Wages, Ownership and Power in the Age of Artificial Intelligence
AI multiplies the value of every hour of work. But who will own that multiplier?
In a recent debate in which I took part on the Portuguese television channel NOW, the discussion began with the familiar question: will artificial intelligence create or destroy jobs?
There is evidence supporting both apparently contradictory predictions. In July, Microsoft announced the elimination of around 4,800 jobs. The company stated that these roles were not being directly replaced by artificial intelligence, although it framed the restructuring in the context of sector-wide changes and the growing impact of this technology.
Around the same time, the International Labor Organization estimated that almost 80 million workers across the 11 ASEAN countries are employed in occupations with some degree of potential exposure to generative AI, with no signs, so far, of large-scale job losses.
Both things can be true. AI can eliminate certain roles, create others, and transform all of them virtually. But the total number of jobs, in itself, tells us little about the decisive issue: who will keep the additional wealth being produced?
The Multiplier Has an Owner
The great civilizational shift brought about by artificial intelligence is not limited to the automation of tasks. It lies in multiplying the value produced by each hour of human labor when combined with intelligent machines.
An architect will take hours to complete work that previously took weeks. A programmer can single-handedly develop what once required an entire team. A lawyer can analyze thousands of documents in a fraction of the time. A doctor can use systems capable of instantly synthesizing clinical information that would be impossible for a human to process at the same speed.
But this multiplier does not float freely in the air. It has an owner.
Someone owns the models, the chips, the data centers, the energy, the networks, the data and the platforms through which this new value is produced.
Even open models do not resolve this issue. The software may be open, auditable and adaptable, but the computing power required to run the most powerful systems still depends on scarce and expensive infrastructure. The center of power may simply move down one level: from software to hardware, from model weights to data centers.
The recent case of DeepSeek illustrates this clearly. The company has made the weights of several of its models openly available (the underlying parameters that make them work), yet its founder, Liang Wenfeng, has seen his fortune, as estimated by Bloomberg, rise to around $36 billion, largely because he owns close to 80% of the company. Sam Altman, by contrast, leads OpenAI without owning an equity stake. Despite heading a far better-known and more valuable company, his fortune is nowhere near Liang’s, and the wealth he does possess comes from other investments. The difference is simple: one leads without owning; the other both leads and owns. Opening the model weights did not dissolve ownership; it merely revealed where ownership remains.
This was the fundamental issue I brought to the debate on NOW: the number of jobs does not reveal how wealth will be distributed, and nothing guarantees that higher productivity will translate proportionately into higher wages.
Paulo Dimas, CEO of the Center for Responsible AI, reached the same conclusion by a different route: power is being transferred from labor to capital, particularly to those who control the “artificial intelligence factories”.
Wages Reward Labor; Ownership Rewards Capital
This distinction is fundamental.
Wages reward labor. Ownership rewards participation in capital. A property right gives its holder the right to receive the income generated by a particular asset.
For much of industrial history, wages did not make most workers wealthy, but they did provide individuals and families with autonomy. They made it possible to pay rent or mortgage payments, educate children, build savings, and make decisions without being permanently dependent on others' will.
It is this autonomy that AI may weaken, even in the absence of mass unemployment.
A person may keep their job and still lose relative economic standing. This will happen when their productivity grows far more rapidly than their wage, with the difference remaining in the hands of whoever owns the multiplier.
The greater the proportion of value generated by intelligent capital, the smaller, in relative terms, the share received by those who depend exclusively on wages.
This is why, in the age of AI, many people will need both: income from labor and income from ownership.
This idea ultimately became one of the panel’s main points of agreement: “wages reward labour; ownership rewards capital”.
From Wealth to Power
Economic concentration is not merely a question of material distribution. Wealth is converted into power.
It makes it possible to finance research, determine priorities, shape markets, influence governments and decide which systems reach society. When that wealth is multiplied by artificial intelligence, its conversion into power also accelerates.
The legal profession provides a particularly revealing example. The question is not merely whether machines will take work away from lawyers. It is also who will own the systems that select strategies, interpret case law, recommend settlements and make certain legal causes economically viable.
