A server can move to space. A patent can move to Ireland. A profit can move through a spreadsheet. Land cannot move.
I want to make the case for a tax before the crisis that makes it unavoidable.
For most of the modern welfare state, public finance has rested on one convenient fact: people work.
Workers earn wages. Governments tax those wages. Employers pay payroll taxes and social-security contributions. Workers spend what remains, generating VAT. Companies employ the workers and pay tax on their profits.
It is an ecosystem built, in a surprisingly literal sense, around the economic relevance of the human being.
Now take the AI industry at its word.
Suppose artificial intelligence does not merely make a programmer or accountant 20 percent more productive. Suppose software performs a growing share of cognitive work, robots eventually perform more physical work, and the amount of human labor required to produce a unit of GDP falls substantially.
Production may rise. Corporate valuations may rise. The owners of models, data centers and robots may become extraordinarily rich. Yet wages can stagnate, employment can shrink, and the tax base supporting the state can erode.
GDP grows while the fiscal foundation underneath it weakens.
Citrini Research and Alap Shah’s 2026 scenario The 2028 Global Intelligence Crisis used a memorable name for this: Ghost GDP. I later carried the term into my report Poland Facing AGI.
The question is simple:
What happens to a state that taxes work when less work is needed?
The tax system is already looking backward
This is no longer only an AGI thought experiment.
The World Bank’s 2026 report Navigating the Age of AI: Implications for Poland’s Economy models a much milder transition. It does not assume the end of human work. Yet even under those restrained assumptions, it identifies the same structural fault line: AI shifts income from labor toward capital, weakening the tax bases built around wages and employment. Revenue from other sources does not automatically replace what is lost.
The significance is not a decimal point in a ten-year forecast. It is that a mainstream institution now recognizes the fiscal system’s dependence on human labor as a policy problem. The bases for income tax and social contributions narrow precisely when governments may need more money to help people through the transition.
Bill Gates has arrived at the same diagnosis. In his recent essay on the turbulence ahead, he argues that governments should “tax AI tokens and robots”. Human workers are taxed continuously, while a machine replacing them is normally treated as a deductible investment. Gates wants to slow displacement and finance retraining and social protection.
Sam Altman went further in 2021. In Moore’s Law for Everything, he predicted that power would shift from labor to capital and proposed taxing the two assets he expected to hold much of the value in that world: companies and land. His illustrative American Equity Fund would receive 2.5 percent of large companies’ value in shares each year and a 2.5 percent levy on privately held land, then distribute the proceeds to citizens.
The builders of the technology are telling us that the relationship between work, income and ownership may change dramatically. But our tax system still behaves as if the payslip will remain the natural center of economic life forever.
It will not.
Do not tax abundance back into scarcity
The first instinct is to tax the machine.
Tax every robot. Tax every GPU. Tax every token produced by a model. The attraction is obvious: if machines replace the taxable worker, make the machine inherit the worker’s tax bill.
But this creates a contradiction. We hope AI will make medical advice, education, engineering and scientific discovery cheaper. Then we propose to increase the price of every unit of machine intelligence. A token tax may raise revenue and slow displacement, which is precisely why Gates supports it. It may also slow beneficial adoption, push computation toward local models that are harder to observe, and invite endless arguments over which computation counts as taxable AI.
A turnover tax has a similar problem. Revenue is easier to observe than profit, but revenue is not rent. A software company with a 60 percent margin and a supermarket with a 2 percent margin can report the same sales. The same turnover tax barely touches one and can consume most of the other’s profit. IMF research on turnover-based taxation highlights exactly this blindness to margins, together with cascading costs across supply chains.
We should tax pollution because we want less pollution. We should tax monopoly rents because we do not need more monopoly rents. But when a technology can make society more productive, the default should not be to tax every additional unit of it.
My starting rule is different:
Tax what cannot leave. Tax what cannot be produced in response to demand. Tax rents before you tax production.
Which brings us to land.
What a land value tax actually taxes
A land value tax is not the property tax most people imagine.
It does not tax the apartment block, factory, shop or house standing on a parcel. It taxes the underlying value of the site as if the improvements did not exist.
Build another floor: the tax does not rise because of the new floor.
Renovate a ruined building: the tax does not punish the renovation.
Replace a car park with homes: the tax does not penalize the construction.
Leave valuable urban land empty: you still owe the same tax as the neighbor who uses an equally valuable site well.
The distinction is the entire point. A conventional property tax says, “Build something useful and your bill increases.” A land value tax says, “This location is valuable whether you use it or waste it.”
Consider two identical empty plots. One is in central Warsaw beside a metro station, offices, restaurants and thousands of potential customers. The other is in a shrinking town, far from jobs and infrastructure. The owner of the Warsaw plot did not create the metro, the legal system, the surrounding businesses, the density or the millions of people who make that location valuable. Society did.
The building is largely private value. The location is largely social value.
LVT returns part of that social value to the society that created it.
The economic case is unusually clean. Land is fixed in supply. Tax work and people may work less. Tax investment and some investments stop making sense. Tax a product and consumers buy less of it. Tax land and the country does not become smaller.
An IMF working paper on the equity and efficiency of LVT starts from this standard result: a land tax does not distort the supply of its base. Its authors also find that a revenue-neutral shift from income and corporate taxes toward land can improve efficiency and reduce the net burden on low- and middle-income households, depending on how the revenue is returned.
This is why the idea survives ideological fashion.
Henry George wanted the public to collect land rent and abolish other taxes. Milton Friedman wanted a much smaller state, yet in a 1978 interview he still called the tax on unimproved land value “the least bad tax”. They disagreed about the size and purpose of government. They agreed that if government must raise money, it should begin with the base least able to disappear.
There is also a reason land keeps returning to the center of the housing debate. A landmark study of 14 advanced economies found that rising land prices explain about 80 percent of the postwar increase in house prices. We often talk as if the housing crisis were only a shortage of bricks, timber and labor. Much of what people are bidding against each other for is access to a finite location.
AI and robotics may make buildings dramatically cheaper to construct. They cannot reproduce what makes a particular location valuable: access to jobs, transport, customers, schools, culture and other people. If construction gets cheaper while the supply of well-connected locations remains limited, more of the gain can be absorbed into land prices.
You can automate construction. You cannot automate proximity.
Why AI makes the old argument stronger
The defining feature of the digital economy is mobility.
Software can be built in California, owned through a subsidiary elsewhere, executed in a third country and sold in Poland. Intellectual property can cross a border with a signature. Accounting profit can be relocated through royalties, debt and transfer prices. Compute can move between clouds.
It can now even move above the clouds. In November 2025, Starcloud put an NVIDIA H100 into orbit and later ran and trained small AI models in space.
Does an orbital data center destroy the case for LVT?
No. It destroys the claim that LVT is a comprehensive tax on AI hardware.
The server may go to space. Warsaw does not.
If a company sells its central Warsaw parcel and moves to a cheaper location, it will pay less LVT. But the valuable parcel has not escaped. Its buyer inherits the bill. That is not avoidance. It is the mechanism working: expensive land moves toward whoever can make the best use of the location.
As more tax bases become mobile, an immobile base becomes more valuable. A 2025 study called Funding Government in the Age of AI compared nine possible revenue strategies across different automation scenarios. Land value taxes and consumption taxes performed best overall on feasibility, incidence, resilience and incentives. The World Bank subsequently highlighted that finding in its Poland report.
LVT will not capture every rent in an AI economy. Model ownership, energy, minerals, spectrum, grid connections and orbital infrastructure can all become bottlenecks. Some will require competition policy, resource charges, public equity or other forms of rent taxation.
