The AI Urban Exclusion Cycle: smart urbanism and inequality in the Global South

1 Introduction

Artificial intelligence (AI) is increasingly embedded within smart city initiatives across the Global South, where governments, development agencies, and technology companies promote digital technologies as tools for improving urban efficiency, governance, and service delivery. From intelligent transport systems to urban surveillance technologies, smart city infrastructures increasingly rely on real-time data collection and data-driven forms of urban governance (Batty, 2018; Cugurullo et al., 2024; Kitchin, 2016; UN-Habitat, 2024). Across many cities in the Global South, smart city agendas are presented as opportunities to modernize infrastructure, improve sustainability, and enhance economic competitiveness. However, emerging scholarship increasingly argues that smart city models transferred from the Global North often fail to account for the historical, institutional, and socio-spatial realities of these cities (Wernick et al., 2024). Critical scholarship has also questioned approaches to smart city development that overlook the political and spatial dimensions of technological transformation and the role of meaningful citizen engagement, highlighting socio-spatially uneven outcomes and the risks of tokenistic participation (Kapitsinis, 2026; Kummitha, 2025). These concerns are especially relevant to the Global South, where smart urbanism intersects with existing socio-spatial and institutional inequalities. The rapid expansion of AI technologies has intensified concerns about an emerging "AI divide" shaped by unequal access to, use of, and benefits from AI technologies (Hammerschmidt et al., 2025).

At the same time, AI technologies are being introduced into urban contexts already characterized by deep inequalities in housing, infrastructure, connectivity, and access to public services (Sampath, 2021; Das, 2025). Many cities continue to experience uneven urban development characterized by infrastructure inequality, expanding informal settlements, and fragmented systems of urban governance (Roy, 2009; Wernick et al., 2024; Alizadeh and Prasad, 2024). Vulnerable urban communities, including low-income populations, informal settlement residents, migrants, and digitally excluded groups, often remain underrepresented within the datasets and digital systems shaping urban decision-making.

Smart city initiatives are often framed as neutral and data-driven approaches to urban management (Batty, 2018; Kitchin, 2016; Alizadeh and Prasad, 2024; UN-Habitat, 2024). However, AI systems depend on infrastructures and institutional capacities that remain unevenly distributed across many cities (UN-Habitat, 2024; Sampath, 2021; Mhlanga, 2026). As a result, the benefits of AI-driven urban development may become concentrated within already connected and economically advantaged areas, while marginalized communities risk becoming increasingly invisible within planning processes. This concern echoes recent arguments that smart city initiatives may reproduce existing inequalities and further marginalize poorer populations in the Global South (Wernick et al., 2024). Unequal access to connectivity, digital infrastructure, and urban data may therefore reinforce existing forms of spatial exclusion rather than reduce them. Despite growing recognition of these challenges, existing research has generally examined data inequality, uneven infrastructure, and governance limitations as separate concerns rather than as interconnected processes through which AI may reinforce urban inequality. We argue that this fragmented perspective has constrained current debates on AI-enabled urban development by obscuring how these dimensions interact to reproduce urban inequality, particularly across cities in the Global South.

Against this background, the aim of this paper is to develop a conceptual framework for explaining how AI-enabled smart city development may reproduce and intensify existing urban inequalities in the Global South. We propose the AI Urban Exclusion Cycle (Figure 1), which conceptualizes this process through the interaction of three interconnected mechanisms: (1) data inequality and urban invisibility, (2) smart infrastructure and uneven urban development, and (3) AI governance, surveillance, and marginalized communities. These mechanisms translate pre-existing inequalities into forms of spatial exclusion that, in turn, reinforce urban inequality and shape subsequent patterns of data production, infrastructure investment, and AI-enabled governance.

The paper makes two main contributions. First, it builds on critical smart city debates on inequality, participation, infrastructure, and governance and contextualizes them within the distinctive socio-spatial and institutional conditions of cities in the Global South. Second, the AI Urban Exclusion Cycle conceptualizes data inequality, uneven infrastructure, and governance limitations as interconnected and mutually reinforcing mechanisms instead of separate problems. In doing so, the framework shifts attention from whether individual AI technologies are inclusive or exclusionary toward the broader urban processes through which technological development can reproduce spatial inequality over time. The framework therefore provides a conceptual basis for examining how these reinforcing dynamics might be interrupted through representative urban data, equitable infrastructure investment, accountable governance, and meaningful public participation.