To paraphrase Yuval Noah Harari: who will fund the causes championed by intelligent machines?
The same question applies to medicine, journalism, education, security, public administration and scientific research. When machines participate in decisions that shape the structure of society, their ownership ceases to be a purely commercial matter.
It becomes an ethical, political and constitutional question.
Taxing Afterward May No Longer Be Enough
The usual response to the concentration of wealth is taxation. The state collects taxes and uses the revenue to finance benefits, public services and redistributive mechanisms.
But taxation provides a benefit. It does not provide ownership.
A benefit depends on a subsequent political decision, can be changed, and gives citizens no stake in the asset that produces the wealth.
For this reason, alongside redistribution, we must begin discussing mechanisms of pre-distribution: rules that shape the way the market distributes income and power before taxes and social benefits are applied.
Redistribution means sharing part of the wealth after it has already been produced and appropriated. In this context, we should focus particularly on one form of pre-distribution: broadening participation, from the outset, in the ownership of the infrastructure that produces the new wealth.
This does not mean abolishing private investment or placing data centers under direct public administration. It means designing models in which the state, communities and citizens can hold stakes, receive dividends or benefit from economic rights associated with infrastructure that uses collective resources.
The Lesson (and the Limits) of Alaska
The Alaska Permanent Fund demonstrates that a society can convert the temporary wealth generated by a natural resource into a lasting financial asset, managed professionally and independently, without losing its public character.
The fund was created through an amendment to the Alaska Constitution, approved by voters in 1976 under the political leadership of Republican Governor Jay Hammond.
The first dividend was distributed in 1982. In 2025, around $619 million was paid to approximately 619,000 beneficiaries, corresponding to $1,000 per beneficiary. The fund’s earnings now account for more than half of the state’s unrestricted general revenue.
The fundamental lesson is not the cheque’s annual value. It is a political decision to transform a resource that will eventually run out into an enduring asset.
The model does not, however, fully resolve the issue of ownership. Citizens receive a dividend, but they do not individually own a share in the fund that they can sell, transfer, or leave to their heirs. Moreover, the use of the fund’s earnings and the allocation of dividends remain subject to political decisions.
Alaska is therefore an inspiration, not a finished model.
An architecture suited to the age of AI would need to go further: it would have to create legally defined and verifiable rights that are resistant to discretionary manipulation and, where justified by the model, transferable between generations.
According to Reuters, citing the Financial Times, OpenAI proposed granting the United States Government a 5% stake and suggested that other major AI laboratories should do the same through a mechanism inspired by the Alaska Permanent Fund. The proposal shows that the industry itself has begun to recognize the political dimension of ownership. But transferring stakes to a central authority does not, in itself, guarantee genuine sharing with citizens.
Ownership concentrated in the hands of a government can also become a form of capture.
No One Protests Against a Dividend They Receive
The expansion of data centers is already encountering growing political resistance.
According to a survey by Data Center Watch, in the first quarter of 2026 alone, at least 75 projects in the United States, valued at approximately $130 billion, were blocked or delayed. In just three months, the figure matched the total recorded throughout the previous year. Local communities point to energy costs, water consumption, noise, lack of transparency, and insufficient benefits for affected areas.
People are not necessarily opposed to technology. They are opposed to bearing the costs without sharing in the benefits.
This leads to a simple political intuition: no one protests against a dividend they receive.
This also matters to investors. A community that participates economically in a project has stronger reasons to accept its installation, protect its continuity, and demand that it be well managed. Distributed ownership is not merely a matter of social justice. It can simultaneously serve as a social license to operate and as a mechanism for investment stability.
Sines: Partners, Not Merely Landlords
Portugal is in a unique position.
Start Campus announced an €8.5 billion investment to develop a major data center in Sines by 2030. Microsoft subsequently announced a $10 billion investment in artificial intelligence infrastructure at the same site, in partnership with Start Campus, Nscale and NVIDIA.
Sines benefits from Portugal’s Atlantic position, submarine cables, energy availability, territory, networks and public decisions recognizing the project’s strategic importance.