That limitation should be stated openly. LVT is not magic. It is the strongest anchor for a system in which much else can move, hide or multiply.
The case for a single-tax horizon
Henry George’s “single tax” is often treated as a historical curiosity. I think it is more useful as a direction of travel.
The principle is radical but simple: stop taxing the creation of wealth and finance the common realm by collecting value created by the common realm.
That does not mean abolishing every other tax next Monday. Nobody can honestly guarantee that land rent, at politically and economically sustainable rates, will immediately finance every pension, hospital, school, army and infrastructure project of a modern European state. An ambitious macroeconomic model published by CEPR explored a system that raised 55 percent of revenue from land and the rest from consumption while abolishing income taxes. That is evidence of the scale worth investigating, not a ready-made budget.
The single-tax horizon means something more disciplined:
Every additional złoty collected from land should buy a złoty of relief from work, construction or productive investment.
Raise LVT and cut payroll taxes. Raise LVT and remove taxes on buildings. Raise LVT and increase the amount people can earn before paying income tax. Keep moving in that direction until the revenue potential, evidence or democratic choice tells us to stop.
If the destination is never a mathematically pure single tax, the system will still be radically better for having followed the compass.
Poland taxes the wrong thing
Poland is an ideal place to begin because its current system is close to the opposite of LVT.
Recurrent taxes on land and buildings are still based largely on area. The system can distinguish residential from business use and maintain elaborate categories of structures, then argue over rates and definitions. But it does a poor job of distinguishing a square meter in central Warsaw from a square meter in a place losing people and opportunity.
We tax the measurable surface while missing much of the economic value.
This is not only a Georgist complaint. The OECD’s 2025 housing review of Poland recommends building mass-valuation capacity and moving toward recurrent taxation based on the market value of residential property and land. The IMF’s latest Article IV report likewise tells Poland to invest in an appraisal system that can replace space-based property taxation with value-based taxation.
Neither institution is formally proposing a pure LVT. Their value-based property proposals include buildings. But the administrative prerequisite is the same: Poland must learn to value locations consistently, update those values and let citizens challenge them.
That is not a reason to wait. It is the first stage of the reform.
A Polish road to LVT
Poland should treat LVT as fiscal infrastructure, not as one more charge placed on top of everything else.
The transition could follow five rules.
1. Build the valuation layer first.
Combine transaction prices, cadastral records, planning rights, access to infrastructure and geographic data into a national mass-appraisal system. Publish the estimated site value of every parcel, the inputs used and the model’s confidence range. Owners must have a simple appeal route, and independent auditors must test whether errors systematically favor particular regions or groups.
Modern states already value millions of heterogeneous properties for lending, insurance, acquisition and taxation. The valuation will never be metaphysically perfect. Neither is taxable corporate profit. The standard should be transparent, contestable and consistently better than taxing square meters almost blindly.
2. Split land from improvements.
Every assessment should show two numbers: the estimated value of the site and the value of what has been built on it. As the land rate rises, the recurring tax on buildings should fall toward zero. The reform should reward the apartment block, factory and renovated tenement rather than treat them as a larger tax target.
3. Make the shift revenue-neutral at each stage.
The first phase should not be sold as a way to collect more money. It should change where the money comes from. Municipal LVT revenue should be matched by explicit reductions in taxes on labor and productive investment through a revised national-local settlement.
The promise must be visible on the payslip and credible in law:
One złoty more from land. One złoty less from work.
4. Phase it in slowly and announce the path in advance.
A permanent land tax is capitalized into lower land prices. That lowers the upfront purchase price for future buyers, but it also creates a real one-time loss for current owners who purchased under the old rules. Pretending otherwise would make the reform dishonest.
Rates should therefore rise over a decade according to a published schedule, while labor and building taxes fall alongside them. Stable expectations matter more than a dramatic first-year rate.
