Will capitalism survive AI?

I have documented elsewhere why AI's organizational revolution never happened. That is not the subject of this essay. The subject is what that non-revolution reveals about a deeper movement: if AI is failing to transform organizations, it is because it is striking at something more structural than them. It is striking at the pillars of the economic system in which those organizations exist.

The empirical diagnosis is laid out in The revolution that never happened. Three years after ChatGPT, the macroeconomic impact is indistinguishable from statistical noise. CEOs announcing "AI-driven" restructurings are dressing up something else. Individuals, meanwhile, are shifting.

This essay takes the other scale. Not the organization, the system. If AI is not transforming companies but is transforming their members, it is not touching the layer of organization. It is touching the layer beneath. The rules of the game. The structural pillars that economists have been describing for two centuries with striking consistency.

Will capitalism survive AI? No one has the answer. But you can no longer write as if nothing has moved.

Three pillars, two centuries of consensus

Before looking at what is cracking, you have to know what is holding. Capitalism, as an economic system, rests on three pillars that economists, liberals and critics alike, describe in roughly the same way since Smith.

The first pillar is the private ownership of the means of production. In 1776, Adam Smith observed in The Wealth of Nations that the wealth of modern societies arises when tools, machines, land, and capital belong to private actors who put them to work.^[1] Ownership is what turns a resource into capital.

The second pillar is wage labour. Karl Marx, in Capital, describes it as the central condition of the system: those who do not own the means of production rent out their labour power to those who do, in exchange for a wage.^[2] Max Weber later showed how that relationship rationalises itself, becomes bureaucratic, regulated, measured by the clock.^[3]

The third pillar is the capture of capital. Joseph Schumpeter put it plainly: the value produced by economic activity flows back to those who invested, in the form of profit, dividend, surplus value.^[4] That capture is what allows capital to accumulate, reinvest, concentrate.

Three pillars. No one disputes the description. Liberal economists and their critics have spent two centuries arguing whether the system is just, sustainable, desirable. About what it is, they agree. That is not the question here.

The question here is that today, in 2026, each of these three pillars is fracturing under the effect of artificial intelligence. Each for different reasons. Each at a different pace. But all three at the same time.

First pillar, ownership distributing itself

You were sold that capital meant the factory, the patent, the machine. Look at your desk. Capital now fits on 600 grams of silicon and a monthly subscription.

The first pillar that is cracking is the ownership of the means of production.

I am writing these lines from a Mac mini sitting on a shelf, at home. It runs a Mistral model locally, hosts the Claude skills I have built, and runs the agents I delegate the tasks I no longer want to do myself. For the cost of half a monthly salary, I have at home a productive infrastructure that would have cost several hundred thousand euros five years ago, and would have required a team of five to operate.

I am not an outlier. Among the hundreds of professionals I work with in training rooms, in 2025, I watched participants build, in a few hours, robots that automate tasks that used to occupy entire teams. A robot that fills out APIDAE entries for a regional tourist information system, where three agents used to spend their weeks on data input. A robot that transcribes meeting audio and produces a clean summary, ready to circulate. A robot that processes expense reports and pre-validates accounting entries. These participants are not engineers. They are communications managers, tourist office directors, team leads, administrative officers.

"In that rude state of society in which there is no division of labour, in which exchanges are seldom made, every man provides everything for himself. As the division of labour advances, a stock of materials and tools becomes indispensable."^[1]

Smith described the necessary concentration of means. AI runs the movement in reverse. Microsoft has quantified the phenomenon at scale: 78% of AI users at work bring their own tools, without waiting for their company to deploy anything.^[5] This is the BYOA, Bring Your Own AI, a silent vote of the individual against the organization. I note here what that displacement does to the pillar.

The precedent exists. Fifteen years ago, BYOD, Bring Your Own Device, saw employees bring their smartphones and laptops into work. IT departments panicked, then resigned themselves. Ownership of the work phone dissolved into the personal property of the worker. But the phone was a communication tool. AI is a production tool. When an employee brings their own AI, they bring their own productive force. And they amplify it at home, in the evening, accumulating prompts, skills, routines, a body of know-how that does not belong to their employer.

Analyst Matt Warren published in April 2026 a formula that captures the move: "Bring your own agent is a form of capital formation at the edge of the company."^[6] The employee, often unknowingly, accumulates a personal stock of productive capital at the periphery of the organization that pays them. When they leave, that capital leaves with them.

