| Subindex | Rank | Country | AI sovereignty score | Compute and cloud agency | Data and knowledge agency | Model, innovation, and application agency | Talent and absorptive capacity | Institutional and regulatory agency | Legitimacy, rights, and sustainability |
|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | Chile | 63.8 | 37.2 | 65.8 | 47.5 | 65.3 | 79.0 | 88.2 |
| 1 | 2 | Brasil | 63.3 | 58.0 | 65.0 | 49.1 | 28.3 | 92.7 | 86.9 |
| 2 | 3 | Uruguay | 62.0 | 50.3 | 61.9 | 43.8 | 55.4 | 82.3 | 78.5 |
| 3 | 4 | Costa Rica | 50.9 | 37.8 | 39.7 | 25.7 | 48.2 | 77.3 | 76.8 |
| 4 | 5 | Colombia | 50.8 | 22.5 | 60.0 | 28.9 | 41.5 | 76.2 | 75.9 |
| 5 | 6 | Argentina | 47.4 | 34.9 | 58.6 | 48.8 | 35.4 | 66.3 | 40.6 |
| 6 | 7 | Perú | 45.1 | 10.9 | 56.3 | 26.2 | 33.5 | 74.5 | 69.1 |
| 7 | 8 | República Dominicana | 41.0 | 13.2 | 49.3 | 25.4 | 27.0 | 69.9 | 61.5 |
| 8 | 9 | México | 35.4 | 14.5 | 63.0 | 35.3 | 33.0 | 42.9 | 23.8 |
| 9 | 10 | Ecuador | 32.5 | 14.2 | 58.9 | 23.4 | 30.3 | 42.0 | 26.0 |
| 10 | 11 | Panamá | 30.4 | 21.5 | 52.6 | 18.0 | 26.5 | 39.3 | 24.2 |
| 11 | 12 | Paraguay | 27.3 | 13.9 | 51.5 | 22.5 | 28.9 | 26.4 | 21.0 |
| 12 | 13 | Cuba | 27.3 | 3.4 | 16.1 | 38.5 | 21.1 | 61.7 | 23.0 |
| 13 | 14 | El Salvador | 25.0 | 19.1 | 25.8 | 26.8 | 24.5 | 34.4 | 19.5 |
| 14 | 15 | Jamaica | 21.9 | 5.7 | 39.2 | 15.3 | 29.7 | 32.2 | 9.3 |
| 15 | 16 | Guatemala | 21.5 | 15.1 | 42.4 | 16.7 | 21.9 | 15.1 | 17.9 |
| 16 | 17 | Honduras | 20.5 | 11.4 | 36.8 | 21.4 | 23.3 | 17.0 | 13.1 |
| 17 | 18 | Bolivia | 20.0 | 17.0 | 27.8 | 23.9 | 23.5 | 17.1 | 10.9 |
| 18 | 19 | Venezuela | 19.8 | 4.3 | 36.3 | 25.5 | 21.9 | 15.2 | 15.7 |
1 Introduction
AI sovereignty has become an attractive but unstable policy term. It is used to describe national control over data centers, local foundation models, data governance, regulatory authority, public-sector resilience, cultural autonomy, industrial policy, and democratic accountability. The literature reviewed here suggests that this instability is not merely a defect to be cleaned up. It is part of the object: AI sovereignty names a struggle over who can shape AI systems, under what dependencies, and with what forms of legitimacy.
For an ILIA-oriented operationalization in Latin America and the Caribbean (LAC), the central implication is that AI sovereignty should not be treated as autarky, branding, or the simple presence of domestic AI assets. It is better understood as effective agency under interdependence: the capacity of states, publics, institutions, communities, and regional coalitions to access, govern, adapt, contest, and benefit from AI systems across a layered technical and political stack. This review therefore asks how the existing literature can be translated into indicators that distinguish formal control from real agency.
2 Corpus and Method
The review draws on an 88-source BibTeX corpus covering AI sovereignty, digital sovereignty, data sovereignty, compute sovereignty, AI readiness, regional development, and critical data studies. A source-synthesis workflow generated 147 structured synthesis units from the machine-readable corpus, including one manually completed synthesis for the AI Index 2026 policy chapter after the automated LLM pass failed with an SSL error. The curation process then grouped sources into included, peripheral, and discarded categories in review-state/corpus-outlook.yaml.
The synthesis uses the configured lens of “ILIA-oriented AI sovereignty operationalization.” Sources were prioritized when they offered one of five things: conceptual boundaries, measurement logic, observable indicators, LAC-specific evidence, or warnings about dependency, legitimacy, coloniality, democratic risk, labor, or environmental cost. The empirical extension added in this version uses the ILIA 2025 country-level workbook (data/ILIA 2025/Datos ILIA 2025.xlsx) and the ILIA 2026 subindicator tracking workbook (data/ILIA 2026/Subindicadores ILIA 2026.xlsx). The 2025 workbook provides country scores; the 2026 workbook is used as a measurement-design source because it documents continuity, adjustments, and new subindicators, but it does not provide comparable country scores in the current repository snapshot.
3 Conceptual Findings
3.1 Sovereignty Is Relational, Not Autarkic
The strongest conceptual convergence is negative: AI sovereignty should not mean complete self-sufficiency. Digital sovereignty literature already warns that sovereignty in the digital domain is plural, contested, and claimed by many actors, including states, Indigenous communities, social movements, and individuals (Couture and Toupin 2019; Pohle and Thiel 2020). Repetto’s review sharpens this point: the proliferation of “sovereigns” in the digital sovereignty literature is not only conceptual stretching, but evidence that sovereignty is increasingly mediated through technical infrastructures and relationships among actors (Repetto 2025).
In AI specifically, this means that “sovereignty” cannot be scored by asking whether a country owns every layer of the stack. Barasa et al. frame sovereignty as a continuum of strategic postures: states may seek to control some layers, steer others through regulation or procurement, and depend on external providers where full control is impossible or undesirable (Barasa et al. 2026). The TPDi agency framing makes a similar move by shifting attention from sovereignty as a status to agency as a combination of access, control, choice, and leverage over AI capabilities (Hawkins, Razavi, et al. 2025). This is especially important for LAC, where full-stack self-sufficiency would be unrealistic for most countries.
3.2 Control Is Not Enough
Roberts’ normative account is essential for avoiding a common measurement trap. A country or firm can have descriptive control over AI infrastructure without legitimate authority over how that control is exercised (Roberts 2024). For ILIA, this implies that sovereignty indicators should not reward control alone. They should also ask whether control is democratically governed, publicly accountable, contestable, and oriented toward social benefit. Santaniello adds a second caution: digital sovereignty often functions as a discursive instrument marked by adversariality, multiversity, latency, instrumentality, and hypocrisy (Santaniello 2025). Declarations of sovereignty may therefore coexist with deep dependency.
The same point appears in regional and critical literature. Mügge shows that AI sovereignty agendas can prioritize state jurisdiction and competitiveness while externalizing harms to the Global South (Mügge 2024). Lehuedé contrasts state and market sovereignty projects with Latin American grassroots digital sovereignty, arguing that only some sovereignty projects break with extractive logics (Lehuedé 2024). For ILIA, this means that “sovereignty” must be audited for its beneficiaries: sovereignty for whom, over what, and at whose cost?
3.3 Dependency Operates Through the Stack
Compute and infrastructure sources translate the abstract sovereignty debate into observable mechanisms. Hawkins et al. separate compute sovereignty into at least three layers: data center location, operator nationality, and accelerator-vendor nationality (Hawkins, Lehdonvirta, et al. 2025). Lehdonvirta et al. provide a directly usable proxy for domestic public cloud AI compute availability, distinguishing no public cloud AI compute, inference-relevant compute, and training-relevant compute (Lehdonvirta et al. 2025). Richardson et al. add that data centers located domestically may still be subject to foreign jurisdiction if they are operated by foreign firms (Richardson et al. 2025).
This creates an important ILIA distinction: local presence is not the same as effective control. An indicator that counts “data centers in country” would overstate sovereignty if it ignores ownership, operator jurisdiction, accelerator supply, cloud service terms, energy and water constraints, and public procurement dependence.
3.4 LAC Sovereignty Is a Development and Capacity Problem
The LAC-specific literature links AI sovereignty to structural development constraints. CEPAL frames digital transformation and AI as tools for escaping development traps, but only if they are coupled with institutional capacity, governance, and regional cooperation (NU. CEPAL 2025, 2018). Farca et al. emphasize infrastructure, cybersecurity, human capital, sustainability, and shared regional models as conditions for AI development in LAC (Farca et al. 2026). Jung and Katz estimate that AI’s economic gains in the region depend strongly on skilled labor and intangible investment, both of which are regional bottlenecks (Jung and Katz 2026).
