The economy of artificial intelligence is not organized around the diffusion of technology, but around the concentration of the conditions that make it operational. Computation, data, energy, digital infrastructures, and advanced industrial capacity are neither fungible resources nor easily replicable. On the contrary, they tend to accumulate in a limited number of global nodes, transforming AI into a lever of structural power rather than a simple factor of innovation. In this configuration, power does not derive from the use of artificial intelligence as such, but from control over the resources that make its existence at scale possible.
The control of critical infrastructures constitutes the first level of this concentration. Training advanced models requires computational capacities that only a few operators can provide on a continuous basis. According to estimates by the International Energy Agency, data centers and advanced computing infrastructures account for approximately 2 percent of global electricity consumption, a share set to grow rapidly with the expansion of generative AI. This energy intensity is not a side effect, but a structural constraint that selects who can sustain technological evolution over time. Where energy is abundant, stable, and low cost, artificial intelligence develops. Where it is not, dependence emerges.
Added to this dimension is control over the semiconductor supply chain. The production of advanced chips is concentrated in a very small number of facilities and territories. According to data from the Semiconductor Industry Association, more than 90 percent of the most advanced semiconductors are produced in a single geographic area. This concentration is not only industrial, but strategic. Those who control access to frontier hardware indirectly shape both the speed and the direction of AI development. Technology, in this sense, becomes a function of industrial geopolitics rather than of open markets.
The combination of computation, energy, chips, and data generates a cumulative effect. Resources do not merely add up, they reinforce one another. Greater computational capacity enables the processing of more data, which in turn improves models, making the infrastructure even more attractive for new information flows and new investments. This mechanism produces a structural advantage that is extremely difficult to offset. The issue is not competing on individual products or applications, but whether one enters a self-reinforcing circuit or remains outside it. In the corporate world, this marks the difference between participating in the platform and remaining a subordinate user.
From this concentration arises an asymmetry of power among states that goes beyond traditional economic metrics. Some political and economic systems possess real margins of choice: they can experiment, slow down, regulate, and steer. Others are forced to adapt to standards and infrastructures defined elsewhere. The asymmetry does not appear as direct imposition, but as inequality of options. It is the difference between deciding and reacting. Even within complex organizations, real power does not reside in the final act, but in the ability to define the agenda and the perimeter of alternatives.
This structure enables a silent instrumentalization of technology. Limiting or conditioning access to advanced computational resources, strategic cloud services, or critical hardware components directly affects the ability to compete and to govern complex processes. There is no need for explicit sanctions or coercive acts. It is sufficient to act on system architecture. In corporate terms, this is equivalent to controlling bottlenecks in the value chain, shaping the behavior of other actors without formally imposing it.
The new global equilibria that result from this configuration are unstable and difficult to reverse. States seek to reduce exposure through industrial policies, public investment, and strategic alliances. However, the logic of scale inherent in artificial intelligence tends to reproduce initial concentration. According to the OECD, more than 70 percent of global investment in AI infrastructure is concentrated in fewer than ten countries. This figure signals a reduction in the mobility of technological power, with a progressive hardening of global hierarchies.
In this scenario, Hannah Arendt’s reflection on power as the capacity to organize collective action offers a decisive interpretative key. Power, in its most effective form, does not coerce, but structures contexts. Artificial intelligence, understood as concentrated infrastructure, exercises precisely this type of power. It does not impose decisions, but renders some trajectories natural and others impracticable. The strength of AI as a geopolitical lever lies in its normalization, in its ability to operate as the invisible background of choices.
The parallel with the corporate world is direct. Organizations that control platforms, standards, and infrastructures do not need to intervene in every decision made by partners. It is sufficient to design the decision-making environment. Leadership thus shifts from command to system configuration. In the same way, geopolitical competition around AI is not played out only in diplomatic arenas, but in the technical design of the architectures that sustain the global economy.
This concentration also generates an unequal distribution of costs and benefits. The economic and strategic benefits of AI tend to concentrate in infrastructural nodes, while environmental, energy, and social costs are often externalized. According to data from the Environmental Justice Atlas, energy-intensive digital infrastructures are increasingly located in areas with lower capacity for political opposition. The separation between places of value and places of impact introduces tensions that are difficult to address with traditional redistributive instruments.
Decision-making processes mediated by artificial intelligence amplify this dynamic. By reducing human cognitive limits, AI promises greater rationality, but risks crystallizing already concentrated power structures. When decisions are optimized within infrastructures controlled by few actors, the space for imagining alternatives narrows. Technical efficiency can turn into implicit destiny. The responsibility of political and managerial elites is therefore not to slow innovation, but to prevent it from closing the space of the possible.
The concentration of the resources on which artificial intelligence is based is not a temporary anomaly, but a structural feature of its development. It cannot be eliminated without calling into question the very logic of the technology. It can, however, be governed, made visible, and subjected to explicit choices. This requires recognizing that technological infrastructures are never neutral, but incorporate a vision of economic and political order.
The new global order emerging around AI is neither stable nor definitive. It is crossed by continuous tensions between concentration and diffusion, cooperation and competition, autonomy and dependence. Enterprises operating in this context know that the greatest risk is not open competition, but unrecognized asymmetry. Likewise, states seeking to preserve margins of sovereignty must question not only what to develop, but which dependencies to accept.
Artificial intelligence makes visible a profound transformation of contemporary power. No longer exercised through spectacular acts, but through the silent management of the resources that make the world function. In this management lies a contest that produces no permanent winners, but demands a new lucidity. Understanding concentration as a geopolitical lever does not mean demonizing technology, but recognizing that every infrastructure is also a choice of order. It is within this awareness, still fragile but decisive, that the space opens for an equilibrium that is not imposed, but continuously negotiated.
Global AI Observatory