Advanced artificial intelligence can no longer be interpreted as an immaterial technology located solely in the domain of software, models, and cognitive interfaces. Its economic trajectory reveals a deeper structure: AI is a form of computational power rooted in physical infrastructures, industrial supply chains, energy networks, mining capacities, logistical systems, and political jurisdictions. Its apparent operational lightness conceals a material subsoil that conditions access, autonomy, productivity, and sovereignty. From this perspective, every response generated by a model, every automated forecast, every algorithm-assisted decision-making process belongs to a concrete chain made up of data centers, semiconductors, electricity, water, fiber optics, copper, lithium, graphite, rare earths, and concentrated industrial capital.

The first interpretive error consists in separating AI from its energy base. According to the International Energy Agency, in the base case global electricity consumption by data centers could double by 2030, reaching approximately 945 TWh; in scenarios of accelerated AI adoption, electricity demand from data centers could exceed 1,700 TWh in 2035. This figure does not merely measure an increase in consumption. It signals the transformation of computation into an infrastructural issue, because data centers concentrate demand in specific territorial areas and can generate local pressures on power grids that are far more acute than their percentage weight at the global level would suggest.

The new cognitive economy therefore rests on a structural tension: the more value appears to move toward symbolic automation, the more dependence grows on material inputs that are scarce, localized, and politically exposed. The IEA found that in 2024 lithium demand increased by nearly 30%, while nickel, cobalt, graphite, and rare earths grew by between 6% and 8%. Even more relevant is the concentration in refining stages: between 2020 and 2024, the average share of the top three refining countries for the main energy minerals rose from 82% to 86%. AI does not depend only on the availability of chips, but on the entire industrial ecology that makes it possible to produce, power, and replace them.

From this derives an essential economic consequence: digital strategy is no longer separable from industrial strategy. A company that adopts AI without measuring its exposure to cloud, semiconductors, infrastructure providers, energy constraints, and data sovereignty transfers a growing share of its decision-making capacity toward systems it does not control. Competitive advantage no longer consists only in the faster use of tools, but in the ability to understand which conditions make those tools available, at what price, under which jurisdiction, and with which future dependencies.

Computational power thus becomes a new economic magnitude, comparable in strategic function to physical capital in previous industrial revolutions. It is not simply a technical resource, but an enabling capacity: it determines who can train models, who can integrate them into production processes, who can scale their use, who can negotiate favorable conditions, and who remains confined to the position of dependent user. The distinction between those who possess computational capacity and those who purchase it as a service tends to reproduce, in a new form, a hierarchy between the center and the periphery of the digital economy.

The OECD’s measurement of domestic availability of public cloud compute for AI confirms this direction. The organization has developed a methodology to estimate the global physical distribution of cloud regions relevant to AI, distinguishing economies according to the effective availability of public computational capacity. This point is institutionally decisive: the geography of compute does not coincide with the geography of AI use. Many economic systems will be able to employ intelligent applications, but only some will possess the infrastructure necessary to develop them, train them, adapt them, and govern them according to their own interests.

AI governance therefore shifts from the regulatory field alone to the infrastructural field. Rules, ethical standards, and accountability frameworks remain necessary, but they do not exhaust the matter. A legal order can regulate the use of artificial intelligence without possessing the material conditions of its production. It can define principles of transparency without controlling computational capacity. It can impose security obligations without dominating the technological supply chains that make systems available. In this gap between norm and infrastructure lies one of the principal fragilities of contemporary institutions.

The political nature of AI does not reside only in algorithmic biases or in its effects on professions. It resides in the very form of the infrastructure. A data center is not a neutral building, a chip is not a neutral component, a cloud platform is not a neutral intermediary. They are devices for organizing economic power. They establish who accesses power, who rations it, who monetizes it, who bears it as a cost, and who transforms it into strategic capacity. Technology incorporates relations of dependence even before producing automated decisions.

For companies, this implies a revision of the way technological risk is defined. It is no longer sufficient to assess model accuracy, data security, and expected productivity. Corporate governance must integrate the risk of supplier concentration, the vulnerability of cloud infrastructures, energy availability, data localization, the resilience of supply chains, and the possibility that computational capacity may become subject to economic or political rationing. AI is not only a factor of efficiency. It is a new balance-sheet, contractual, and institutional exposure.

Industrial policy itself is changing its object. The World Economic Forum has observed that governments across different regions, from Singapore to Saudi Arabia, from Ireland to Kenya, are competing to attract data center investment through tax incentives, accelerated permitting, and dedicated infrastructural zones. This race is not only about growth in the digital sector. It concerns the ability to preside over an essential node of the future economy, because data centers attract ecosystems of cloud, fintech, e-commerce, automation, and artificial intelligence. The localization of computation becomes a variable of national development.

Within this framework, labor does not simply disappear under the effect of automation. It is repositioned within a more opaque chain, in which the visible work of users and professionals is accompanied by infrastructural work, extractive work, maintenance work, energy work, annotation work, regulatory work, and financial work. AI productivity appears as the yield of the algorithm, but it incorporates a plurality of social and material contributions that are often removed from ordinary economic representation. The question of value therefore returns to the center: not only who produces outputs, but who controls the conditions of automated production.

Money and finance also enter this transformation. The construction of computational capacity requires large-scale investment, long-term energy contracts, privileged access to capital, and alliances among hyperscalers, chip producers, infrastructure funds, and states. Compute tends to become a financializable, programmable, and strategically allocable asset. In a phase of relative scarcity, computing power can assume functions analogous to a productive reserve: not money in the strict sense, but a capacity to generate future economic options. Those who possess it can decide the timing, prices, and priorities of innovation.

The most delicate institutional point concerns the anesthesia of judgment. AI produces ordered decision-making surfaces: scores, forecasts, classifications, scenarios, recommendations. These surfaces can induce organizations to confuse the form of rationality with the substance of responsibility. Every automated output, however, has a material and political underside: energy consumed, contractual dependencies, mineral supply chains, competing jurisdictions, proprietary standards, industrial concentration. Managerial and public responsibility begins when AI adoption is measured not only by what it makes possible, but by the kind of economic order it makes more probable.

Advanced artificial intelligence therefore opens a phase in which the economy of knowledge and the economy of matter are no longer separable. Its center is not only the model, but the set of conditions that allow the model to exist, operate, and scale. The decisive question does not concern the quantity of intelligent tools available, but the distribution of the power necessary to produce and govern them. The algorithm promises abstraction, but its effectiveness is decided in the density of infrastructures, in the geography of resources, in the quality of institutions, and in the capacity of economic systems to see the material subsoil of their own intelligence.

Global AI Observatory