Brazil provides the densest country case. Belli’s KASE framework and Brazil stack analysis provide a sovereignty-oriented structure that includes connectivity, data, compute, talent, governance, and regulatory capacity (Belli 2025; Belli and Gaspar 2025). Neto et al. show that Brazil’s 2024 AI plan moves toward AI sovereignty through data, algorithms, and compute investments, but remains vulnerable on connectivity, regulation, and cybersecurity (Johansson Neto et al. 2024). Grohmann et al. add that sovereignty debates in Brazil cannot be separated from labor, class, and transnational platform power (Grohmann et al. 2025).
4 Toward an ILIA-Oriented Operationalization
The review suggests that ILIA should add an AI sovereignty module or cross-cutting sovereignty lens rather than simply rename existing readiness dimensions. The module should measure effective agency under interdependence. It should combine capability indicators, dependency indicators, governance indicators, and normative risk indicators.
| Dimension | What It Measures | Candidate Indicators | Key Sources |
|---|---|---|---|
| Stack-layer capability | Whether a country has usable capacity across AI infrastructure layers | domestic cloud AI compute tier; data center capacity; accelerator access; public-sector compute access; local model or adaptation capacity | (Lehdonvirta et al. 2025; Hawkins, Lehdonvirta, et al. 2025; OECD 2023; Richardson et al. 2025) |
| Effective control and jurisdiction | Whether capacity is locally governable rather than merely located locally | operator nationality; applicable foreign jurisdiction; public procurement dependency; contractual audit rights; data residency plus data access safeguards | (Hawkins, Lehdonvirta, et al. 2025; Richardson et al. 2025; Roberts 2024) |
| Data and knowledge sovereignty | Whether local data, languages, and knowledge systems can shape AI systems | local datasets; Indigenous/community data governance; local language resources; public data infrastructure; data interoperability | (Hummel et al. 2021; Hutchinson et al. 2024; Kwok 2026; Couture and Toupin 2019) |
| Institutional agency | Whether public institutions can steer, contest, and procure AI systems | AI procurement standards; technical audit capacity; regulator staffing; algorithmic transparency registers; public-sector redundancy | (Stix 2022; Ferrari 2024; Langer and Haag 2026; NU. CEPAL 2025) |
| Regional leverage | Whether countries can act through regional coordination rather than isolated national capacity | shared compute facilities; regional standards coordination; cross-border data governance; Caribbean resilience mechanisms; joint procurement | (Farca et al. 2026; Greene and Herakleous 2025; Zaballos et al. 2025) |
| Economic and talent capacity | Whether societies can absorb and shape AI economically | AI investment share; skilled labor and tertiary education; R&D capacity; domestic AI firms; economic complexity | (Jung and Katz 2026; Cazzaniga 2024; Rodríguez Pita et al. 2025) |
| Democratic legitimacy and risk | Whether sovereignty projects increase accountable agency rather than surveillance or capture | independent oversight; auditability; public participation; labor protections; environmental reporting; documented rights safeguards | (Roberts 2024; Mügge 2024; Grohmann et al. 2025; Lehuedé 2024) |
This table implies three scoring principles. First, score access, control, choice, and leverage separately. A country may have access to hyperscaler AI compute but little control over terms, weak choice among providers, and limited leverage in procurement. Treating these as distinct sub-indicators prevents inflated sovereignty scores. Second, distinguish domestic presence from domestic agency. Local data centers, cloud regions, or AI labs should receive higher scores only when they are accompanied by enforceable jurisdictional safeguards, operator diversity, public audit rights, environmental transparency, and local skill formation. Third, score sovereignty as a portfolio. For many LAC countries, sovereignty may be best achieved through regional pooling, public procurement rules, local datasets, sector-specific models, and negotiated interdependence rather than national foundation-model races.
5 Operationalization Based on ILIA Data
The theoretical architecture above is broader than what can be measured immediately with the ILIA workbooks in this repository. The feasible index should therefore be read as a first empirical translation, not a complete measure of AI sovereignty. It uses ILIA 2025 country-level subindicator scores where they already exist and uses the ILIA 2026 tracking workbook to identify where measurement is improving. This distinction matters because some of the most important theoretical variables, such as operator nationality, foreign jurisdiction over cloud providers, contractual audit rights, zero-rating dependence, or public procurement lock-in, are not yet directly measured in the available ILIA data.
