The industrial question opened by advanced artificial intelligence no longer concerns only the replacement of human labor, the automation of processes, or the increase of productivity. It concerns the formation of a new economic infrastructure, in which power is concentrated among those who organize the informational conditions within which firms, consumers, and institutions make decisions. The market does not disappear, but is progressively administered by computational architectures that select visibility, access, reputation, prediction, and the allocation of attention. In this transformation, competition continues to exist formally, but increasingly takes place within environments regulated by private actors capable of determining the operational conditions of competition.
Digital platforms anticipated this dynamic. They are no longer merely distribution channels, but systems of economic ordering. The European Union, through the Digital Markets Act, recognized this transition by qualifying certain large operators as gatekeepers, that is, platforms that provide essential access services between businesses and end users. The European Commission has identified search engines, app stores, marketplaces, operating systems, social networks, advertising services, and other platform services as central nodes of this new regulated contestability. Initially, the issue appeared to concern access to digital markets. With artificial intelligence, it moves to a deeper level: access to the capacity to decide.
The platform was already an invisible customs post. Artificial intelligence tends to transform it into an invisible factory of decision. It does not merely display products, classify content, or facilitate transactions. It interprets signals, anticipates behavior, produces recommendations, automates choices, generates languages, measures performance, and directs managerial action. Power no longer resides only in the possession of data, but in the ability to translate them into decision-making environments. Those who control these environments do not necessarily impose an explicit order. They make certain alternatives more probable, certain strategies more legible, certain operators more visible, certain behaviors more rewarded.
The vertical integration of decision-making processes constitutes the structural point of the transformation. In mature industrial economies, production, distribution, credit, communication, and regulation belonged to relatively distinguishable spheres. In the platformized and post-AI economy, these functions tend to converge into unified architectures. The same infrastructure can host the market, observe transactions, profile users, suggest the price, distribute advertising, manage payment, govern reputation, and train predictive systems on the behaviors collected. In economic terms, this is not only a matter of efficiency. It is the concentration of many market functions within a single computational device.
The OECD has indicated that market power in the digital economy cannot be understood only through prices, shares, or traditional sectoral boundaries. Network effects, closed ecosystems, economies of scale and scope, control of data, and the ability to define conditions of access produce forms of dominance that are less visible but deeper. Artificial intelligence radicalizes this condition because it adds generative and predictive capacity to the platform. The gatekeeper does not only control the threshold. It contributes to defining the very grammar of possible choices.
For firms, especially those that do not own computational infrastructures, dependence takes on a new form. It is no longer only dependence on a supplier, a logistics chain, or a commercial channel. It is dependence on systems that establish which information matters, which metrics prevail, which signals are converted into value, and which behaviors become economically rational. A firm may believe it is freely optimizing its own strategy while merely adapting to the coordinates produced by an external environment. Subordination does not appear as command, but as operational normalization.
Labor is affected by the same logic. AI does not act only as a technology of substitution, but as a system for recomposing the labor function. A growing part of value no longer arises from the direct execution of tasks, but from the ability to set up, verify, correct, interpret, and govern computational procedures. However, this recomposition does not automatically distribute autonomy. If AI systems are embedded in proprietary infrastructures, human labor risks becoming an operational complement to external decision-making models. Professional qualification, then, does not coincide with the use of the tool, but with the ability not to be entirely absorbed by its logic.
Money and the measurement of value are also indirectly involved. In platformized economies, value is not expressed only by monetary price, but also by indices of visibility, reputation, engagement, data accessibility, predictive quality, and computational capacity. The metric becomes infrastructure. What is measured tends to be governed; what is not measured tends to be marginalized. The traditional monetary function is not abolished, but flanked by private evaluation systems that condition access to revenues even before the monetary market expresses its own judgment.
The material dimension of this economy must not be obscured by its immaterial appearance. The International Energy Agency estimates that data centers consumed approximately 415 TWh of electricity in 2024, equal to about 1.5% of global electricity consumption, and projects that they may reach approximately 945 TWh by 2030 in the base scenario. Computational power, therefore, is not a pure digital abstraction. It is an industrial, energy, and geopolitical resource. Those who control computing capacity, cloud infrastructures, chips, energy, and data control an increasing share of future economic production.
UNCTAD has drawn attention to the unequal distribution of the benefits and costs of the digital economy, emphasizing how many economies contribute to the material base of digitalization without capturing a proportional share of its added value. This asymmetry becomes more relevant in the age of AI, because the distance between those who provide resources, data, labor, and infrastructures and those who own the models, platforms, and computational capacity tends to widen. The problem is not only technological. It is institutional. It concerns the conditions through which value is extracted, accounted for, retained, and redistributed.
Karl Polanyi showed that the market is never a natural order, but an institutional construction. The AI economy confirms this insight in a new form: the market appears spontaneous because the infrastructure that orders it remains hidden. Algorithms are not simple neutral instruments. They incorporate criteria of relevance, priority, exclusion, recommendation, and calculation. In this sense, every major computational architecture is also an implicit economic order. It does not necessarily replace the law, but produces an additional regulatory layer: continuous, adaptive, and often opaque.
The governance of artificial intelligence cannot therefore be reduced to the technical safety of models or to the protection of personal data. It must examine the economic structure of the systems that mediate collective decisions, market access, the organization of labor, and the distribution of value. Regulation limited to outputs risks failing to see the infrastructure. Industrial policy limited to adoption risks reinforcing external dependencies. Business strategy limited to efficiency risks mistaking optimization for autonomy.
What is at stake is not the rejection of platforms or artificial intelligence. It is the construction of institutional conditions capable of preventing economic decision-making from becoming entirely delegated to non-contestable private infrastructures. Firms will have to preserve informational sovereignty, internal competencies, plurality of channels, critical capacity regarding metrics, and control over strategic data. Institutions will have to distinguish between innovation and concentration, between access and dependence, between automation and the governance of decision. The new economic order of AI will not be determined only by those who develop the most powerful models, but by those who will be able to establish within what limits those models may organize the very field of economic freedom.
Global AI Observatory