What Does AI Sovereignty Mean for Latin America?

With the rapid advances in AI over the past few years and the changing geopolitical order, the topic of sovereignty has become more relevant than ever. Tech companies and their technologies have moved at an extraordinarily rapid pace—but since they are primarily based in the US and China, the data their models are trained on does not particularly account for the specificities of other countries in the Global Majority, including their languages and cultural nuances. This is not trivial: these models are having a profound impact on societies, and many policymakers and researchers have been thinking about these impacts from a sovereignty perspective.

These tensions show specific characteristics depending on the region and country. In Latin America, in particular, while there may be valuable reasons to aim for AI sovereignty, there may also be constraints that require strategic thinking.

This piece will explore AI sovereignty in Latin America through the case of Latam-GPT, a recent regional multistakeholder initiative that aims at better positioning Latin America in the global discussion about AI governance.

The concept of AI sovereignty

Much of the conversation around AI sovereignty has been centered around the infrastructure and data used to develop foundation models. As a recent report by Stanford's Human-Centered Artificial Intelligence center explains, each layer in the AI technology stack involves differing interpretations of what sovereignty means, as well as different degrees of control.

Yet, as with previous discussions around digital sovereignty, there is no shared definition. And this leads different actors to use the term in different ways, sometimes with no clear goals in mind, and often to describe ideas that are quite different—and incompatible.

Considering the necessary interdependencies that AI involve and the resulting constraints for many countries, some researchers have recently been exploring the more pragmatic and achievable concept of "strategic interdependence": the idea that sovereignty should not be binary—sovereign or not sovereign—and should instead focus on a state's ability to make deliberate decisions about how AI is integrated, governed, and used, in line with its national goals. Similarly, others have developed the concept of "managed interdependence," focusing on strategic alliances and partnerships to reduce risks throughout the AI stack.

The idea of AI sovereignty is obviously appealing. Full autonomy might be safer and more culturally aligned, but it is also extremely costly—and access to the necessary funds and capabilities can be pretty challenging in some regions. As explained in a recent piece in Tech Policy Press, a productive definition of AI sovereignty should start with the question: "What parts of the AI supply chain must a nation own, control, or govern, and what parts can a nation safely partner with, rent, or share?"

A recent initiative in Latin America illustrates exactly these tensions: Latam-GPT.

Latam-GPT: good intentions, unanswered questions

Latam-GPT is a recent initiative led by Chile's National Center for Artificial Intelligence (CENIA) and backed by 65 institutions from 15 countries. Launched in February 2026, aside from the consortium it brought together, the initiative involves the development of a large language model built specifically for Latin America. At its launch event, attended by Chile's president, it was described as "the first model developed by and for Latin America." The project was funded by the Development Bank of Latin America and the Caribbean (CAF, for its acronym in Spanish), the Inter-American Development Bank (IDB), and the Chilean Ministry of Science, Technology, Knowledge and Innovation.

Latam-GPT also focuses on developing local capacities and addressing the regional representation gap. Its stated aim is to enable the region to become an active player in AI development—rather than merely a consumer—to move closer to sovereignty, and to finally have a seat at the table. According to its website, “Latam-GPT represents a concrete step toward greater regional technological autonomy and an informed, situated contribution to the global debate on the future of Artificial Intelligence.” It has also been pointed out that Latam-GPT does not intend to compete with frontier AI labs, but rather aims to produce "a model specific for Latin America and the Caribbean, with the necessary cultural requirements and challenges, such as understanding its dialects, history, and cultural aspects."

The core goal of the initiative, as highlighted on the Latam-GPT website, is to achieve "sovereignty" for the region, offering "an open alternative to the dominance of big tech companies, demonstrating that the region has the necessary capacity and autonomy to develop advanced projects on AI." The project has been largely praised, but has also received some criticism. While it is a valuable initiative, there are f questions worth posing––particularly around the idea of AI sovereignty and the region's goals.

First, while the project's main goal is to achieve sovereignty, from media pieces and interviews with project members it is usually not sufficiently clear what the model actually is. Latam-GPT is not built from scratch: it is based on LLaMA (Meta's open-source LLM) and trained using a continued pre-training (CPT) technique. The initiative also uses Amazon Web Services’ infrastructure. Although these details are available in the website's FAQs, it was barely mentioned during the launch and has not been clearly communicated in interviews with project members. What's more, it was not mentioned at all during the launch that, since Latam-GPT is not currently a chatbot, the team had to develop Copuchat—a public-facing interface designed to collect data to train the model. This chatbot is built on OpenAI's API.