The pillar is not falling. It is shifting.

I am not saying concentrated capital is dead. I still depend on OpenAI, Anthropic, Google for the frontier capabilities my Mac mini cannot reach. The frontier model is still rented. But the trajectory is clear. In January 2025, the Chinese startup DeepSeek released a model, R1, that matches the performance of GPT-o1 on reasoning benchmarks, for $5.576M in "official training," against several billion for the comparable models.^[7][8] The figure is understated, the company itself acknowledges that it does not include prior R&D, and independent estimates put the total cost well above.^[9] What matters is that an order of magnitude has shifted.

Mistral is doing the same on the European side with its Mixtral and Mistral Large models, released in open weight, freely downloadable, runnable on consumer hardware.^[10] The gap between proprietary models and open ones is closing faster than most analysts forecast.

If the trajectory holds, in two or three years, I will not need to rent anything. My Mac mini, my internet connection, and an open model that competes with the best proprietary ones. The means of production will be at home. Not at the employer's. Not at Microsoft's. At home.

Worth flagging. This displacement touches the qualified tertiary, the work made of text, code, image, reasoning. It does not touch heavy industry: an automotive assembly line, a refinery, a blast furnace remain means of production that no individual employee can internalise, and they will stay that way. The pillar is not collapsing everywhere. It is moving where the productive tool fits in a backpack.

Smith never imagined that the means of production could sit on a personal shelf.

Second pillar, wage labour emptying itself

Wage labour is not a natural condition of the economy. It is a historical arrangement. It lasts as long as there is an advantage in renting yourself out.

The second pillar that is cracking is wage labour.

I have several WhatsApp groups with professionals I have met in training, on missions, at conferences, over the past five years. Around 80% of them are now self-employed. That ratio made no sense three years ago. It became unremarkable in 2026.

These people did not go independent because they could not find work. Most had comfortable positions. They went independent because they understood one simple thing: with AI, they produce in two days what used to take two weeks, and there is no longer any structural reason to hand that differential back to an employer. They keep the differential. They redirect it to free time, to personal projects, to a life. Not to an organization.

There is a second reason, which keeps coming up in those conversations. When you produce in two days what used to take two weeks, the bottleneck is no longer production. It is the chain of decision. The quote waiting on a signature. The strategy note waiting on a committee. The arbitration waiting on the next meeting. Generative AI does not break those cycles. It makes them visible. They become unbearable for someone who can now move at a different pace. Going independent is not just about keeping the productivity gain. It is about being able to execute it without going through four levels of hierarchy.

"As soon as labour in its immediate form has ceased to be the great well-spring of wealth, labour time ceases and must cease to be its measure. The mass of workers must themselves appropriate their own surplus labour. Capital itself is the moving contradiction, in that it presses to reduce labour time to a minimum, while it posits labour time, on the other side, as the sole measure and source of wealth."^[11]

Marx named this phenomenon before it was technically possible. In the Fragment on Machines of the Grundrisse, in 1858, he describes a future moment when wealth ceases to rest on individual labour time, because the machine produces more than human labour ever could. He called that "a miserable foundation" compared to what large industry made possible. The description is analytical: it traces the mechanics of a system that empties itself of its own measure of value.

But Marx thought of the machine as the property of capital, and of any overcoming as a collective and revolutionary movement. He did not anticipate that it would be the individual who, equipped with their own productive tool, would put the pillar under stress against their own company, by staying in their seat or by leaving. What Marx located at the scale of large industry, generative AI is now operating at the scale of qualified tertiary work, in a few years, without social revolution, by a silent, contractual, individual dissolution.

Wage labour empties itself from both ends.

It is not just the independents. It is also those who stay salaried. A colleague runs a team of about ten people. Six months ago, he replaced most of their tasks with AI agents he configured himself. The team shrank through natural attrition without backfill. No one further up the chain asked him to rehire. No one saw the need. The service is running. With fewer people.

The pillar empties from the top: those with the rarest skills go independent, because the economic rationale of subordination collapses once the productivity differential can be kept for oneself. Put differently, they go independent because they can now produce alone the entire vertical chain of tasks they used to be embedded in. And it empties from the bottom: those who handled the repetitive tasks are simply not replaced. In the middle, a shrinking band reorients itself, retrains, or waits.

"The rational capitalistic organization of free labour, and only this, was born here. The presupposition is the separation of household and business, which entirely dominates modern economic life."^[12]

Weber described the rationalization of labour as a process of separation, measurement, bureaucratic framing. With AI, that separation becomes reversible. The household becomes a place of production again. The business is no longer the only place where value is made. The capitalist rationality of measured working time runs into a force that overflows it from below: the individual who produces at home what the organization no longer knows how to measure.

A methodological note. This dissolution of wage labour does not touch all sectors at the same speed. It is dramatic in the qualified tertiary, slower in trades where the body is necessary, almost absent in care, in-person education, physical craft. That uneven pace is central. It is best read through the five adoption levels of AI, which I will treat in a dedicated essay.

When my interlocutors go independent, it is not only an economic calculation. It is the expression of a will they had been carrying for a long time, without being able to act on it. No boss. No employees. No hierarchy. AI made that will workable. The general mechanism, perceived capability before desire, is unfolded in Where there's a way, there's a will.

Third pillar, value lodging itself elsewhere

If ownership is distributing and wage labour is emptying, where is the value going? Look in the statistics. You will not find it.

The third pillar that is cracking is the capture of capital.

It is the hardest to diagnose, because it shows up at the macroeconomic scale, not in a training room or a WhatsApp group. I have documented in detail the statistical silence of AI in The revolution that never happened. The headline points: Daron Acemoglu, the 2024 Nobel laureate in economics, estimates the AI-driven gain in total factor productivity at 0.66% over ten years.^[13] MIT NANDA documents that 95% of generative AI pilots in companies produce no measurable return.^[14] McKinsey reports that only 32% of companies are in the scaling phase, and 7% in full deployment.^[15]

In 2025, the world invested $581.7 billion in AI, up 130% year on year, according to Stanford's AI Index.^[16] The impact on GDP is statistically indistinguishable from zero. The novelty is not the figure. The novelty is that this disappearance of value from the books is not a measurement accident. It is a structural consequence of the displacement of the first two pillars.

"The process of creative destruction is the essential fact about capitalism. It is what capitalism consists in and what every capitalist concern has got to live in."^[17]

Schumpeter theorised a mechanism by which capitalism erodes itself: creative destruction wears down the rents of dominant players. But he assumed value would always emerge somewhere, captured by new entrants, by new capitals. Today, the value AI produces does not seem to lodge anywhere in the official accounts. Not in the profits of adopting companies. Not in the wages of workers. Not even clearly in the revenues of AI providers, whose stock-market valuations are an order of magnitude above their actual revenues.

If the first two pillars shift, the third has nothing left to capture.

If the means of production individualise, and if wage labour dissolves, the value produced no longer flows back into the classical capture circuits. It stays where it is produced. With the individual. In their free time, in their personal projects, in their independence. It takes the form of time saved, parallel projects, new autonomy, things that do not appear in GDP and do not show up in corporate accounts.

It is not that capital has stopped capturing. It is that capital captures elsewhere, otherwise, or not at all. And no one yet has the statistical tools to measure what is happening in that elsewhere.

Three fractures, one movement

Three pillars. Three fractures. The private ownership of the means of production distributing itself in private rooms and home offices. Wage labour emptying from both ends. The capture of capital finding less and less value to capture in the official statistics.

I am not writing that capitalism is dead. I have no intention of burying a system that has survived two world wars, the Great Depression, the oil shock, the collapse of the Soviet bloc, the internet, and 2008. Capitalism is probably more resilient than its commentators tend to believe.

But you can no longer write as if nothing has moved. The three pillars that economists have been describing for two centuries have not fallen. They are also not intact. They are fractured at three different points, by three different mechanisms, in the same technological movement. And that movement is not just another productivity wave. It touches structure, not conjuncture.

Will the system hold? Will it find, as it always has, counter-tendencies to absorb the shock and rebuild its pillars? Or will it tip into something else we do not yet have a name for? No one can say in 2026. But anyone telling you they know is selling something other than analysis.

One question remains, and it is the one I want to leave with the reader. If the three pillars are wavering, if value no longer lodges where it used to, if I can produce alone, from home, with my own infrastructure, and keep the productivity differential for myself, then something is changing that goes beyond economics. AI lets me bypass the organization. It lets me, in time, bypass wage labour. It already lets me bypass the expert, in widening domains. It lets me do alone what used to require otherness, cooperation, subordination, confrontation.

Jean-Jérôme DANTON