting than it first appears.

At the end of December 2025, CPP Investments managed C$780.7 billion supporting the Canada Pension Plan. Norway's Government Pension Fund Global was worth 22.683 trillion kroner at the end of June 2026.

These institutions are very different. Neither turns a Canadian or Norwegian into somebody checking NVIDIA shareholder resolutions over breakfast.
They demonstrate something simple. Large populations can already have claims on productive capital through institutions standing between the person and the company.
This is old technology.
The interesting part is how many of those claims begin with work. A worker gets paid. Some money enters a pension plan. An employer contributes. A retirement account buys shares. Private savings buy a fund.

Years pass.

The worker who originally sold labor now owns a small claim on thousands of other people's labor, buildings, machines, intellectual property and future profits.
There is a bridge from worker to owner.
We rarely describe employment this way because most people experience the process from the wrong end. Payday feels less like capital formation and more like surviving another month.

Payday is rent day with better marketing.

Yet employment has been one route through which households without inherited wealth gradually acquire assets. That creates a second-order problem if AI reduces labor income.

The obvious loss is the wage. The less obvious loss is part of the mechanism through which the wage earner might have become an owner.
A person can lose twice. First, machine capital weakens the price of his labor. Then weaker labor income reduces his capacity to accumulate claims on machine capital.
For somebody who already owns substantial capital, the loop can run in the other direction.

Successful AI investment produces returns. Returns buy more assets. Those assets can finance more compute, companies, infrastructure and intellectual property. More capital generates more claims on future production.
None of this requires every AI company to succeed.
Many will fail. Some will burn mountains of money proving that adding the word intelligence to a product does not suspend arithmetic. Humans managed the same trick with the word internet a quarter-century ago.
The mechanism requires only that successful machine capital be unevenly owned.
A worker competing with machines can lose leverage today and lose part of his route into ownership tomorrow.

An old problem with a new asset

Societies have faced versions of this before.
Natural resources made the ownership question difficult to ignore because the asset was physically sitting under somebody's feet.

Oil provided an especially rude version of the problem. A resource that took geological time to form could be extracted in decades, sold, and leave behind enormous wealth or very little depending on who held the claims.

Alaska built one answer. The Alaska Permanent Fund converted part of the state's oil wealth into financial assets, and eligible residents receive a yearly dividend. The 2025 dividend was $1,000. A resident does not receive a personal section of pipeline or a few barrels of crude. The productive claim has been turned into a financial institution and, eventually, money in a bank account.

Norway built another answer at a much larger scale. Petroleum revenues helped create a fund that now owns financial assets around the world. The fund has become so large that questions about Norwegian oil wealth are now partly questions about ownership of companies that have nothing to do with oil.

AI is not petroleum.
The analogy breaks quickly if pushed too far.
Oil is scarce, physical and tied to geography. Software can be copied. Compute can be manufactured. Models can cross borders. A government can assert rights over minerals under national territory more easily than over every useful inference produced by a privately funded model.
The useful part of the comparison is narrower.
A society encountered a productive asset capable of generating wealth beyond the wages of the people directly working on it and had to decide where the ownership claim should sit.
AI may force a similar decision, except it can arrive disguised as thousands of purchasing decisions. No parliament needs to vote on "who shall own artificial intelligence."

Procurement departments can answer the question one invoice at a time.

Public money, private compounding

This is also why public compute is interesting and insufficient.

Europe is already putting public money into AI infrastructure. The European Commission says investment by the Commission, member states and associated countries in supercomputing infrastructure and AI Factories will reach €10 billion over 2021 to 2027.
In July 2026, the EU opened a call for up to seven AI Gigafactories, backed by as much as €10 billion in EU and national funding and intended to mobilize at least €20 billion in private investment.

The machines can be public at one layer and private at the layer where the compounding happens.
Capital has always been good at finding the floor where the cash register sits. A publicly supported data center can give startups, researchers and companies cheaper access to compute. If one of those companies becomes enormously valuable, ownership of that company is a separate matter.

This is where ideas often bundled together begin to split apart.
A sovereign fund gives citizens an indirect economic claim through the state. A pension fund gives contributors and beneficiaries claims through a retirement institution.
Citizen equity could attach financial assets more directly to individuals.

Public compute can make a productive input widely available without distributing ownership of businesses using it. Capital grants could give people assets at some point in life and let them decide what to do with them. Dividends distribute cash from an asset.
These arrangements can all put money into a household.
They do not create the same relationship with production.

Property has anatomy

The word ownership hides too much.
Three questions are useful.
Who receives the income?
Who controls the asset?
Who can sell, transfer or inherit the claim?

Those rights can belong to different people.