The following sections elaborate the three interconnected components of the AI Urban Exclusion Cycle. Section 2 examines data inequality and urban invisibility, Section 3 explores uneven smart infrastructure development, and Section 4 discusses AI governance and surveillance as mechanisms that reinforce spatial exclusion. Section 5 outlines pathways for breaking the cycle through more inclusive and human-centered urban AI, while Section 6 concludes the paper.

2 Data inequality and urban invisibility

We argue that data inequality and urban invisibility constitute the first reinforcing mechanism of the AI Urban Exclusion Cycle, through which pre-existing urban inequalities become embedded in AI-enabled planning and governance. AI systems depend heavily on the collection, classification, and analysis of urban data, while smart city infrastructures increasingly rely on sensor networks, digital platforms, and real-time data systems to support urban services, infrastructure management, and planning (Kitchin, 2014). Consequently, the inclusiveness of AI-driven systems depends fundamentally on the quality and representativeness of urban data (UN-Habitat, 2024). In many cities across the Global South, urban data remain uneven, incomplete, and shaped by longstanding social and spatial inequalities.

Vulnerable urban communities are frequently underrepresented within formal systems of data collection and urban mapping. As Kitchin (2014) argues, big urban data reflects existing technological infrastructures and institutional priorities that determine which populations become visible through data. Informal settlements are often absent or only partially represented within official planning records despite housing large urban populations. Roy (2009) further argues that urban informality is linked to unequal systems of governance and recognition rather than simply poverty. Consequently, marginalized communities may remain partially invisible within datasets increasingly used to guide urban governance.

Where data gaps exist, vulnerable populations risk being overlooked in urban decision-making and infrastructure investment. Elwood and Leszczynski (2013) note that geospatial technologies shape which forms of knowledge become visible within digital systems. In this context, the absence of reliable data is not merely a technical limitation but also a political condition that determines which places and populations become visible within AI-enabled urban governance. Populations excluded from datasets may also face institutional invisibility, particularly where access to services increasingly depends on digital identification and data-driven targeting (UNDP, 2025; Eubanks, 2018).

We contend that mapping practices represent one of the principal mechanisms through which data inequality is translated into spatial exclusion. Wealthier districts are more likely to generate the high-quality data required for AI-driven management systems, whereas low-income and peripheral areas often remain less visible due to limited connectivity and weaker institutional presence. As a result, AI systems may unintentionally reinforce uneven urban development by privileging already connected populations.

Digital inequality further deepens these challenges. Unequal access to smartphones, internet connectivity, and digital literacy continues to shape participation in AI systems across many developing contexts (UNDP, 2025). In cities where digital participation increasingly shapes access to services and representation, unequal technological access risks creating new forms of exclusion layered upon existing socio-spatial inequalities.

We therefore suggest that debates on urban AI should move beyond questions of data availability alone. Instead, data inequality should be understood as a structural mechanism through which AI-enabled planning reproduces spatial exclusion. As AI becomes more integrated into urban governance and development practice, questions regarding who is visible within datasets, who is excluded, and whose realities are reflected within AI-driven planning become increasingly important for achieving more equitable urban development outcomes.

3 Smart infrastructure and uneven urban development

We argue that uneven smart infrastructure development constitutes the second reinforcing mechanism of the AI Urban Exclusion Cycle. While the first mechanism explains how pre-existing urban inequalities enter AI-enabled systems through unequal data, the second shows how these inequalities become embedded within the spatial distribution of digital infrastructure and technological investment. Across many cities in the Global South, governments and development agencies increasingly invest in AI-enabled infrastructures such as smart mobility systems, digital public services, automated traffic management, and urban monitoring platforms. In this context, AI is increasingly positioned as part of the emerging infrastructure of urban governance, raising important questions regarding unequal urban access to digital capacity and technological resources (UNDP, 2025).

However, the distribution of smart urban infrastructure is rarely spatially neutral. AI-driven systems often develop unevenly across cities, concentrating in areas with stronger connectivity, greater institutional investment, and higher economic value. Affluent districts and newly developed urban zones are frequently prioritized for the deployment of smart technologies and communication infrastructure (Wernick et al., 2024). By contrast, low-income neighborhoods and peripheral urban areas may remain weakly connected to the digital infrastructures increasingly shaping urban life. In such contexts, smart urban development may contribute to forms of "unequal abundance," where the benefits of digital transformation remain concentrated within already connected populations while marginalized communities experience fewer gains from digital development (UNDP, 2025). We therefore contend that infrastructure investment should not be viewed simply as a technological decision but as an urban process that shapes who benefits from AI-enabled development.

