Based on an opinion document provided by the editorial team.
A year ago, artificial intelligence still looked like a benchmark race. The conversation was organised around a simple question: who had the best model. By June 2026, that reading is no longer enough. The decisive question is no longer only who develops the most advanced AI. What matters above all now is who can sustain it.
Artificial intelligence has stopped being a purely digital industry. It has become a heavy industry. It depends on scarce chips, abundant energy, cooling water, fibre optics, land, permits, long-term capital and the state's ability to coordinate all of that. The model that answers in seconds rests on infrastructure that takes years to build.
That is why the new AI gap no longer resembles the old internet-access gap. It is not enough to have connected users, technical talent or creative startups. The critical difference lies in whether a country or region can train, run, audit and govern AI systems from infrastructure that is its own, shared or at least negotiated from a position of strength.
AI as power infrastructure
AI is becoming a layer of power. It does not only automate tasks: it reorganises markets, accelerates scientific research and transforms defence, health, education, finance and public administration. Whoever controls its infrastructure does not only control a technology. They control a growing portion of the cognitive capacity available to a society.
The new map does not divide the world between those who use AI and those who do not. Everyone will use it. The real divide is between those who control compute, energy, data, models and rules, and those who merely rent access to intelligence designed and governed elsewhere.
The United States: intelligence as national infrastructure
The United States remains the centre of gravity of global AI. It has leading labs, hyperscalers, venture capital and privileged access to advanced chips. But in 2026, American AI can no longer be read only as private innovation. It increasingly appears as national infrastructure.
Large data centres consume electricity comparable to that of a city. That makes technological debate also territorial, tariff-related and political. Who gets priority on the grid? Who pays for expansion? Which productive projects are displaced? The magic of chatbots now carries behind it a material dispute over energy, water and permits.
Washington also wants to export AI to the world while restricting certain capabilities for national-security reasons. That turns technological dependence into political dependence. For a Latin American or European company, using a US model may remain efficient. The problem is that access may become uncertain because of a regulatory decision made far from its own market.
China: autonomy under pressure
China is moving through the opposite mirror. External restrictions did not stop its progress; they reshaped it. Instead of relying entirely on frontier hardware, it accelerated a home-grown ecosystem built on efficiency, adaptation and continuity. The lesson is not that compute stops mattering, but that when brute scale becomes harder, optimisation becomes strategy.
China is showing that technological sovereignty does not always mean having the best thing in the world. Sometimes it means not being stoppable by an outside decision. That logic explains part of the turn toward open or semi-open models, efficient architectures and tools adaptable to available hardware.
Europe: regulating is not enough if you do not build
Europe understood earlier than many that AI could not be left entirely to product logic. The AI Act consolidated a clear regulatory position. But by 2026 that advantage is also revealing its limit: regulation without infrastructure is not enough.
European concern is no longer only about preventing abuses or bias. It is about preventing the entire continent from depending on clouds, chips and models designed elsewhere. That is why the language of technological sovereignty is growing stronger: AI factories, sovereign cloud, regional compute, targeted public procurement and strategic open source.
Latin America: using is not the same as developing
Latin America has talent, universities, renewable energy and an urgent need to raise productivity. But the region is still confusing adoption with development. Integrating external APIs is not the same as building sovereignty. Automating public procedures with imported models may improve efficiency, but it does not change a country's structural position if the data, infrastructure and rules remain under someone else's control.
The risk is becoming a computational periphery: lots of superficial adoption, little genuine capacity. Governments buying packaged solutions. Companies plugging chatbots into internal processes. Startups building on top of external models. Plenty of marginal improvement, little regional strategy.
Argentina: more compute, but also better judgement
Argentina has real assets: public universities, a scientific tradition, a software culture, technical talent, energy, a favourable time zone and a habit of building under constraint. But more compute alone is not enough as a strategy.
The demand from local researchers and entrepreneurs for more compute capacity is reasonable. Without GPUs, cloud credits or shared infrastructure, Argentine talent is condemned to testing small ideas or depending on foreign APIs. But Argentina cannot play this race like an infinite hardware auction. If it tries, it will always arrive late and buy expensive.
The opportunity lies elsewhere: combining public compute with technical intelligence. Yes, infrastructure matters, but so does mastering fine-tuning, quantisation, efficient inference, local auditing, quality datasets and concrete applications in health, education, industry, administrative justice and science. In low-capital countries, efficiency stops being a secondary virtue. It becomes a survival strategy.
What is at stake
AI has entered a stage that is less glamorous and more decisive. The age of spectacular demos has given way to the age of power permits, data centres, audits, supply chains and institutional design. That does not make AI any less revolutionary. It makes it more similar to the major infrastructures that already reordered the world.
For countries such as Argentina, the discussion cannot be reduced to enthusiasm or fear. Nor simply to regulating or deregulating. The relevant question is what place we want to occupy in the global intelligence chain. We can be only a market and a source of cheap energy for someone else's infrastructure, or we can build a capacity of our own, partial but real, before the wave passes over us.
Sources: