Wealth depends upon controlling what’s scarce. The scarce thing was once a person. Then it was a wage. Soon it will be a machine that does not need a wage at all. Follow the substitution and the whole shape of the modern economy comes into focus—not as a story of progress interrupted by occasional cruelty, but as a single continuous business, conducted by recognisably the same kind of people, using recognisably the same technique, on whatever happened to be the bottleneck of the moment.
Start with the hinge, because the hinge is the argument in miniature and it is dated.
1833 AND 1834: THE SAME PARLIAMENT, TWO MOVES
In August 1833, Parliament voted £20 million to compensate slave-owners. Something in the region of forty percent of annual government expenditure, handed to the people who had held human beings as capital, on the principle that the loss of that capital should not inconvenience them.
Fourteen months later, in August 1834, the same Parliament passed the Poor Law Amendment Act. The new law abolished outdoor relief for the able-bodied poor and imposed the workhouse test. Its governing principle had a name—less eligibility—and the name was honest. The condition of a person on relief had to be worse, deliberately and materially worse, than the condition of the lowest-paid independent labourer. Relief could not compete with the worst wage, because if it could, the worst wage would have to rise.
Read the two acts together, because that is how they were written. First the state paid the owners for the loss of their human capital. Then it legislated that the labouring poor must be kept beneath the worst wage so they would take it. Compensation above, discipline below. Eighteen months apart, same building, largely the same men.
That is the transition in its entirety. Leverage by ownership became leverage by wage. The enslaved person had been leverage because he was property: his labour was pure surplus, and his existence set a floor under everyone else. After abolition, the same class needed the same leverage without the property, because property is expensive to maintain and was becoming awkward to defend in public. The reserve army is cheaper. You do not have to house, feed, or insure the unemployed. You only have to let them exist as a warning.
A wageless man is leverage through wage suppression. He is not outside the economy looking in. He is an input to it.
KALECKI NAMED THE MECHANISM IN 1943
None of this is a conspiracy theory. It is a published economic argument, made by a serious economist in a mainstream journal during the war, and it was ignored for exactly as long as it was inconvenient.
Michal Kalecki’s “Political Aspects of Full Employment” asked the question his colleagues mostly avoided: if a government can sustain full employment through spending, why do business leaders oppose it so reliably? His answer was not that they were stupid. It was that they understood what full employment does. Under a regime of permanent full employment, he wrote, “the sack” would cease to play its role as a disciplinary measure. Full employment removes the fear that keeps wages down and workers compliant. Business opposition to it, therefore, was never really about economics. It was about power. And the doctrine of “sound finance,” he argued, had a social function: it made the level of employment depend on the “state of confidence”—which is to say, on the preferences of the people who own things.
He was writing in 1943. It took about thirty years for the lesson to be applied properly.
The instrument that applied it was the Phillips curve, the observed trade-off between unemployment and wage growth, and then its rebranding as the natural rate of unemployment by Milton Friedman in 1968. Once you accept that there is a floor beneath which unemployment cannot safely fall without wages accelerating, you have given the reserve army a scientific charter. It stops being a disciplining device and becomes a technical necessity. The NAIRU—the non-accelerating inflation rate of unemployment—becomes the number the central bank steers toward, and every decimal point of it is a wage somebody does not get.
That is the modern version of the whip. The whip is gone. The mechanism is a target, published quarterly, defended by people in nice suits who genuinely believe they are doing macroeconomics.
THE FLOOR WITH A HOLE IN IT
The racial buffer did not retire when the plantation did. It got a desk in the welfare office.
Ira Katznelson’s When Affirmative Action Was White lays out the mechanics. The New Deal’s two signature labour protections, the Fair Labor Standards Act and the Social Security Act, excluded agricultural and domestic workers. That was not an oversight. It was the price of Southern Democratic votes, and it was paid in Black labour, because the majority of Black workers in the South were agricultural or domestic. The floor of the new welfare state was built with a hole in it, and the hole was exactly where Black workers stood.
So the buffer was written into the architecture. The white worker got a floor. The floor was the Black worker. And the floor was defended, because the alternative was to look up.
The same pattern held in the private sector. Taft-Hartley in 1947 narrowed the strike. The unions that survived were largely the ones that had already drawn the colour line. And when Reagan broke the air traffic controllers in 1981, firing eleven thousand of them and hiring replacements, he did not merely win a dispute. He reset the norm. After PATCO, the strike stopped being a weapon and the reserve army got its old job back.
Every one of those moves was a labour policy. Every one of them suppressed a wage. Not one of them mentioned race in the text. All of them used it.
WHAT IS BEING ENCLOSED NOW
Here the argument leaves the archive and starts reading the trend lines, so hold it loosely and check the numbers yourself.
Every previous era of accumulation was built on enclosing something. The enclosures took the commons. The plantation took the body. The factory took the working day. The platform took the taxi medallion, the hotel, and the corner shop, and turned all three into rent.
What is being enclosed now is inference: the capacity to take an input and produce a decision, a design, a diagnosis, a draft, a route, a recommendation. It is the newest thing to be scarce, and it is being bought up while most people are still arguing about whether it works.
Inference has a physical substrate, because it has to. It runs on compute, which means chips. It runs on power, which means electricity. It runs on data, which means the recorded residue of everything people have already done. And it runs on land and water and cooling, because a data centre is a building before it is a model.
Every layer of that substrate is concentrating. The leading-edge fabs are a handful of firms. The lithography that makes them possible is one firm. The design tools are three. The training runs that set the frontier belong to a small number of companies with the balance sheets to pay for them, and the compute divide in AI research is already documented: a shrinking set of institutions can compete at the top at all.
The energy layer is moving fastest and gets discussed least. Lawrence Berkeley National Laboratory put US data centre consumption at 176 TWh in 2023, about 4.4 percent of national electricity, and projected 325 to 580 TWh by 2028. The IEA’s 2025 report put global data centre demand at roughly 415 TWh in 2024 and projected it to more than double to about 945 TWh by 2030. Those are not forecasts about software. They are forecasts about power purchase agreements, transmission lines, water rights, and land.
So who controls the energy and the raw materials is the landlord of inference. That is not a metaphor, and it is worth being precise about why. A landlord is someone who owns the thing you cannot route around. You can write better code. You cannot write a better transmission line. You can improve the model. You cannot improve the supply of EUV lithography machines. The bottleneck is physical, the physical bottleneck is ownable, and the rent on it gets collected the way rent has always been collected: quietly, monthly, from people who have no alternative.
THE ROBOT DOES NOT ASK FOR A WAGE
Now the part that changes the class relation rather than just relocating it, and the part most of the commentary walks straight past.
Slavery was crude because it required owning the person, and owning a person is a maintenance problem. Wage labour was refined because it required owning only the means of production and renting the person’s time, with the reserve army setting the price. In both cases the person doing the work had to be managed, and management is friction. It requires supervisors, ideology, and the constant low-grade work of keeping people in their lanes.
Inference takes the person out of the transaction. Not out of the world—out of the transaction. If a model does the work, there is no wage to suppress, because there is no wage. The worker’s bargaining position is not disciplined by a reserve army. It is bypassed. You cannot strike against a machine you do not operate, owned by a company you do not work for, running on power you cannot afford to buy.
Who controls the inference earns from the labour of the robot. That is the new extraction, and it has one property the old ones never had: it does not require the cooperation of the extracted. The reserve army worked because the unemployed still had to be fed and the employed still had to be frightened. A model needs neither.
Which is why the ownership question is the only question, and why everything else being debated is downstream of it. Not will the models be good. Not will the models be safe. Who owns the robot owns the future, and that will not be settled by the technology. It will be settled by who holds title, who wrote the rules, and who has the class shield. The same three things that settled the last two rounds.
THE QUEUE IS STILL BEING KEPT
So where does that leave the buffer? Still working, because the queue still needs ordering.
The category is still doing what it was built to do: telling the people being displaced whose fault it is. In the 1670s the answer was the Black bond-servant. In the 1980s it was the welfare queen. In the 2020s it is whoever the algorithm flags, whoever the migration statistic names, whoever the culture war has queued up for the week. The room has changed its furniture. The room is still the variable, the occupants are still the accident, and the performance is still the mechanism.
And the landlord of inference is not a new kind of person. He is the same kind of person the courtiers were: somebody who got close to the thing everybody needs and made himself the toll booth. Bentinck had the king’s ear. The Gladstones had the plantation. The Heywoods had the bank. The next one will have the substation, the fab, and the weights. The costume changes. The fee is still payable by anyone who can afford it, which is exactly why the category is still needed: to keep the queue orderly while the fee goes up.
Yesterday they owned the worker. Today they own the wage. Tomorrow they own the machine. The only thing that has ever changed is the name on the deed.
References
Arrieta-Ibarra, I., Goff, L., Jiménez-Hernández, D., Lanier, J., & Weyl, E. G. (2018). Should we treat data as labor? Moving beyond “free.” AEA Papers and Proceedings, 108, 38–42.
Christophers, B. (2020). Rentier capitalism: Who owns the economy, and who pays for it? Verso.
Friedman, M. (1968). The role of monetary policy. American Economic Review, 58(1), 1–17.
International Energy Agency. (2025). Energy and AI. IEA. https://www.iea.org/reports/energy-and-ai
Kalecki, M. (1943). Political aspects of full employment. Political Quarterly, 14(4), 322–330.
Katznelson, I. (2005). When affirmative action was white: An untold history of racial inequality in twentieth-century America. W. W. Norton.
Lawrence Berkeley National Laboratory. (2024). 2024 United States data center energy usage report. LBNL.
Marx, K. (1867). Capital: A critique of political economy (Vol. 1), Chapter 25: The general law of capitalist accumulation.
Phillips, A. W. (1958). The relation between unemployment and the rate of change of money wage rates in the United Kingdom, 1861–1957. Economica, 25(100), 283–299.
Poor Law Amendment Act 1834, 4 & 5 Will. 4 c. 76 (UK).
Slavery Abolition Act 1833, 3 & 4 Will. 4 c. 73 (UK).