Smart infrastructure depends heavily on electricity, internet connectivity, digital devices, geospatial data, and institutional capacity, all of which remain unevenly distributed across many cities (Almulhim, 2025; Das, 2025). Existing infrastructure deficits therefore shape who can meaningfully benefit from AI-driven systems and who remains excluded from them. Such constraints can limit the effective implementation of smart city initiatives, particularly in rapidly urbanizing contexts facing infrastructure strain and limited institutional capacity (Maatallah et al., 2026).

AI-driven transport technologies may improve traffic management and public transport efficiency in highly connected urban districts. Yet underserved communities often continue to face inadequate transport infrastructure, weak digital connectivity, and limited integration into formal transport systems. As a result, the benefits of smart mobility may become concentrated among already connected populations, while vulnerable residents continue to experience exclusion from essential urban services and opportunities.

The concentration of smart infrastructure within economically strategic urban zones also raises important questions regarding development priorities. This pattern reflects broader forms of splintering smart urbanism, where advanced digital infrastructures become concentrated in selected urban enclaves while socio-spatial fragmentation deepens elsewhere (Alizadeh and Prasad, 2024). In some cases, smart city investments are linked more closely to economic competitiveness or global city branding than to the everyday needs of vulnerable urban communities (Datta, 2015). Consequently, technological modernization may become more visible in commercially valuable districts than in areas experiencing housing insecurity, infrastructure deficits, or inadequate public services.

We therefore suggest that debates on AI-enabled urban development should move beyond celebrating technological innovation alone and pay greater attention to the geography of digital infrastructure. Within the AI Urban Exclusion Cycle, uneven smart infrastructure represents the second reinforcing mechanism through which pre-existing urban inequalities become translated into unequal access to urban opportunities, public services, and technological investment. Without more inclusive approaches to infrastructure planning, smart urban development risks reinforcing uneven urban geographies, allowing digitally connected communities to benefit disproportionately from technological investment while marginalized populations remain excluded from emerging forms of urban opportunity and participation. This highlights the need for just smart cities that prioritize socio-spatial justice alongside technological innovation.

4 AI governance, surveillance, and marginalized communities

We argue that AI governance and surveillance constitute the third reinforcing mechanism of the AI Urban Exclusion Cycle. While the first mechanism focuses on unequal urban data and the second on uneven smart infrastructure, the third highlights how governance practices, institutional capacity, and regulatory frameworks shape the ways in which AI systems are deployed, monitored, and experienced across different urban communities, ultimately determining who benefits from AI-enabled urban development and who becomes disproportionately exposed to its risks. Beyond service delivery and infrastructure management, AI technologies are increasingly embedded within systems of urban governance, monitoring, and public security. Across many smart city initiatives, governments and institutions are adopting AI-enabled surveillance technologies, predictive analytics, facial recognition systems, and automated monitoring platforms to support urban management and public safety objectives.

However, the expansion of AI-enabled governance systems also raises important concerns regarding accountability, regulation, and the unequal impacts of surveillance technologies on vulnerable urban populations. Recent scholarship further argues that smart urban governance in the Global South must be understood through questions of rights, inclusion, and socio-spatial justice rather than technological efficiency alone (Alizadeh and Prasad, 2024). Public trust is increasingly central to AI governance, particularly when automated systems influence policing, welfare access, migration management, or public security decisions (UNDP, 2025). Their effectiveness and urban impacts depend fundamentally on governance readiness, including institutional capacity, regulatory frameworks, and the availability of high-quality data (Mhlanga, 2026). In cities characterized by unequal access to rights, services, and formal recognition, data-driven governance systems may disproportionately affect communities already experiencing forms of marginalization or social vulnerability.

