Artificial intelligence is often described as an immaterial technology, capable of expanding indefinitely through software, data, and ever more sophisticated computing capacity. This representation, however, obscures the material foundation on which the entire system rests. The AI economy is, first and foremost, an energy economy. Every advance in models, every increase in scale, every promise of extended automation translates into growing demand for electricity, infrastructure, and physical resources. Energy thus emerges as the true structural parameter of artificial intelligence, the factor that defines its limits, conditions its distribution, and reshapes its economic and institutional implications.
Consumption data make this tension difficult to ignore. According to the International Energy Agency, data centers worldwide consumed approximately 460 terawatt-hours of electricity in 2022, just under the annual consumption of a country such as France. More recent estimates indicate that, driven by the expansion of generative AI, this demand could exceed 1,000 terawatt-hours by 2026. Training large-scale models requires weeks of continuous computation on extremely high energy-density infrastructures, while daily inference at global scale multiplies consumption on a permanent basis. Performance gains do not follow a linear curve, but an exponential one, making it increasingly difficult to separate technological growth from pressure on resources.
This dynamic introduces an energy competition that extends far beyond the market. Access to abundant, stable, and low-cost energy becomes a necessary condition for AI development. Not all economies, nor all organizations, possess this condition. Areas with fragile power grids, high dependence on fossil fuels, or elevated energy costs start from a structurally disadvantaged position. By contrast, territories capable of combining production capacity, renewable sources, and infrastructural stability acquire a systemic advantage. In the language of political economy, energy becomes a factor of technological selection, determining who can scale and who must slow down.
Environmental impacts are not a side effect, but an integral part of this equation. Studies published by Nature Climate Change and MIT show that the carbon footprint of large language models can reach, in training alone, emissions equivalent to hundreds of transcontinental flights, especially when the energy used derives from fossil sources. The paradox is evident: a technology often presented as a tool for optimization and sustainability risks generating new environmental externalities if not embedded within a coherent energy strategy. In the corporate world as well, experience shows that local process optimization can produce systemic costs that emerge only in the medium term, when infrastructures are pushed beyond their sustainability thresholds.
In this context, energy policies take on direct strategic relevance for the development of artificial intelligence. The location of data centers increasingly follows the availability of renewable energy and stable grids, as demonstrated by recent concentrations of investment in Nordic countries or regions rich in hydroelectric power. Regulatory choices regarding emissions, incentives, and infrastructure define the perimeter within which AI can grow. Energy ceases to be an external variable and becomes an integral part of technological governance. For firms, this means incorporating the energy dimension into industrial strategy, alongside human capital and intellectual property.
The thought of Hans Jonas offers a useful interpretative key to grasp the scope of this transformation. Jonas formulated the principle of responsibility in response to the growth of technological power, arguing that every capacity to intervene in the world imposes a duty toward the future. Applied to AI, this principle makes clear that energy choices are not neutral. Consuming large amounts of energy today to enhance intelligent systems produces effects that extend over time, affecting environmental, social, and geopolitical balances. Computational power thus becomes an ethical issue even before it is a technical one.
In the corporate world, this responsibility translates into a revision of decision-making models. Organizations that make intensive use of AI can no longer assess their investments solely in terms of efficiency or immediate economic return. Energy and environmental impact become strategic variables, capable of generating regulatory, reputational, and operational risks. Individual cognitive limits make it difficult to grasp the complexity of these interdependencies, but ignoring them exposes organizations to sudden shocks, as demonstrated by recent tensions in global energy markets.
The distributive dimension completes the picture. Territories that host energy-intensive infrastructures bear environmental and social costs often disproportionate to local benefits, while the economic value generated by AI concentrates elsewhere. This asymmetry reproduces, in amplified form, the dynamics of global value chains: those who provide the material base remain marginal relative to those who control application and rent. For firms and institutions, the issue concerns not only where to locate infrastructures, but how to redistribute burdens and benefits in a manner compatible with the social resilience of the territories involved.
Energy thus becomes the concrete measure of the future of artificial intelligence. Not an absolute limit, but a reality criterion that forces confrontation with the finiteness of resources. The rhetoric of unlimited technological growth enters into tension with the materiality of the systems that make it possible. Recognizing this constraint does not mean halting innovation, but bringing it back within a horizon of conscious decision.
Energy policies integrated with industrial strategies can transform constraint into leverage. Investing in efficiency, renewable sources, and less energy-intensive computational architectures represents not only an ethical choice, but a decision in favor of economic resilience. Organizations that are able to anticipate this convergence between energy and artificial intelligence will build an advantage that is less immediate, but more stable over time.
The AI economy thus occupies a space of tension between power and limit. On one side, the drive toward ever greater computational capacity; on the other, the awareness that every increase in power carries a material and political cost. Within this space emerges a responsibility that cannot be delegated either to algorithms or to the market. Energy, from a technical variable, becomes a measure of the meaning with which the present decides its technological future.
Global AI Observatory