Is all this a problem in itself? Not necessarily, since building a model from scratch is challenging and costly. But it can be misleading when the initiative's main framing and narrative is about sovereignty and its launch is presented as a historical milestone for the region. From all the media coverage and the interviews about the project, the more nuanced goals around sovereignty and the actual role of foreign companies in the project are not visible enough.

Second, the goals of the project deserve more careful definition. According to a piece for Tech Policy Press, for example, the main objective is to demonstrate “...that Latin America can develop technical capabilities in frontier technologies” and that “its value should be measured not against the dominant models, but against the counterfactual of a region that remains purely a consumer of AI technologies developed elsewhere.” If the model cannot compete with frontier AI models, Latin America will still be a consumer of AI technologies, even with Latam-GPT, as people are likely to choose the dominant models.

Latam-GPT website also states that with its model “diverse actors will be able to build their own specific AI solutions, democratizing the access to tools that are crucial for competitiveness and regional development.” The datasets are still not available. But even with access to them, who will have the necessary infrastructure (i.e., compute power, engineering resources, expertise, etc.) to be able to use this training data? Is the region prepared––economically and in terms of infrastructure––to make a good use of these datasets? Is this the type of democratization the region currently needs?

Third, while the collaborative dimension of this initiative is impressive––it has brought together experts and organizations from multiple sectors across 15 countries–– there is a real sustainability question, which others have already raised. CAF, IDB, and the Chilean government contributed funding for the initiative's creation and model development, yet CENIA has not published any updates on how it will be funded going forward or how the collaborative structure will be sustained. This is a valid concern, particularly given the many regional initiatives that launched with momentum and then stalled for lack of funding.

It also remains unclear how the project plans to secure the continued involvement of the contributors whose work is essential to Latam-GPT's development. If people are volunteering their time, how sustainable is that really? And is it the most strategic use of funds and human resources to invest in building a model? Transparency is key—not only to genuinely empower users and civil society organizations in the region but also to attract investment and new partners. What are the initiative's next steps and plans for the future? Will it continue to rely on foreign models? And what kind of sovereignty is it actually aiming for?

All this points to the need for an open regional conversation about what Latin America actually wants to achieve and what is feasible. What layers of the stack are most strategic for Latin America and why? What are the specific realities in each country? Do they all need a shared LLM?

Latin America is not alone in its effort towards sovereignty. Countries such as India and Singapore have developed their own models—but the context, resources, and needs in each case are different, and their goals around "AI sovereignty" vary accordingly. Canada, for instance, is not building a model, but has addressed sovereignty openly in its Sovereign AI Compute Strategy, specifying what needs to be procured locally.

There are two crucial features of the Latam-GPT initiative: its collaborative structure—spanning multiple sectors and countries, with a strong presence of academics, civil society, and governments—and its curated corpus. If Latam-GPT is not designed to compete with AI companies abroad, and CENIA representatives have confirmed that is not the goal, it's worth asking whether there are alternatives to building a model that could leverage the consortium already created. Rather than focusing funds on model development, the consortium could jointly advocate for frontier AI companies to incorporate this corpus when building their models—while continuing to expand and refine it, in partnership with additional organizations across the region. Public-private partnerships could also be an interesting option to explore. Such a consortium could be key to ensuring the corpus is accurate and representative, and to pressing companies abroad––some of which have been expanding to the region––to take the region's cultural nuances seriously.

Before launching new efforts to achieve AI sovereignty, Latin American leaders owe the public the opportunity to engage in a transparent conversation and must be able to answer all the questions above. Clearly defining what AI sovereignty means in the region is essential, alongside a strategic articulation of the specific goals an initiative is trying to achieve. What can the region realistically own? How much agency is feasible? Would the answers differ by country? And what is the most strategic, considering the region's goals?

Clarifying these goals would also help lawmakers and policymakers across the region understand what actually needs to be regulated and why, establishing specific requirements for companies coming to the region, tailored to each layer of the stack. The diverse ecosystem CENIA has started to create with Latam-GPT is a promising space to host this conversation.