The resulting operationalization maps seven theoretical dimensions into six measured subindices. “Effective control and jurisdiction” is not dropped; it is treated as a partially observed dimension proxied through compute, cloud, cybersecurity, data protection, and international standards indicators. This is analytically conservative: the index can identify capacity and institutional readiness for sovereignty, but it should not be interpreted as a direct measure of legal or corporate control over the AI stack.
| Theoretical dimension | Measured ILIA 2025 subindicators used now | ILIA 2026 measurement status | Justification and literature |
|---|---|---|---|
| Compute and cloud agency | Nube; Capacidad de infraestructuras de HPC; Número de GPUs; Capacidad de GPUs; Centro de Datos Certificados; IXP; Servidores de Internet Seguros | Continuity for HPC, GPUs, cloud, certified data centers, IXP, and secure servers; new or adjusted 2026 variables add AI processing data centers, data center energy consumption, advanced storage, and an AI basic basket | Compute sovereignty is measurable through domestic cloud and accelerator availability, but exact control requires attention to provider and chip-vendor dependence (Lehdonvirta et al. 2025; Hawkins, Lehdonvirta, et al. 2025; Richardson et al. 2025). |
| Data and knowledge agency | Disponibilidad Barómetro de Datos; Capacidad Barómetro de Datos; Gobernanza Barómetro de Datos | 2026 retains Barómetro de Datos and adds Statistical Performance Indicator variables for data use, services, products, sources, and data infrastructure | Data sovereignty is not only data protection; it concerns the capacity to produce, govern, and reuse local data infrastructures (Hummel et al. 2021; Hutchinson et al. 2024; Kwok 2026). |
| Model, innovation, and application agency | Productividad Open Source; Calidad Open Source; Cantidad de Patentes; Proporción de Desarrolladores de Software; Relevancia de Producción de Software; Número de Inversiones Privadas; Valor Total Estimado Inversiones Privadas; Empresas de IA; Gasto en I+D; Desarrollo de Aplicaciones | 2026 keeps software, investment, firm, and application indicators while refining adoption and AI application variables | Sovereignty requires the capacity to adapt and build systems, not only consume external models. Open source, domestic firms, applied development, and R&D capacity are practical proxies for agency (Belli 2025; Hawkins, Razavi, et al. 2025; Rodríguez Pita et al. 2025). |
| Talent and absorptive capacity | Educación Temprana en Ciencias; Educación Temprana en IA; Habilidad en Inglés; Concentración Habilidades IA Ingeniería; Licenciados STEM; Demanda Cursos de IA; Ranking QS Magíster; Ranking QS Doctorados; Ranking Acreditadas Magíster; Ranking Acreditadas Doctorados | 2026 retains the core talent structure and adds gender-disaggregated or gender-specific measures for science, English, IA skills, STEM graduates, and IA course demand | Effective agency depends on skilled labor, advanced training, and the ability to absorb AI into productive and public-sector uses (Jung and Katz 2026; Farca et al. 2026; Cazzaniga 2024). |
| Institutional, regulatory, and regional agency | Strategy existence, age, authority, update, evaluation, coordination, infrastructure, data, regional cooperation, institutional existence, ISO participation, AI international agreements, AI risk mitigation, personal-data regulation and enforcement, cybersecurity measures | 2026 refines strategy, implementation, standards, AI regulation, cybersecurity, data protection, and international linkage indicators; adds legal risk classification and regulatory exploration variables | Institutional sovereignty means the ability to steer, audit, coordinate, and enforce. Merely having assets is insufficient without regulation, standards capacity, and public-sector institutions (Stix 2022; Ferrari 2024; Karanicolas 2025; Belli 2025). |
| Legitimacy, rights, and sustainability | Ética y Gobernanza; Participación Ciudadana; Metodología Multistakeholder; Perspectiva de Género; Sostenibilidad en Estrategia; Protección de datos y privacidad; Seguridad, Precisión y Confiabilidad; sustainable data center standards; digital infrastructure sustainability; Sustentabilidad; renewable energy share | 2026 reorganizes this area under responsible, safe, and sustainable AI; it adds AI incidents and annual incident growth while retaining privacy, safety, data-center sustainability, digital infrastructure sustainability, and clean-energy indicators | Sovereignty should be assessed by the quality of agency it creates. Democratic participation, rights safeguards, labor and social concerns, and ecological constraints prevent sovereignty from becoming a purely statist or corporate project (Roberts 2024; Mügge 2024; Grohmann et al. 2025; Lehuedé 2024). |
| Effective control and jurisdiction | No direct country-level ILIA 2025 measure; proxied through cloud/compute, data protection enforcement, cybersecurity, standards, and procurement-relevant institutional variables | 2026 improves infrastructure observability, but operator nationality, foreign jurisdiction, contractual rights, and vendor lock-in remain measurement gaps | This is the largest gap between the literature and the present dataset. Compute located in a country can still be controlled through foreign firms, foreign law, and hardware supply chains (Hawkins, Lehdonvirta, et al. 2025; Richardson et al. 2025; Roberts 2024). |
6 Methodology for Constructing the AI Sovereignty Index
The preliminary index is built from the Puntaje ILIA sheet in the ILIA 2025 data workbook. This sheet is used rather than raw values because its subindicator scores are already harmonized onto the ILIA scoring scale. The analysis does not re-normalize the raw variables and does not add any new imputation beyond what is already embedded in the ILIA score sheet. Missing values, if present after ILIA scoring, are ignored within a subindex mean; a country is retained when it has at least 40 percent of the selected subindicators available for that subindex.