Future phases are not an abstract possibility. In May 2026, Nscale announced an expansion of its agreement with Microsoft and Start Campus, including the installation of more than 66,000 NVIDIA Rubin GPUs from late 2027 onwards and an additional investment of €695 million: €230 million for shared infrastructure and €465 million for a second 200 MW building. When we discuss the next phases, we are referring to developments that have not yet been completed and decisions that can still be made.
None of this means retrospectively altering contracts or commitments that have already been made. It means considering future phases and new projects.
Portugal should not limit itself to collecting taxes, rents, fees or compensation payments. It should examine ways of converting part of the value created into lasting assets for the country and its citizens.
Several solutions are possible: public equity stakes, national or regional funds, royalties converted into financial assets, participation units allocated to citizens, or mixed models that combine public, private, and community capital.
The specific architecture must be discussed rigorously. But the principle can be expressed simply:
We cannot be merely landlords. We must become partners.
We have made Sines a project of national interest; what remains is to turn the Portuguese people into economic participants in the project, rather than merely people with an interest in its success.
Sines could become our Alaska. This does not mean mechanically copying its fund, but rather applying to artificial intelligence, with even greater justification, the same question Alaska applied to oil: how can a territorial advantage and a strategic resource be converted into ownership that benefits several generations?
A New Infrastructure of Trust
Ownership only truly exists when it can be identified, proven, audited, and defended.
This is where the economic question reconnects with ethics and digital governance.
It is not enough to promise future distributions. We need legally defined rights or titles, transparent rules and mechanisms enabling every beneficiary to verify the existence of their rights and monitor the corresponding income.
In practical terms, any proposal for sharing the wealth generated by artificial intelligence should answer three simple questions. Is there a legally enforceable right, rather than merely a promise of distribution? Can each holder or beneficiary verify it without relying on the administrator’s goodwill? Is the rule resistant to silent alteration, or can it be hollowed out through a discretionary decision? These are three robustness tests, not a complete architecture. The specific architecture, including its legal and technological forms, requires a separate discussion.
Distributed ledger technologies and smart contracts can help automate certain rules, preserve transaction histories and make concealed alterations more difficult. But they do not replace the law, the courts, the state or democratic legitimacy.
Technology does not create justice by itself. It can, however, make verifiable what the law and democracy decide to protect.
Ownership is not simply that which power can never take away. In extreme circumstances, power can take almost anything. Ownership is, above all, that which power cannot remove in secret, erase without leaving a trace, or confiscate without being held accountable.
In reunified Germany, for example, the legislation adopted the restitution of property expropriated by the authorities of the German Democratic Republic between 1949 and 1990 as a general principle, subject to exceptions such as good-faith acquisitions or the material impossibility of restitution. The issue has particular resonance for me: in 1989, when I worked for Schering AG, I visited East Berlin shortly before the fall of the Wall and observed the visible effects of decades of severe restrictions on private property. By contrast, expropriations carried out under Soviet occupation between 1945 and 1949 were excluded from restitution, an exclusion upheld by the Federal Constitutional Court. The contrast illustrates the division of labor between technology and politics: documentary memory makes it possible to formulate and prove a claim; democracy decides which rights it recognizes and how they should be remedied. Confiscation may win the day. The record may win the decade.
Traditional trust depended, to a great extent, on the quality of the people holding positions of authority. Modern trust requires rules that remain verifiable even when those people fail.
This does not weaken the state. It frees the state from serving as the sole repository of every form of trust.
The Question of the AI Age
Regulating artificial intelligence is essential. We need systems that are safe, responsible, explainable, and subject to the law.
But this answers only the question of how artificial intelligence should operate.
Another question remains:
Who owns the machine?
A society can build technically responsible artificial intelligence while simultaneously allowing its ownership and benefits to become concentrated in a very small number of organizations.
AI regulation and the distribution of its infrastructure are not alternatives. They are two dimensions of the same system of governance.
The great question of the age of artificial intelligence is not merely who will have a job. It is who will own the machines that multiply the value produced by labor, because those who control this multiplier also tend to concentrate wealth and power.
And a multiplication of power on this scale requires modern governance built upon a distributed, verifiable infrastructure of trust that is protected against arbitrary alteration.