The economy of artificial intelligence makes visible a structural transformation of productive factors. Scarcity no longer concerns only natural resources, skilled labor, or financial capital, but a technical capacity that conditions the entire set of economic and institutional processes: computing power. Within the AI ecosystem, computation is not a neutral support, but an enabling infrastructure that determines who can fully participate in value creation and who remains in a subordinate position. Computational availability thus becomes a systemic constraint, capable of orienting development trajectories, competitive structures, and power relations.
The contemporary production function incorporates computation as a multiplicative factor. According to estimates by the McKinsey Global Institute, the large-scale adoption of artificial intelligence systems could contribute up to 4.4 trillion dollars annually in global value added, yet this value is strongly concentrated in contexts that possess adequate computational infrastructures. Training large-scale models requires millions of GPU hours, high energy consumption, and a level of technical orchestration that is not widely accessible. The result is a reorganization of productivity: where computation is abundant, innovation accelerates; where it is limited, even skills and data remain underutilized.
The investments required to sustain this capacity reinforce mechanisms of selective accumulation. According to the Semiconductor Industry Association, the cost of building a single high-end AI data center now exceeds one billion dollars, while the production of advanced chips requires multi-year investments and highly specialized supply chains. This explains why more than 60 percent of global advanced computing capacity is concentrated in a few areas of North America and East Asia. This concentration is not episodic, but cumulative: infrastructure, data, and human capital tend to co-locate, reinforcing existing advantages and making entry by new actors increasingly difficult.
This dynamic produces structural barriers that do not take the form of explicit prohibitions, but of economic and technical thresholds. Access to computation often occurs through cloud platforms, at variable costs and under contractual constraints that transfer decision-making power toward infrastructure providers. Firms that cannot afford direct investments remain dependent on external capacity, adapting their business models to the conditions imposed by the ecosystem. It is a form of functional dependence that does not eliminate competition, but channels it within boundaries defined by those who control computation.
The macroeconomic implications are significant. When computation becomes a critical productive factor, growth tends to polarize. According to the OECD, sectors with high AI intensity show productivity rates up to three times higher than those with low computational intensity, with direct effects on income distribution and demand for skilled labor. Aggregate growth may appear positive, but accompanied by rising inequalities among sectors, territories, and economic systems. Computational scarcity thus acts as a new mechanism of structural selection.
The thought of Karl Polanyi offers a useful interpretative key to read this transformation. Polanyi showed how the commodification of elements that perform a systemic function produces imbalances that the market alone is unable to correct. Treating computing power as an ordinary commodity risks ignoring its infrastructural role. Computation is not merely a productive input, but a condition for the functioning of the entire economic system. When its allocation is left exclusively to market logics, rigidities emerge that require institutional interventions of rebalancing.
In the corporate world, this dynamic translates into a growing tension between efficiency and strategic control. Outsourcing computational capacity enables speed and flexibility in the short term, but introduces a dependence that conditions future choices. Human cognitive limits are compensated through automation, but at the cost of reduced decision-making autonomy. Leadership, in this context, does not coincide with the maximum exploitation of computation, but with the capacity to integrate it without turning it into an irreversible constraint.
The concentration of computational power also raises issues of accountability. Those who control computation also control access to advanced predictive capabilities, with direct effects on markets, public policies, and governance systems. According to the World Economic Forum, a growing share of strategic decisions, from the management of supply chains to the assessment of financial risk, is today mediated by high computational intensity models. However, the transparency of these processes remains limited, making it more difficult to attribute responsibility in cases of error or failure.
The geopolitical dimension of computational scarcity is equally relevant. States and regions lacking adequate infrastructures see their capacity to orient industrial and scientific development reduced. Economic sovereignty becomes increasingly linked to the ability to guarantee stable access to computation, energy, and semiconductors. Recent restrictions on the export of advanced chips and industrial policies dedicated to AI show how computing power is now considered a strategic resource, comparable to energy or transport networks.
The new computational scarcity is therefore not a contingent technical limit, but a structural element of the artificial intelligence economy. Recognizing it means shifting analysis from enthusiasm for applications to the governance of the material conditions that make them possible. This does not imply imagining a total democratization of computation, an unrealistic objective, but avoiding that its concentration produces economic and institutional rigidities that are difficult to correct.
Within this space of tension, computational capacity reveals itself as a measure of contemporary power. Not because it automatically determines economic outcomes, but because it delimits the field of possibilities. Governing this scarcity means recognizing that artificial intelligence is not an autonomous engine of growth, but an infrastructure that requires political, industrial, and organizational decisions. It is within this awareness, more than in any promise of technological abundance, that the ground opens for a less fragile and more responsible economic equilibrium.
Global AI Observatory