The index has six measured subindices:
- Compute and cloud agency. Physical and network infrastructure for AI compute, cloud, interconnection, and secure digital services.
- Data and knowledge agency. Data availability, capacity, and governance.
- Model, innovation, and application agency. Local software, open-source, patents, AI firms, investment, R&D, and applied AI development.
- Talent and absorptive capacity. Basic, professional, and advanced AI-related human capital.
- Institutional and regulatory agency. Strategy, institutions, standards participation, regulation, personal-data governance, cybersecurity, and international linkage.
- Legitimacy, rights, and sustainability. Public participation, multistakeholder process, gender, privacy, trustworthy AI, data-center sustainability, infrastructure sustainability, and clean energy.
Within each subindex, all selected ILIA 2025 subindicators receive equal weight. The overall preliminary AI sovereignty score is the unweighted average of the six subindices. This equal-weight choice follows two principles. First, the literature treats sovereignty as a portfolio of complementary capacities rather than a single dominant layer (Rodríguez Pita et al. 2025; Belli 2025). Second, the available data do not yet justify a more precise empirical weighting scheme. A later version could test alternative weights, for example emphasizing compute and institutional control, or applying factor analysis once a larger longitudinal dataset exists.
Formally, for country (c), subindex (s), and selected ILIA subindicators (I_s), the subindex is:
[ S_{c,s} = {i I_s(c)} x{c,i} ]
where (x_{c,i}) is the ILIA 2025 score for country (c) on subindicator (i), and (I_s(c)) is the set of non-missing selected subindicators for that country and subindex. The preliminary sovereignty score is:
[ AIS_c = {s=1}^{6} S{c,s} ]
The outputs below should be interpreted as a first diagnostic ranking. They are useful for finding patterns, strengths, and bottlenecks, but they do not yet measure all dependency mechanisms highlighted by the literature. In particular, they under-measure foreign jurisdiction over cloud providers, operator nationality, accelerator supply-chain dependence, public procurement lock-in, and the degree to which communities can contest AI deployments.
7 Preliminary Results
7.1 Country Ranking
The ranking should be read as a composite view of agency-relevant capacities already captured by ILIA. Countries near the top combine several advantages at once: infrastructure and compute, data and innovation capacity, human capital, institutions, regulatory tools, and legitimacy or sustainability safeguards. A country can still rank lower if it has one strong layer but large gaps elsewhere; this is intentional because the literature treats sovereignty as a multi-layered portfolio rather than a single asset.
7.2 Subindex Profile
The heatmap is often more informative than the final score. It shows whether a country’s sovereignty profile is balanced or asymmetric. For example, an economy may have relatively strong governance but weak compute capacity, or good talent indicators but weaker data-governance and infrastructure foundations. These asymmetries are analytically important because they identify where sovereignty-oriented policy could add most value.
7.3 Layer Tensions
The compute-governance comparison directly reflects one of the core findings of the literature: infrastructure without governance does not equal sovereignty, and governance without infrastructure may leave countries dependent on external providers. The most desirable position is the upper-right quadrant, where countries combine material capacity with institutional agency. Countries in the lower-right quadrant may have infrastructure but weaker steering capacity; countries in the upper-left quadrant may have governance ambitions but insufficient stack-layer capability.
The radar chart highlights that high scores can come from different profiles. Some countries may be relatively infrastructure-led, while others score through institutional and regulatory capacity or broader innovation and talent ecosystems. This supports a policy reading in which sovereignty strategies should be tailored to bottlenecks rather than copied from a single national model.
7.4 Coverage Check
| Subindex | Countries | Minimum_coverage | Mean_coverage | Maximum_coverage | |
|---|---|---|---|---|---|
| 0 | Compute and cloud agency | 19 | 100.0 | 100.0 | 100.0 |
| 1 | Data and knowledge agency | 19 | 100.0 | 100.0 | 100.0 |
| 2 | Institutional and regulatory agency | 19 | 100.0 | 100.0 | 100.0 |
| 3 | Legitimacy, rights, and sustainability | 19 | 100.0 | 100.0 | 100.0 |
| 4 | Model, innovation, and application agency | 19 | 100.0 | 100.0 | 100.0 |
| 5 | Talent and absorptive capacity | 19 | 100.0 | 100.0 | 100.0 |
The coverage table checks whether the preliminary ranking is driven by a narrow subset of variables. Because the analysis uses ILIA’s already scored Puntaje ILIA sheet, most missingness has already been handled upstream. Still, coverage remains part of the audit trail: future versions should report whether new ILIA 2026 variables improve the observability of compute, data infrastructure, gender gaps, incidents, and sustainability.
8 What This Adds to Existing Readiness Indices
Existing AI readiness indices are valuable but insufficient for sovereignty. AIPI, for example, captures digital infrastructure, human capital, innovation, and regulation in a way that is useful for adoption readiness (AI Preparedness Index (AIPI), n.d.; Cazzaniga 2024). But readiness is not agency. A country can be ready to adopt AI while remaining dependent on foreign cloud, proprietary models, external standards, and extractive data pipelines.
The sovereignty lens adds four variables that readiness indices usually underplay: dependency depth, contestability, leverage, and normative quality. Dependency depth asks who controls infrastructure, data, models, standards, and procurement terms. Contestability asks whether public institutions and affected communities can audit, challenge, or redirect AI systems. Leverage asks whether countries can bargain, coordinate regionally, or create fallback options. Normative quality asks whether sovereignty increases democratic, social, and ecological agency rather than merely expanding state or corporate power.
This is where critical literature is not ornamental but operational. Data colonialism, decolonial AI, popular digital sovereignty, and infrastructural power name mechanisms that can become indicators: data extraction, epistemic dependence, labor invisibility, community exclusion, infrastructure chokepoints, and environmental externalities (Couldry and Mejias 2019; Mohamed et al. 2020; Grohmann and Costa Barbosa 2026; Ishkhanyan 2026). The preliminary index does not yet capture all these mechanisms, but it creates a reproducible baseline for asking where ILIA already measures sovereignty-relevant capacities and where the next measurement round should add sharper dependency variables.
9 Limitations
The corpus is strong on conceptual, regional, and infrastructure sources, but uneven in three areas. First, many measurement proposals are recent working papers or policy artifacts, so their indicators need validation before being adopted. Second, source-level synthesis produced some malformed trace strings in a minority of files; the final argument therefore relies primarily on audited core syntheses and direct machine-readable source checks. Third, LAC-specific empirical evidence is still concentrated in Brazil, CEPAL regional reports, Caribbean readiness/resilience work, and infrastructure investment studies. More country-level evidence would be needed before producing a full operational dataset.
The empirical index adds two further limitations. The first is construct undercoverage: ILIA 2025 provides strong proxies for capacity, institutions, talent, data governance, and sustainability, but it does not directly measure operator nationality, foreign jurisdiction, accelerator-vendor dependence, cloud contractual terms, public procurement lock-in, or zero-rating dependence. The second is temporal comparability: the 2026 workbook documents a richer measurement design, including several new variables, but country-level 2026 scores are not present in this repository snapshot. The visualizations therefore show a 2025 baseline rather than a 2025-2026 time series.
10 Conclusion
The reviewed literature supports a concrete shift: AI sovereignty for ILIA should be operationalized as effective, legitimate agency across the AI stack under conditions of interdependence. This avoids two weak alternatives: treating sovereignty as total autonomy, which is unrealistic for most LAC countries, or treating it as generic readiness, which misses dependency and power.
The ILIA-based operationalization developed here shows that such a measure can already be built from existing subindicators, provided its limits are explicit. The preliminary index combines compute and cloud agency, data and knowledge agency, model and innovation agency, talent and absorptive capacity, institutional and regulatory agency, and legitimacy, rights, and sustainability. It should be read as a sovereignty-relevant capacity index, not yet as a complete dependency or control index. The next measurement frontier is to add direct variables on cloud ownership, foreign jurisdiction, accelerator dependence, procurement leverage, and community contestability, so that ILIA can distinguish not only who is ready to adopt AI, but who has meaningful agency over the terms of adoption.
11 References
AI Preparedness Index (AIPI). n.d. Https://www.imf.org/external/datamapper/datasets/AIPI; IMF.
Barasa, Hilda, Peichin Tay, Keegan Mcbridge, Alexander Iosad, and Jacob Mökander. 2026. Sovereignty in the Age of AI: Strategic Choices, Structural Dependencies and the Long Game Ahead. Paper. Tony Blair Intitute for Global Change.
Belli, Luca. 2025. “Exploring the Key AI Sovereignty Enablers (KASE) of Brazil, to Build an AI Sovereignty Stack.” In The Quest for AI Sovereignty, Transparency and Accountability, edited by Luca Belli and Walter B. Gaspar. Springer Nature Switzerland. https://doi.org/10.1007/978-3-032-02762-7_2.
Belli, Luca, and Walter B. Gaspar. 2025. “AI Sovereignty, Transparency, and Accountability: An Overview.” In The Quest for AI Sovereignty, Transparency and Accountability, edited by Luca Belli and Walter B. Gaspar. Springer Nature Switzerland. https://doi.org/10.1007/978-3-032-02762-7_1.
Cazzaniga, Mauro. 2024. “Gen-AI.” Staff Discussion Notes 2024 (001): 1. https://doi.org/10.5089/9798400262548.006.
Couldry, Nick, and Ulises A. Mejias. 2019. “Data Colonialism: Rethinking Big Data’s Relation to the Contemporary Subject.” Television & New Media 20 (4): 336–49. https://doi.org/10.1177/1527476418796632.
Couture, Stephane, and Sophie Toupin. 2019. “What Does the Notion of ‘Sovereignty’ Mean When Referring to the Digital?” New Media & Society 21 (10): 2305–22. https://doi.org/10.1177/1461444819865984.
Farca, Alberto, Salma Jalife Villalón, Ricardo Martínezgarza, Pau Puig Gabarró, and Enrique Iglesias Rodriguez. 2026. “Development and Use of Artificial Intelligence in Latin America and the Caribbean.” IDB Publications, ahead of print, May. https://doi.org/10.18235/0014054.
Ferrari, Fabian. 2024. “State Roles in Platform Governance: AI’s Regulatory Geographies.” Competition & Change 28 (2): 340–58. https://doi.org/10.1177/10245294231218335.
Greene, Tira, and Demetris Herakleous. 2025. Data Embassies: An Innovative Approach to Strengthening Digital Resilience in the Caribbean. No. 84445. Naciones Unidas Comisión Económica para América Latina y el Caribe (CEPAL).
Grohmann, Rafael, and Alexandre Costa Barbosa. 2026. “Popular Digital Sovereignty and Digital Real Utopias: Lessons from the Homeless Workers’ Movement in Brazil.” Convergence, May, 13548565261456920. https://doi.org/10.1177/13548565261456920.
Grohmann, Rafael, Edemilson Paraná, Jonas C. L. Valente, and Roseli Figaro. 2025. “AI Policy Debates in Brazil: Struggles over Regulation, Governance and Labour.” Global Political Economy 4 (2): 215–27. https://doi.org/10.1332/26352257Y2025D000000037.
Hawkins, Zoe Jay, Vili Lehdonvirta, and Boxi Wu. 2025. AI Compute Sovereignty: Infrastructure Control Across Territories, Cloud Providers, and Accelerators. {{SSRN Scholarly Paper}} No. 5312977. Social Science Research Network. https://doi.org/10.2139/ssrn.5312977.
Hawkins, Zoe Jay, Rebecca Razavi, Meredith Hodgman, Johanna Weaver, Vili Lehdonvirta, and Mercedes Page. 2025. From AI Sovereignty to AI Agency: Measuring Capability, Agency & Power. Discussion {{Paper}}. Tech Policy Design Institute.
Hummel, Patrik, Matthias Braun, Max Tretter, and Peter Dabrock. 2021. “Data Sovereignty: A Review.” Big Data & Society 8 (1): 2053951720982012. https://doi.org/10.1177/2053951720982012.
Hutchinson, Jonathon, Milica Stilinovic, and Joanne E. Gray. 2024. “Data Sovereignty: The Next Frontier for Internet Policy?” Policy & Internet 16 (1): 6–11. https://doi.org/10.1002/poi3.386.
Ishkhanyan, Artur. 2026. “Digital Colonialism, Techno-Sovereignty, and Infrastructural Power.” Frontiers in Political Science 8 (May). https://doi.org/10.3389/fpos.2026.1811790.
Johansson Neto, Germano P., Viviane Farias da Costa, and Walter Britto Gaspar. 2024. “Brazil’s Artificial Intelligence Plan (PBIA) of 2024: Enabler of AI Sovereignty?” The African Journal of Information and Communication (AJIC), no. 34 (December): 1–15. https://doi.org/10.23962/ajic.i34.20424.
Jung, Juan, and Raúl L. Katz. 2026. Impacto económico de la inteligencia artificial en América Latina: transformación tecnológica y rezago en materia de inversión y capacidades laborales. CEPAL.
Karanicolas, Michael. 2025. “Developing AI Standards That Serve the Majority World.” In The Quest for AI Sovereignty, Transparency and Accountability, edited by Luca Belli and Walter B. Gaspar. Springer Nature Switzerland. https://doi.org/10.1007/978-3-032-02762-7_15.
Kwok, Jenny C. Y. 2026. “Data Colonialism and Indigenous Languages in AI: A Critical Review of Existing Initiatives and Their Struggles with Data Sovereignty.” AI & SOCIETY, ahead of print, May. https://doi.org/10.1007/s00146-026-03091-w.
Langer, Paul F., and Stefan Haag. 2026. “AI Sovereignty: Redundant Use of Large Language Models for Public Sector Resilience.” Public Organization Review, ahead of print, May. https://doi.org/10.1007/s11115-026-01021-4.
Lehdonvirta, Vili, Boxi Wu, Zoe Jay Hawkins, Celine Caira, and Lucia Russo. 2025. “Measuring Domestic Public Cloud Compute Availability for Artificial Intelligence.” OECD Artificial Intelligence Papers, ahead of print, October. https://doi.org/10.1787/8602a322-en.
Lehuedé, Sebastián. 2024. “An Alternative Planetary Future? Digital Sovereignty Frameworks and the Decolonial Option.” Big Data & Society 11 (1): 20539517231221778. https://doi.org/10.1177/20539517231221778.
Mohamed, Shakir, Marie-Therese Png, and William Isaac. 2020. “Decolonial AI: Decolonial Theory as Sociotechnical Foresight in Artificial Intelligence.” Philosophy & Technology 33 (4): 659–84. https://doi.org/10.1007/s13347-020-00405-8.
Mügge, Daniel. 2024. “EU AI Sovereignty: For Whom, to What End, and to Whose Benefit?” Journal of European Public Policy 31 (8): 2200–2225. https://doi.org/10.1080/13501763.2024.2318475.
NU. CEPAL. 2018. Datos, algoritmos y políticas: la redefinición del mundo digital. April.
NU. CEPAL. 2025. Superar las trampas del desarrollo de América Latina y el Caribe en la era digital: el potencial transformador de las tecnologías digitales y la inteligencia artificial. CEPAL.
OECD. 2023. “A Blueprint for Building National Compute Capacity for Artificial Intelligence.” OECD Digital Economy Papers, ahead of print, February. https://doi.org/10.1787/876367e3-en.
Pohle, Julia, and Thorsten Thiel. 2020. Digital Sovereignty. {{SSRN Scholarly Paper}} No. 4081180. Social Science Research Network.
Repetto, Eric. 2025. “Concept ‘Stretching’ or Concept Innovation? A Review of the Usages of Sovereignty in the Digital Sovereignty Literature.” Policy & Internet 17 (3): e70011. https://doi.org/10.1002/poi3.70011.
Richardson, Aris, Haley Yi, Michelle Nie, et al. 2025. How Sovereign Is Sovereign Compute? A Review of 775 Non-U.S. Data Centers. arXiv:2508.00932. arXiv. https://doi.org/10.48550/arXiv.2508.00932.
Roberts, Huw. 2024. “Digital Sovereignty and Artificial Intelligence: A Normative Approach.” Ethics and Information Technology 26 (4): 70. https://doi.org/10.1007/s10676-024-09810-5.
Rodríguez Pita, María del Pilar, Jorge Emiliano Pérez Martínez, and Alberto Urueña López. 2025. “From Concept to Method: A Framework for Measuring Digital Sovereignty.”
Santaniello, Mauro. 2025. “Attributes of Digital Sovereignty: A Conceptual Framework.” Geopolitics 0 (0): 1–22. https://doi.org/10.1080/14650045.2025.2521548.
Stix, Charlotte. 2022. “Foundations for the Future: Institution Building for the Purpose of Artificial Intelligence Governance.” AI and Ethics 2 (3): 463–76. https://doi.org/10.1007/s43681-021-00093-w.
Zaballos, Antonio García, Pau Puig Gabarró, Enrique Iglesias Rodriguez, and Rafael Pérez Colón. 2025. “Capturing the Data Center Opportunity in Latin America and the Caribbean.” IDB Publications, ahead of print, October. https://doi.org/10.18235/0013774.