Latest

Artificial Intelligence

Sovereign AI in Latin America: 11 projects across eight countries

Latin America has 11 sovereign AI projects across eight countries, with progress in models, supercomputing and its own infrastructure.

Sovereign AI in Latin America: 11 projects across eight countries
Sovereign AI in Latin America: 11 projects across eight countries

Latin America has 11 sovereign artificial intelligence initiatives spread across eight countries, according to version 2.0 of the Sovereign AI Index from the Center for a New American Security (CNAS), a global map that tracks 184 government-backed projects aimed at reducing their technological dependence and increasing their decision-making capacity over the development and use of artificial intelligence (AI).

Brazil, Chile, Colombia, Mexico, Argentina, Uruguay, Paraguay and the Dominican Republic already have concrete initiatives related to computing infrastructure, language models or national programs for access to AI resources. The region is still far behind Europe, with 79 projects, and Asia-Pacific, with 54, but it is ahead of Eurasia, with eight; Africa, with seven; and Canada, with three.

The report, with data updated to June 2026, offers a revealing picture of how governments, including those in Latin America, are trying to gain room for maneuver against the United States and China, which account for 90 percent of the computing capacity needed to develop and deploy frontier AI models.

The situation poses a paradox, since countries want to build more sovereign AI, but much of the technology needed to do so still depends on foreign companies and suppliers. According to CNAS, more than three-fifths of the projects tracked worldwide declare at least one foreign partner, and four out of five of those partners are US companies. NVIDIA, for example, supplies the GPUs used in 45 percent of all infrastructure projects registered globally.

What is sovereign AI?

Although there is no single, agreed definition, the CNAS index understands a sovereign AI project as any government-backed initiative explicitly linked to national strategic interests and supported by material public investment in computing infrastructure, models or data ecosystems.

Mexico and Brazil hold 95% of AI patents in Latin America

The reasons driving these projects vary between national security, economic development, cultural and linguistic preservation and the desire to achieve greater autonomy from foreign technological dependence. Often, several of these objectives appear combined within the same strategy.

The report classifies projects into three categories that reflect fundamental layers of the artificial intelligence value chain. The first is infrastructure, which includes data centers, supercomputers, GPU clusters and programs for access to computing capacity. This category accounts for 59 percent of all initiatives identified.

The second corresponds to models, meaning government efforts to develop or adapt foundation models, which account for 32 percent. Finally there is data, a category that encompasses initiatives to build national datasets and accounts for just 9 percent of projects.

Latin America: 11 initiatives in eight countries

Within this global landscape, Latin America appears as a region where activity is still incipient, though growing. The index counts 11 sovereign AI initiatives spread across eight Latin American countries, far below Europe, with 79 projects, and Asia-Pacific, with 54, but ahead of Eurasia, with eight; Africa, with seven; and Canada, with three.

This comparison confirms that global investment in sovereign AI remains concentrated especially in the Middle East and East Asia, which together account for more than 80 percent of declared investment. Regions such as Latin America, meanwhile, are moving forward with considerably more modest resources and ambitions.

Among the Latin American projects identified by CNAS are:

  • Mexico-Coatlicue. A supercomputing project for science and government, currently in the announcement phase.
  • Dominican Republic-CEIA-RD. A national artificial intelligence center focused on computing and training, currently in development.
  • Colombia-IA Hecha País. An initiative to develop models for biodiversity, climate and health. It has a model in the pilot phase.
  • Brazil-SoberanIA. A Portuguese-language artificial intelligence project supported by domestic computing capacity, with the model already operational.
  • Brazil-Santos Dumont upgrade. An expansion of computing capacity for artificial intelligence and scientific research. It is operational.
  • Brazil-Jairu. An artificial intelligence research cluster at the University of São Paulo, currently operational.
  • Chile-Latam-GPT. A regional model led by Chile with collaborators from several countries. Its first version has already been released.
  • Paraguay-Yguazú Digital. A computing center for artificial intelligence developed with Taiwan, currently announced.
  • Uruguay-Antel Pando expansion. A project to provide AI computing capacity to governments and companies, currently in the procurement process.
  • Argentina-Clementina XXI + IPAC. A supercomputer and an access program for research. The system is operational and CNAS counts it as two initiatives sharing the same infrastructure.

Latam-GPT

The case of Latam-GPT is particularly illustrative of the challenges facing Latin America. It is a model led by Chile but conceived as a regional effort, with collaborators from different countries. Its approach shows how the scale needed to develop certain artificial intelligence capabilities can lead Latin American countries to seek regional partnerships instead of betting exclusively on national developments. Cooperation makes it possible to share resources, knowledge, data and technological capacity in a sector where the infrastructure needed to train and deploy advanced models requires ever-larger investments.

One of the index's main contributions is to show that the rhetoric of technological independence does not always translate into real autonomy. According to CNAS, more than three-fifths of the projects tracked worldwide declare at least one foreign partner, and four out of five of those partners are US companies.

The dependence is especially visible in infrastructure. NVIDIA, for example, supplies the GPUs used in 45 percent of all infrastructure projects registered globally.

Dependence

Dependence also appears in the model layer. Among projects that rely on open-weight models, Meta's Llama family is by far the most used, followed by Mistral, from France; Google, from the United States; and Alibaba, from China.

The report itself concludes that, even in the model layer, building sovereign AI usually rests on US technological foundations. In this way, a country can develop a national model, train it with its own data, adapt it to its language and deploy it within its borders, but still depend on chips, architectures, software or models developed by foreign companies.

No country can achieve full sovereignty

For the report's authors, no country, not even the United States, can achieve total control over all the complex inputs needed to make frontier artificial intelligence possible. Even the most technologically advanced economies depend on chips and manufacturing equipment produced abroad.

For this reason, CNAS argues that, for most economies, the real question is not whether they can achieve full sovereignty in artificial intelligence, something the authors consider unfeasible, but how to manage their technological dependencies.

This means choosing which layers of the technology chain are strategic for each country and accepting, at the same time, a certain degree of dependence on external suppliers in those areas where developing their own capabilities would be too costly or complex.

Sovereignty in artificial intelligence thus appears less as a situation of absolute independence and more as a strategy aimed at increasing states' decision-making capacity within a deeply internationalized technology chain.

What it means for Latin America

For Latin America, the index offers two complementary readings. On the one hand, it confirms that the region has begun to move. The governments of Brazil, Chile, Colombia, Mexico, Argentina, Uruguay, Paraguay and the Dominican Republic already have concrete initiatives related to computing capacity, language models or both. Many of them are also geared toward the specific needs of the countries and the region, such as language preservation, public health, scientific research, biodiversity or environmental management.

On the other hand, the volume and scale of these projects remain modest compared with those developed in the Middle East or East Asia, reflecting existing limitations in financing, energy infrastructure and installed computing capacity.

The CNAS map thus shows a Latin America that is beginning to build its own artificial intelligence capabilities, but doing so within a highly concentrated global technological infrastructure. The report makes clear that, rather than aspiring to total self-sufficiency, the challenge for Latin American countries, as for the rest of the world, will be to design partnerships, deploy technologies and build governance frameworks that allow them to exercise significant decision-making capacity over how artificial intelligence is used within their borders, even when much of the underlying technological infrastructure continues to come from abroad.

Alfredo Collosa

Author

Alfredo Collosa

Alfredo Collosa is a tax specialist. Public accountant. Bachelor's degree in Business Administration. Official Master's in Public Finance and Tax Administration from UNED/IEF in Spain. AMLCA Anti-Money Laundering Certificate (Florida University, US) and member of MetaLaw

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading