Seven technology companies now account for nearly a third of the S&P 500 index, compared to 12% ten years ago, a rise now driven by artificial intelligence.
Faced with it, a legitimate question quickly emerged: how many jobs will it transform or eliminate? But it masks another, almost absent from the debate: who will benefit from the productivity gains it produces? The first concerns work, the second on the sharing of value. One cannot be treated without the other.
To understand what is happening, let’s remember the robots. For two centuries, accumulated knowledge could only be monetized through a human being: a skill sold for a given time, remunerated by a salary that paid contributions and was spent locally. The wage was the instrument, imperfect but real, which transformed private productivity into widely distributed income. Late 20th century robotization began to break this link: by automating repetitive manual tasks, it shifted income from labor to capital. It has accelerated the servicialization of the economy with its cohort of professions with low added value but also the injunction to move towards the top of the ladder of qualifications: cognitive, design, relational.
AI targets precisely these professions. It automates non-routine intellectual work: controlling, analyzing, writing, translating, advising, towards which robotization had pushed employment. And the asset that captures the gains no longer looks like a robot. This represented an expensive, physical object, attached to a factory, which supported technicians, a supply chain, and manufacturers in many countries. The income remained diffuse. AI is software: almost zero reproduction cost, instantaneous global deployment, produced by a handful of often foreign players, with extreme returns to scale and few local employees. The income does not evaporate: it changes pockets, and much fewer pockets.
This is the real problem in society. Capitalism holds up because the value it creates is redistributed widely enough so that everyone finds their share and accepts the rules. When the gain is concentrated to this extent and more and more, it is no longer just a question of fairness: it is a question of cohesion and legitimacy. A society in which an ever-thinner fraction captures an ever-larger share of the wealth ends up eroding the consent that holds it together. The capture of value is not a moral abstraction but a concrete political risk.
The potential gains from AI are not of a new nature. Productivity remains productivity. On the other hand, the sharing vector is evolving. Robotization had twisted this transmission belt, AI threatens to break it, by automating the segment which remained the basis of salaries.
If the instrument for distributing value changes, another one must be forged. Our economy shares value between capital, through dividends, and labor, through wages. When the labor channel weakens, we must find a third way that allows us to continue to socialize part of the value created. The societal dividend, this part of profits that a company chooses to voluntarily allocate to the general interest, can play this role. Not out of charity or under constraint, but because a value multiplied by technology must find its way back to the collective in another way to preserve social cohesion.
None of this is inevitable. It all depends on whether AI replaces work or augments it. And if it increases it, who pockets the surplus: the employee, or the supplier who rents the tool? Decoupling is not a law of nature: it is the product of our choices. This is why we must ask the question now, alongside that of employment, before it is resolved without us when tech investors demand a return on their colossal investments. The temptation to capture an increasing portion of the value will then be great. It is up to us, business leaders, to put AI in its rightful place and decide how to share the gains it will allow.
