AI faces ecological and energy limits. Constraints are pushing more sober and local models, particularly in the global South, which are redefining performance and sustainability.
Artificial intelligence is gaining ground at an unprecedented pace in developed economies. But as its uses multiply, a question long relegated to the background becomes central: what is the real cost of this acceleration?
Behind the technological promise, the limits are becoming more and more tangible: energy and water consumption, centralized infrastructures, dependence on ever heavier models. As AI spreads, its footprint ceases to be marginal and becomes structural and increasingly problematic.
From then on, the question is no longer just who adopts AI the fastest, but who is able to design it in a sustainable way. And in this area, the countries of the Global South perhaps have more to teach us than we imagine.
Exploding AI energy costs a model under pressure
The ecological paradox of AI is now documented. Data centers already represent a significant and rapidly growing share of global electricity consumption. To make matters worse, their energy demand could more than double by 2030. Added to this pressure are significant water needs for cooling, as well as a continued growth in electronic waste, the health and environmental impacts of which are largely externalized to the most vulnerable countries. In reality, AI mainly benefits advanced economies, while a growing share of its externalities weighs on other regions. This asymmetry is neither sustainable nor neutral.
Second entrant AI why leaving later becomes a strategic advantage
This is where what we too rarely call second-mover advantage comes into play. Paradoxically, in many countries of the Global South, the absence of heavy and carbon-intensive infrastructure is not a handicap to be corrected, but paradoxically a blessing in terms of sustainable development. Where advanced economies are today forced to backpedal on systems designed without environmental constraints, others, who arrived later in this area, can integrate sobriety, decentralization and resilience from the design stage.
Technical constraints AI the hidden engine of frugal innovation
Constraints – intermittent connectivity, high costs, energy instability – naturally guide technological choices. They favor lighter models, specialized systems, treatments as close as possible to uses. Edge analytics, small language models, sectoral or linguistic AI: these approaches deliver high value with a much smaller footprint. They operate outside of the permanent cloud, are resistant to outages, and adapt to local contexts. In other words, they achieve the feat of adapting to local constraints to respond effectively to real, measurable, and often urgent needs.
Sectoral and local AI for concrete use cases that challenge global models
And it’s not just a posture. The concrete use cases are already there. In agriculture, on-board tools make it possible to diagnose diseases or optimize irrigation directly in the field. In energy, predictive systems improve the maintenance of mini-grids or off-grid batteries. In health, linguistically adapted applications operate without heavy infrastructure and strengthen access to care. This AI is neither spectacular nor universal. She is useful. And robust.
AI performance changing metrics to integrate sobriety and trust
This change also forces advanced economies to exercise lucidity. It also shakes up our certainties and invites us to greater humility. For years, AI performance has been evaluated primarily based on scale: number of parameters, computing power or size of infrastructure. These quantitative indicators are no longer enough. Other, more qualitative criteria become essential: inference efficiency, contextual robustness, energy sobriety, explainability, trust. Measuring AI solely by power ignores its systemic costs – and its fragilities.
Sustainable AI the end of the spectacular model in favor of efficiency
The Global South gives us here a great lesson in frugal, responsible and sustainable AI. This does not point towards a marginal path. It reveals an alternative, credible and already operational trajectory. A trajectory where “bigger” is not synonymous with “better”, where design takes precedence over repair, where constraint becomes a driver of innovation. This is not a question of catching up. It’s a lesson in method. The AI of tomorrow may be less impressive on paper. Above all, it will be more adapted, more sustainable and more aligned with available resources. In a constrained world, this form of clarity could well become the new frontier of performance.
For public and private decision-makers, the challenge is no longer to run behind power, but, as the countries of the Global South show us, to choose architectures consistent with economic, energy and social realities. This shift in focus, from volume towards accuracy, could take hold more quickly than we imagine. And transform a trajectory perceived as peripheral into a new global standard.
