Advanced artificial intelligence is structurally transforming the architecture of the contemporary economy, affecting the relationships between markets, institutions and decision-making power. The competitive paradigm that characterized industrial and financial capitalism, based on a relatively horizontal plane of actors coordinated by prices and formal rules, is gradually giving way to a more vertical configuration. In this new structure, power tends to concentrate in computational nodes capable of processing, anticipating and orienting decisions on a systemic scale. AI does not operate as a simple instrument of efficiency, but as a cognitive infrastructure that redefines who decides, on what grounds and with what margins of autonomy.

Computational capacity thus becomes a primary strategic resource. It is not merely a matter of possessing financial capital or economies of scale, but of controlling models, data and algorithmic architectures capable of transforming raw information into operational foresight. According to estimates by the International Energy Agency and the OECD, the training of large artificial intelligence models requires infrastructure investments and energy consumption accessible to an increasingly limited number of actors. This material constraint produces an asymmetry that is not only economic, but epistemic. Those who govern the most advanced models do not simply compete more effectively, but occupy a position of cognitive command, influencing the behavior of others even before it manifests itself in the market.

In this context, formal competition tends to coexist with a reduction in effective competition. Markets remain theoretically open, but access to the real conditions of competition becomes unequal. Firms and institutions operating outside major algorithmic infrastructures find themselves reacting to dynamics they do not control and that are often not fully transparent. The asymmetry does not concern only the availability of information, but the ability to transform it into decision orientation. Some actors participate in the economic game, while others define its operational coordinates in real time.

The centralization of decisions does not occur through explicit political acts or visible concentrations of capital, but through the widespread adoption of predictive and optimization systems. Investments, pricing policies, supply chain management, labor allocation and divestments are increasingly entrusted to models that maximize complex objectives under predefined constraints. Studies by the Bank for International Settlements show how the extensive use of similar algorithms in financial markets has contributed to a growing homogeneity of behavior, increasing apparent efficiency but also systemic vulnerability. Economic decision-making thus tends to shift from a space of negotiation among subjects to a process of calculation internal to opaque architectures.

This transformation also redefines barriers to entry. These are no longer solely economic or regulatory, but cognitive. Access to large volumes of data, advanced computing capacity and specialized expertise becomes a necessary condition to compete in increasingly broad sectors of the economy. Innovation risks being channeled along trajectories already traced by actors who can afford large-scale experimentation, while smaller actors operate under conditions of informational scarcity and technological dependence. The result is an economic structure that is formally dynamic, but substantially stratified.

Corporate governance also reflects this mutation. Leadership models adapt to a reality in which decision-making power is mediated by systems that promise rationality and precision, but incorporate assumptions that are difficult to verify. The role of the decision-maker shifts from the act of choice to the validation of algorithmic outputs, with a consequent redefinition of responsibility. Trust in the machine becomes a critical organizational variable, while the capacity to question the results produced by models tends to diminish, especially in contexts of high competitive pressure.

This dynamic raises issues that go beyond economic efficiency. Karl Polanyi highlighted how an economy disembedded from social relations risks producing profound dislocations. Computational hierarchies represent a possible contemporary manifestation of this disembedding. An economic order that self-organizes according to technical criteria may appear internally coherent, but fragile with respect to its long-term social and institutional consequences. Decisions that are optimal for the model can generate cumulative effects on employment, income distribution and trust that fall outside short-term metrics.

At the macroeconomic level, the concentration of decision-making power simultaneously amplifies efficiency and risk. The ability to rapidly coordinate large quantities of resources can reduce frictions and waste, but the convergence of decisions increases exposure to systemic errors. When many actors rely on the same models or similar logics, deviations become less frequent but more costly. The economy takes the form of a stable but rigid structure, in which adaptive capacity depends on the diversity of available cognitive architectures.

For entrepreneurs and public decision-makers, this scenario imposes a strategic reflection on the relationship between autonomy and integration within large algorithmic ecosystems. The choice does not concern only the adoption of advanced technologies, but the degree of dependence on external cognitive infrastructures and the ability to preserve proprietary decision spaces. At stake is also the definition of economic value. Value that is immediately measurable and optimizable tends to be privileged by models, while forms of value linked to the interpretation of emerging needs or long-term horizons risk remaining invisible.

The trajectory toward a reduction in effective competition is not inevitable, but it is consistent with the current distribution of computational power. Artificial intelligence infrastructures function like all fundamental infrastructures, delimiting what is possible and what is not. Treating them as neutral tools means renouncing any inquiry into the hierarchies they produce. In an economy in which the algorithm tends to occupy the invisible apex of decision-making processes, the central issue becomes the distribution of the power to define the economic and institutional future, even before the outcomes that such a future will generate.

Global AI Observatory