Sonatel, Anglia Ruskin University, Internet Society

The Global Dialogue on AI Governance represents a critical opportunity to steer global Artificial Intelligence governance toward a more inclusive and equitable trajectory, than one driven purely by market incentives. This is an essential opportunity, as Artificial Intelligence is not only transforming economies and societies; but it will also fundamentally reshape the architecture of the Internet by altering data flows, interconnection dynamics, infrastructure localization, and the distribution of value within the digital economy. In this context, Least Developed Countries, landlocked developing countries, and underserved regions are at a critical juncture. Without targeted and coordinated action, AI risks not only exacerbating existing inequalities but also creating a new structural divide based on access to computation, data localization, interconnection capacity, and the governance of digital resources.

Safe, secure and trustworthy AI;AI capacity-building;Social, economic, ethical, cultural, linguistic and technical implications of AI;Interoperability of governance approaches

Social, economic, ethical, cultural, linguistic and technical implications of AI AI offers significant opportunities to increase productivity in sectors such as health, agriculture, education and finance. However, these opportunities remain unevenly distributed, as AI systems are largely developed outside local contexts and rely on datasets that do not sufficiently reflect local economic, social, and cultural realities. This creates a disconnect between the use of AI and the creation of value, as technologies are deployed locally while economic and strategic benefits are captured elsewhere. Furthermore, the limited representation of local languages reduces the accessibility and relevance of these technologies for local populations. AI capacity-building This dynamic is further reinforced by a structural pattern of talent mobility, in which highly skilled AI professionals from these regions are increasingly integrated into global technology ecosystems in the global north. While this reflects the quality of local talent, it also highlights a deeper imbalance, as these skills contribute more to global innovation than to the development of solutions tailored to local needs. As a result, the ability of countries to build and sustain their own AI ecosystems is significantly constrained. Safe, secure and trustworthy AI The safety and trustworthiness of AI systems are closely linked to the ability of countries to understand, control, and evaluate them. However, in many developing regions, infrastructures, data, and AI systems are largely externalized, limiting transparency and reducing local capacity for risk assessment and oversight. The externalization of data processing also constrains the ability to adapt AI systems to local contexts and to ensure compliance with national regulatory frameworks Strengthening local capacities for evaluation, supervision, and risk management becomes essential for building trustworthy AI systems that are aligned with local needs and conditions.

Artificial Intelligence is profoundly transforming interconnection dynamics through the emergence of distinct traffic patterns, including large-scale training flows between hyperscale data centers and latency-sensitive inference traffic. This evolution is an emerging issue that increases the importance of proximity between data, computation, and users, while simultaneously reinforcing the concentration of infrastructure in a limited number of global hubs. In this context, a new form of inequality is emerging. This divide goes beyond connectivity and encompasses access to computational resources, the ability to process data locally, positioning within interconnection networks, and the capacity to capture value generated by AI systems. Least Developed Countries are particularly vulnerable to this transformation due to pre-existing structural constraints, including dependence on international transit networks, high bandwidth costs, limited bargaining power, and difficulties in attracting critical infrastructure such as data centers and cloud services. At the same time, the progress achieved over the past decade in localizing Internet traffic through investments in Internet Exchange Points and Content Delivery Networks is now at risk. The increasing reliance on remote computing infrastructure may bypass local interconnection systems, increase dependence on international traffic, and reintroduce asymmetric architectures.

AI governance is intrinsically linked to issues of sovereignty, equity, and power dynamics. The concentration of infrastructure, data, and computational capacity in the hands of a limited number of global players reduces the ability of countries to exercise effective control over their digital resources. In many cases, data is generated locally but processed and exploited elsewhere, limiting opportunities for local value creation. This dynamic is further reinforced by the externalization of computational capacity and the outward mobility of talent, resulting in a situation where data, computation, and skills are progressively displaced from local ecosystems. In this context, countries risk becoming providers of data without being producers of intelligence, raising critical concerns regarding equity, inclusion, and fair participation in the AI economy. To be effective, the Global Dialogue on AI Governance must move beyond high-level principles and provide concrete mechanisms that enable effective implementation in Least Developed Countries and underserved regions. It is essential that the Dialogue clearly reflects that Africa and underserved regions must not be left behind in the global AI transformation. This requires bridging the gap between political commitments and operational capacity by providing practical tools, actionable frameworks, and targeted support mechanisms. Regulators must be supported by integrating AI into decision-making processes, equitable access to and sharing of data must be facilitated, and sustained efforts must be made to support capacity-building and skills development.

Addressing the new challenges posed by AI, discussed above, requires promoting the development of local and regional infrastructure, including data centers, computational capacity, and distributed AI solutions, to reduce dependency on external systems. It also requires adapting interconnection policies to the specific characteristics of AI, ensuring that existing investments in local ecosystems are preserved and strengthened. At the same time, data governance frameworks must be developed to ensure equitable access, responsible use, and fair value distribution. Strengthening human and institutional capacities must be prioritized, alongside the promotion of balanced partnerships that support technology transfer, co-development, and the development of local ecosystems. Artificial intelligence represents a profound transformation of the economic and technical structures of the Internet, while simultaneously redefining global power dynamics. In this context, for Least Developed Countries, landlocked developing countries, and underserved regions, the challenge is no longer limited to access to technology but extends to their ability to participate meaningfully in value creation, to retain control over their data, to develop their infrastructure, and to mobilize their talent.

Current dynamics reveal a structural imbalance in which data is generated locally but processed elsewhere (data controllers and processors in different jurisdictions), computational capacity is concentrated in global hubs, and talent is increasingly integrated into external ecosystems. As a result, data, infrastructure, and human capital are progressively externalized, limiting the ability of countries to build sustainable and autonomous AI ecosystems. In this perspective, the Global Dialogue on AI Governance must recognize that inclusion cannot be achieved without explicitly addressing the geography and topology of AI, interconnection dynamics, and mechanisms of value creation. It is essential that public policies and governance frameworks not only expand access, but also ensure greater localization of processing, equitable use of data, and the development of local human capacities. Otherwise, existing inequalities will not only be reproduced but deeply embedded in the very foundations of the AI economy, with the risk of placing entire regions in a position of long-term structural dependency.

It is essential that Community Groups, NGOs and Academia can find a forum, through the AI Dialogue, where to exchange data and experiences with International organizations, Industry and member states, so that the true international impact of AI dynamics can be measured and appropriate recommendations can be based on unbiased evidence matched with sound measurement practices and sound Econometrics tools.

Structural disadvantages in many regions already exist independently of AI, and AI may further exacerbate them. Some countries face structural disadvantages in the global Internet ecosystem (Gupta et al., 2014; Fanou et al., 2017) due to geography, development levels, and infrastructure, including: a) dependence on transit networks controlled by external operators, b) high costs of international bandwidth, c) limited bargaining power of specific countries and operators in interconnection agreements and d) difficulty attracting data centers, CDNs, cloud and facing barriers in both hard and soft, e.g. digital skills, digital infrastructures. AI traffic patterns may further reinforce these disparities. Regions without such infrastructure are likely to face long-term dependence on distant centers and should be the primary focus and voices to be included in the dialogue. The best way to reach them is to focus on a bottom up approach starting with relevant NGOs, and academia members working in the field with the relevant underserved communities.

Large events are less suited for this type of reach and activities.The best approach would be to focus on localized community-based engagement formats. This approach will more effectively increase participation into the AI Global Dialogue

The Internet has been widely studied through the lenses of interconnection, infrastructure, and economic organization. Early research emphasized peering and transit arrangements as the core mechanisms shaping global connectivity (Internet Society, 2022), often conceptualized through a core-periphery model in which central networks dominate traffic flows while peripheral networks depend on upstream providers (Giovannetti & Sigloch, 2015 and D'Ignazio & Giovannetti, 2017). While AI promises economic and social benefits, it also risks exacerbating existing digital gaps (UNCTAD 2021, OECD 2025). The concentration of AI infrastructure in a few global hubs, leaves underserved and landlocked regions dependent on costly international transit. This could create a new "AI divide," where some countries lack equitable access to AI services and the ability to capture value (IMF, 2024). This brings to the fore the necessity for research that examines how AI-driven technological innovations, economic incentives, and policy frameworks interact to shape Internet interconnection and market power globally, with a particular focus on underserved regions. Particular attention should be given to AI-driven traffic, emerging satellite networks, and hyperscale deployment patterns, as well as the role of international governance frameworks starting by using ITU datasets in evaluating the impact of policy interventions see for example data collected on IXPs: https://beta.datahub.itu.int/data/?e=GRC&i=20375 and on Wholesale Telecom/ICT Services, Price regulation of access to Internet exchange points (IXPs) https://beta.datahub.itu.int/data/?e=GRC&i=100028&s=5363 Examples of policies, practices, platforms, or approaches that offer concrete solutions to addressing AI challenges, also stressing the need to look at the actual topology of interconnection to assess global market power, and way to discuss cost allocation for data unfractured sharing are addressed in ITU (2025) ITU (2021) and ITU (2025, b).