5. Defer, do not exempt, when income is the problem.
The strongest human objection is the retired person who owns a home on now-valuable land but has little cash income. The answer should not be forced sale, nor a giant exemption that destroys the base. Eligible owner-occupiers should be able to defer the liability, with a transparent interest rate, until the property is sold or transferred. The tax becomes a claim against the asset rather than a demand against the monthly pension.
The OECD documents this approach in countries including Denmark and Ireland. Liquidity can be protected without allowing valuable land to leave the tax system forever.
Poland would not be testing the underlying concept on another planet. Estonia taxes land without taxing the buildings and other improvements on it, using assessed land values and local rates. Its implementation is imperfect and its revenue is modest, but it proves that separating the site from the structure is an administrative choice, not a fantasy.
LVT also needs planning reform. A tax cannot produce apartments on a site where the law forbids apartments. Cities must allow denser development around transport and infrastructure, then collect part of the location value that public decisions create. Permission to build and taxation of land should work as one system.
The objections we should welcome
The case becomes stronger when its costs are admitted.
“Owners will pass it on to tenants.”
A landlord will always try to charge the highest rent the market permits, with or without LVT. Because a land tax does not reduce the supply of land, its long-run burden is capitalized mainly into the site’s price rather than creating less land. But tenants will not be protected by theory alone in every local market. Housing supply, competition and planning still matter.
“Companies will move.”
A registered headquarters can move. A parcel cannot. If a business genuinely no longer benefits from an expensive location and moves, that is efficient. Another owner takes the site and the liability. What disappears is the incentive to warehouse high-value land while waiting for everyone else’s investment to raise its price.
“Valuation will become political.”
It can. So can every tax base. The defense is open data, reproducible models, frequent revaluation, independent oversight and an inexpensive appeal process. Hiding the model would be fatal. Publishing it turns every owner, bank and researcher into a potential auditor.
“It will crash land prices.”
It will reduce them relative to a world without the tax because buyers will subtract future liabilities from what they are willing to pay. This is not an accidental side effect. It is how we move part of the price of access from a private windfall to public revenue. The transition must be gradual because the loss to existing owners is real.
“Land cannot finance the whole state.”
Perhaps not. The honest response is to measure the tax base, publish scenarios and find the limit in practice. The dishonest response is to use uncertainty as an excuse to keep taxing work first. A single-tax horizon is not a promise that arithmetic will obey ideology. It is a commitment to exhaust the least destructive base before reaching again for the payslip.
Who owns the future?
And then there is the truly strange possibility: the AI companies may be right.
Suppose intelligence becomes abundant machine capital. Suppose systems can substitute for humans across most economically valuable tasks. In that world, the central divide is no longer between higher-paid and lower-paid workers.
It is between people who own the productive capital and people who do not.
Debating whether the marginal income-tax rate should move two points up or down becomes almost comical if labor receives a radically smaller share of national income. Ownership dominates taxation. Who owns the models matters. Who owns the energy matters. Who owns the scarce complements to intelligence matters.
And land remains the universal complement. Every person still needs somewhere to live. Every physical business needs a location. Every city, port, mine, transmission line and terrestrial data center occupies a finite piece of the world. When technology makes many produced goods cheaper, the things it cannot reproduce become more important, not less.
LVT alone will not democratize ownership of AI. It will not solve monopoly, govern algorithms or build sovereign compute. But it can prevent one of the oldest forms of rent from swallowing part of every new productivity gain. It can finance the transition without punishing the people who continue to work or the companies that continue to build.
If the strongest AI predictions fail, land value tax remains one of the least damaging ways to fund a state, encourage construction and return socially created value to society.
If those predictions come true, it becomes part of fiscal survival.
That is an unusually good bet.
Stop taxing what we need more of: work, homes, enterprise and useful technology.
Start with what cannot hide, cannot leave and cannot be made by a machine.
The future may produce unlimited intelligence. It will not produce another Warsaw.
If intelligence becomes abundant, tax scarcity.