A pension beneficiary has an economic interest in a portfolio but does not personally manage it.
A fund manager may exercise voting rights without being the ultimate beneficiary of the money.
Citizens may be described as owners of a sovereign fund but have no individual asset they can sell.
A shareholder can own stock yet exercise almost no influence over a company.
A founder can own a minority position and retain substantial control through voting structures.

Property has anatomy. Think of it like the sum of body parts.

Income rights are one bone. Control is another. Transfer, inheritance, collateral and exit sit elsewhere in the skeleton. Debt adds another organ to the creature. A lender may have no vote over ordinary operations and still possess contractual rights that become very powerful when the borrower cannot pay.

This makes the attractive phrase "broaden ownership" much less comforting.

Give every citizen shares in a national AI fund and a new set of problems appears immediately.

Can the shares be sold? If yes, people facing debt, illness or ordinary bad luck may sell them. Ownership can reconcentrate. If no, what exactly does ownership mean when the supposed owner cannot dispose of the asset?
Can the shares be inherited? If yes, family wealth begins compounding across generations. If no, the claim behaves less like ordinary property.
Can people borrow against them? Who votes? Who selects management? Can the government change the rules? Can somebody leave the country and keep the claim? Who qualifies in the first place?

A policy that begins as a machine-ownership problem can end in an argument about citizenship, family law and borders. Birth certificates start to resemble share certificates. That is not an objection to broad ownership. It is an objection to pretending the noun solves the problem.

Owning without deciding

Norway offers a useful reminder. The fund is often described as belonging to the Norwegian people. That does not mean millions of citizens log into an app and decide how the fund votes at corporate meetings. Institutions stand in between. This is probably unavoidable at scale. A modern portfolio can contain thousands of companies. Most people have better uses for Thursday evening than studying executive compensation proposals from a chemical manufacturer in Osaka.

Delegation creates its own power. The person with the economic claim may not be the person exercising control. That tension already exists in pensions, mutual funds, index funds and sovereign wealth funds. Machine capital would make it more consequential if capital income becomes more important relative to wages.

A society could spread economic ownership widely and leave operational control highly concentrated. It could distribute dividends without distributing votes or it could distribute votes that almost nobody exercises. It could create a public fund whose managers become one of the most powerful institutions in the country. It could hand assets directly to citizens and watch some sell early while others accumulate.

Different arrangements move power around.
None removes it.

There is also a serious objection to the ownership argument itself. If ownership can be so indirect, fragmented and weak, perhaps income is what matters after all. If the citizen gets his $1,000, why care whether it arrives as a dividend, benefit or tax credit?
At low levels of wealth, the objection has force. A family short on rent does not need a seminar on capital structure. It needs money.

The difference appears with time. Income pays for consumption. Ownership keeps a claim alive after this month's payment is gone. Two households receiving identical monthly cash can drift apart if one owns the source and the other depends on a recurring political decision.
The difference may be invisible this month. Compounding is patient. The tax collector arrives late Redistribution has another structural feature. It usually arrives after ownership has already done some of its work.
Before an AI company pays corporate tax, somebody financed it, owned shares as their value increased. Somebody may have sold or borrowed against some of them.
Somebody may have used the wealth to buy another company. Creditors may already have received interest.
Somebody may pass the remaining assets to children.

Tax systems can reach many of these events. Capital gains, estates, property and corporate profits can all be taxed.

The sequence still matters. Ownership begins upstream. Much redistribution happens downstream. Broadly owned funds can finance businesses their beneficiaries dislike. Public institutions can become bureaucratic, politicized or captured.mPrivate capital can move faster and fund risks governments avoid. Citizens can make terrible decisions in their capacity as owners with the same enthusiasm they sometimes bring to other collective decisions.

There is no ownership structure that removes conflict.
Different structures move the fight to different places: taxation, fund governance, eligibility, contribution rules, transfer rights, inheritance and creditor protections.
The fight does not disappear. It changes address.

The civilization of beneficiaries

There is a future in which all of this works reasonably well.
Machines become extremely productive. Goods become cheaper. Governments collect enough revenue to provide healthcare, housing support and a substantial basic income.
Most people work less. Poverty falls. The population is materially comfortable. And ordinary people own almost none of the productive capital.

This is not the usual dystopia.

Nobody is starving beneath a giant holographic advertisement for the company that replaced him.
There may be excellent food. Good apartments. Cheap transport. Personal AI systems that know exactly how someone likes his coffee, music and pornography. Plenty of leisure. A person could live better than most kings in history and still possess almost no claim on the productive system supporting him.
He would be a beneficiary. That word is doing a lot of work. A beneficiary can be treated generously. A beneficiary can also be dependent. Dependence does not automatically produce misery. Human beings spend much of life dependent on families, employers, states, utilities and strangers who keep sewage systems functioning.

