Latin America accounts for 14% of global visits to generative AI web solutions and ranks third worldwide in AI application downloads. That figure sits alongside another: the region attracts 1,12% of global AI investment. The Latin American Artificial Intelligence Index 2025 — produced by CENIA and CEPAL — calls that combination a «rather Latin American paradox». The paradox has a history: the region adopts technologies that others produce. What AI changes is the cost of that pattern. The consequences of operating without capabilities of its own arrive sooner and are harder to reverse. The evidence cut-off is May 2026; the sources are primary — ILIA 2025, LAVCA, BID Lab and CEPAL — cited with their exact figures.
The maturity map · pioneers, adopters and explorers
ILIA 2025 assesses nineteen countries across five dimensions: enablers, adoption, human capital, research and innovation, and governance. The results define three groups, and that classification dismantles the image of a homogeneous «region» with a single problem.
| ILIA 2025 group | Countries | Assessment |
|---|---|---|
| Pioneers · more than 60 points | Chile, Brazil, Uruguay | Established dense ecosystems |
| Adopters | Colombia and Costa Rica among those that narrowed the gap most | Closing the gap with the leaders |
| Explorers | More than a third of the nineteen countries assessed | The next cycle is determined here · narrowing the gap or widening the intra-regional divide |
«Latin America is behind in AI» is too broad a statement to base any decision on. The structural bottlenecks, by contrast, are precise, and all come from the same index (ILIA 2025): 90% of regional supercomputing is concentrated in Brazil; more than half of the countries lack critical infrastructure and advanced training; thirteen of the nineteen do not develop early-stage AI skills; eleven lack related doctoral programmes. Brazil and Mexico account for 68% of the region's researchers and, together with Chile, Argentina and Colombia, around 90% of publications. A region that aspires to sovereign AI with its computing capacity, talent and research residing in two countries has an architecture problem before it has an adoption problem.
Public policies · from document to budget
Between 2023 and 2026, the region moved from building agendas to formalising strategies, and from strategies to debating regulation. The change that matters is a different one: some countries began backing their policies with budgets, interministerial coordination and state monitoring capacity. Announcements stopped being the news; execution started to be.
| Country | Instrument | Status as of May 2026 | Note |
|---|---|---|---|
| Chile | National AI Policy 2021-2030 · risk-based bill (2025) | 177 initiatives coordinated by 14 ministries · 78% implemented or completed as of April 2025 | The regional benchmark for effective coordination |
| Brazil | Brazilian AI Plan 2024-2028 · PL 2338/2023 | Public budget of BRL 23.000 million · active parliamentary debate | The most comprehensive case · strategy, investment, infrastructure and regulation in parallel |
| Colombia | CONPES 4144 (2025) · risk-based bill | Institutionalisation at an unusually rapid pace | Emphasis on bias, privacy and labour market transformation |
| Peru | Law 31814 and Supreme Decree 115-2025-PCM · ENIA 2026-2030 | Law and implementing regulations in force | One of the region's few operational frameworks in force · risk classification, prohibitions, transparency, human oversight |
| Uruguay | National AI Strategy 2024-2030 | Formalised on a legal basis | Aligned with UNESCO, OCDE and the Consejo de Europa Framework Convention · legal coherence beyond its scale |
| Argentina | Interministerial Working Group on AI · sectoral programmes | Modular framework, without a consolidated federal architecture | Recognised scientific strength |
| Mexico | Senate-led initiatives | Institutional assembly, without a comprehensive federal strategy | Market capacity, talent and demand |
CEPAL and OCDE converge on a diagnosis worth citing without softening it: many national strategies lack budgets, monitoring and state capacity for implementation. Eight countries report no AI applications for consultation or the co-creation of public policy (CEPAL, 2025), and government use remains limited, in many cases, to chatbots and lightweight automation. Elegant documents abound. Auditable systems, specialised teams and ex post traceability are scarce.
Three investment tracks worth reading together
AI investment in the region runs along three tracks with distinct dynamics. Mixing them produces misguided diagnoses; reading them together reveals where the real deficit lies.
| Track | Figures and source | Assessment |
|---|---|---|
| Venture capital | USD 4.500 million and 751 transactions in 2024 · USD 1.800 million in the first half of 2025, with 54% at early stages (LAVCA, 2025). Mexican startups surpassed Brazil in funding raised for the first time in fifteen years · nearly 500 startups raised their first round in eighteen months | Healthy pipeline. Substantial scaling requires capital on a different scale |
| Public investment | Brazil · BRL 23.000 million for 2024-2028 · Argentina · R&D&I calls open in 2026 · the rest · figures pending confirmation or not publicly specified | The track where the most strategies lack budgets |
| Hyperscalers | Microsoft · BRL 14.700 million in cloud and AI in Brazil over three years and USD 1.300 million in Mexico · AWS · new region in Chile by the end of 2026 | The underlying computing capacity on which any discussion of sovereign AI depends |
The common denominator across all three tracks is the same gap. 1,12% of investment versus 14% of use (ILIA 2025). Global capital funds regional adoption below its share of use, and that gap is closed through infrastructure, R&D and deliberate public procurement — a scale that venture capital alone cannot reach. Without computing capacity that is owned or accessible at competitive cost, all plans depend on infrastructure controlled by someone else.
Talent · two types of training with very different outcomes
ILIA 2025 distinguishes two simultaneous trends in regional human capital. The distinction has operational relevance for any organisation with AI in production: it shows which profiles are abundant and which will be hard to retain.
| Indicator | Data | Source |
|---|---|---|
| AI tool literacy | Sustained growth · more users, lightweight automation, increasing fluency | ILIA 2025 |
| Advanced specialisation | Gap with the global average has widened since 2022, due to an accelerating loss of specialists | ILIA 2025 |
| Startups using generative AI | 85% | BID Lab, 2025 |
| Startups using predictive technologies | 75% | BID Lab, 2025 |
| Startups that consider the available solutions fully suited to their local needs | 12% | BID Lab, 2025 |
The gap lies in specialisation. Adoption has already happened
The region is producing users of AI tools at a good pace, but few professionals capable of auditing models, assessing systematic biases, designing real technical governance or sustaining frontier research. For those operating AI systems in production, the consequence is direct: talent that can verify that models do what they claim to do — with independent technical rigour, with verification delegated to the provider as the exception rather than the rule — is the scarcest resource and the hardest to retain in the region.
Labour market impact. The mistake of assuming symmetry
BID warns that generative AI models can raise productivity while exacerbating inequalities. The mechanism is well known: those already working in cognitive, digital and better-paid tasks capture the benefits of assisted automation first. Historical regional evidence provides a mirror. The introduction of robots in the United States negatively affected employment and wages in Colombia and Brazil in previous periods, while employment in Mexico increased because of its integration into North American production chains (CEPAL, 2025). AI's impact on employment depends on the country's productive structure and its position in the global value chain, as well as on the technology. It is a design parameter, and governments and businesses must incorporate it into their analysis from the outset.
The 2026-2030 cycle hinges on execution
The region has strategies. What is scarce is what makes them verifiable: budgeted implementation, shared computing infrastructure, quality open data, public procurement that creates early demand for local solutions, training intermediate and advanced talent at scale, and risk-based rules with real auditing, traceability and effective human oversight. None of these elements can be resolved with a document.
The region consumes, integrates and adapts quickly · invests, regulates and trains slowly. The missing critical mass must be built, and the 2026-2030 cycle will show who built it.