The distinction appears in bargaining power.
An owner and a beneficiary have different relationships with the institution sending the money. One has a claim attached to an asset. The other has a claim attached to a rule.
Rules can be excellent. They can also change.

Another possibility

The alternative is not a nation of miniature venture capitalists voting on model architecture from their kitchens.
Broad ownership could be boring. A child is born. Along with the health card, tax number and the other paperwork announcing that a new person now exists, there is an account. The family does not choose the companies inside it. Nobody hangs the first statement on the refrigerator. Some returns stay invested. Some may eventually become income.
Years pass. The child grows up knowing the account mostly as another line on a screen. At eighteen, nothing cinematic happens. There is simply an asset that was not financed by a first job.
An arrangement like that would bring problems of its own. Governance could become remote. Citizenship would become financially valuable in a new way. Immigration debates could acquire another zero on the end. Politicians would fight over whatever institution managed the assets. Wealthier families would still find better ways to accumulate private wealth around the public claim.

Ownership would not abolish hierarchy.

It could change where ordinary people sit inside it. There is a neglected possibility between preserving employment forever and replacing lost wages with transfers forever. People could lose some economic importance as workers and gain some economic importance as owners.
Whether that produces a better society cannot be answered by the word ownership alone.

The mechanism matters because the default does not wait for a philosophical agreement.
The default is yesterday's ownership structure buying tomorrow's productive assets.

The second line

Go back to the phone. The salary deposit arrived on Thursday. A retirement contribution left on Friday. The worker barely noticed. He had groceries to buy.
This small arrangement linked two economic identities.
He worked. Then a piece of what he earned crossed the bridge and began owning things. The amounts were usually tiny. The bridge was unequal. Many people never crossed much of it at all.
It existed.

If machine intelligence reduces the value of labor, the first line on the screen may shrink.
The second becomes more important.
A civilization in which human labor matters less will still have to decide who holds claims on the capital doing more of the work. It may choose deliberately.
More likely, thousands of individually sensible decisions will choose before anyone recognizes a choice was being made. A procurement contract, a pension contribution, a bond issue. Nothing that looks like a constitutional moment.

Then one morning somebody will open the same phone screen and notice that the first line matters much less than it used to.
The second will matter more. And if there is no second line, the machines will still know exactly where to send the money.

Sources

1. Board of Governors of the Federal Reserve System, Distributional Financial Accounts, Q2 2026. Updated September 18, 2026. The top 1 percent held 50.9 percent of U.S. corporate equities and mutual-fund shares; households in the 90th to 99th wealth percentiles held 37.2 percent; the bottom half held 0.6 percent. Together, the top 10 percent held 88.1 percent. Federal Reserve data via FRED
2. CPP Investments, "CPP Investments Net Assets Total $780.7 Billion at Third Quarter Fiscal 2026." February 13, 2026. CPP Investments reported net assets of C$780.7 billion as of December 31, 2025. CPP Investments Q3 Fiscal 2026 results
3. Norges Bank Investment Management, Government Pension Fund Global, Half-Year Report 2026. August 12, 2026. The fund was valued at 22.683 trillion Norwegian kroner as of June 30, 2026. Government Pension Fund Global Half-Year Report 2026
4. Alaska Department of Revenue, Permanent Fund Dividend. The official 2025 Permanent Fund Dividend was $1,000. Alaska Permanent Fund Dividend
5. European Commission, "AI Factories." The Commission states that EU, member-state and associated-country investment in supercomputing infrastructure and AI Factories will reach €10 billion over 2021 to 2027 through the EuroHPC Joint Undertaking. European Commission AI Factories
6. European Commission, "EU launches AI Gigafactories call to boost Europe's computing capacity and unlock more than €30 billion in investment." July 30, 2026. The call covers up to seven AI Gigafactories, supported by up to €10 billion in EU and national funding and expected to mobilize at least €20 billion in private investment. European Commission AI Gigafactories announcement
7. Reuters, "SoftBank raises $11.1 billion in world's biggest high-yield corporate bond sale." September 24, 2026. The bond issuance was intended to finance SoftBank's AI investments, including its OpenAI commitment and other technology investments and acquisitions. Reuters on SoftBank's $11.1 billion bond sale
8.  Reuters, "AI borrowers face tough sell in risky corners of US credit market." September 30, 2026. Reuters reported rapid growth in AI-related leveraged borrowing and increasing investor preference for infrastructure businesses with tangible assets and contracted revenue.Reuters on AI borrowing and credit markets