AI-driven policing technologies use historical crime data and predictive algorithms to identify locations or populations considered to be at higher risk of criminal activity. While these systems are often promoted as tools for improving public safety and resource allocation, their outcomes depend heavily on the quality and representativeness of the underlying datasets. In contexts where policing practices have historically concentrated on low-income or marginalized neighborhoods, predictive systems may unintentionally reproduce existing patterns of surveillance and institutional attention. Governance practices therefore become a critical mechanism through which existing inequalities are translated into unequal forms of surveillance and institutional oversight. We contend that the central issue is not the technology itself, but the governance arrangements through which AI systems are designed, deployed, and regulated.

Increasingly integrated into transport systems, public infrastructure, border management, and urban surveillance platforms, facial recognition technologies can expand the capacity of governments and institutions to monitor movement and public activity across cities. However, concerns remain regarding data privacy, consent, transparency, and algorithmic bias, particularly for populations already facing legal, economic, or social vulnerability (Almulhim and Yigitcanlar, 2025). Migrants, informal workers, undocumented populations, and low-income residents may be especially exposed to the consequences of inaccurate or unregulated monitoring systems.

This regulatory gap is particularly significant in developing countries, where AI adoption has often advanced more rapidly than institutional capacity, privacy protection, and accountability mechanisms (Sampath, 2021). Legal frameworks governing data protection, algorithmic accountability, public transparency, and digital rights often remain fragmented or underdeveloped. In lower-capacity governance contexts, institutional safeguards may lag behind AI adoption, while institutions may become increasingly dependent on externally developed AI systems and foreign technology providers that local authorities cannot easily audit or regulate (UNDP, 2025). Such dependence may undermine local policy autonomy and limit the contextual relevance of AI-enabled urban governance systems (Mhlanga, 2026). In the absence of robust oversight, vulnerable populations may face heightened exposure to opaque forms of automated decision-making with limited opportunities for participation or accountability.

We therefore suggest that debates on AI-enabled urban governance should move beyond questions of technological capability and pay greater attention to institutional accountability, regulatory capacity, and public oversight. Within the AI Urban Exclusion Cycle, AI governance and surveillance represent the third reinforcing mechanism through which pre-existing inequalities become institutionalized within AI-enabled decision-making and public administration. Without stronger safeguards, human oversight, and context-sensitive implementation, AI-enabled governance systems risk reinforcing unequal relationships between visibility, surveillance, and urban power across cities in the Global South.

5 Breaking the AI Urban Exclusion Cycle: toward inclusive and human-centered urban AI

We argue that breaking the AI Urban Exclusion Cycle requires coordinated interventions across all three reinforcing mechanisms of the framework, extending beyond isolated technological improvements. As AI continues to shape urban governance, infrastructure, and service delivery across cities in the Global South, the central challenge is not whether cities should adopt AI, but how AI-enabled urban systems can be designed and governed to promote spatial justice, social inclusion, and equitable urban development.

While efficiency and automation often dominate smart urban agendas, urban development also involves social relations, informal practices, and unequal experiences of infrastructure that cannot always be fully captured through data-driven systems alone. Participatory planning and community engagement may improve the inclusiveness of AI-enabled urban initiatives by identifying infrastructure gaps, exclusionary practices, and local priorities that remain less visible within large-scale urban datasets (UN-Habitat, 2024). Participatory approaches may also strengthen public trust by allowing communities greater influence over how AI systems are designed and governed within urban environments.

The development of more inclusive datasets is equally important. Expanding the visibility of informal settlements, peripheral communities, migrants, and digitally excluded populations within urban data systems may help reduce some forms of exclusion embedded within AI-driven planning processes. Representative datasets, transparency, and human oversight are increasingly emphasized as important elements of AI governance systems (UNDP, 2025). However, implementing inclusive urban AI remains challenging across many contexts in the Global South where institutional capacity, digital infrastructure, regulatory oversight, and financial resources remain uneven.

Ultimately, we contend that breaking the AI Urban Exclusion Cycle requires moving beyond technological optimization toward integrated approaches that combine representative urban data, equitable infrastructure investment, accountable AI governance, and meaningful public participation. Smart city initiatives that prioritize participation, spatial justice, and equitable access to infrastructure are more likely to contribute to inclusive urban futures than approaches focused primarily on optimization and competitiveness. From this perspective, the future of urban AI in the Global South will depend not only on technological innovation, but also on the willingness of governments and development institutions to interrupt the reinforcing dynamics of exclusion identified in the AI Urban Exclusion Cycle and embed principles of social inclusion within digital urban transformation.

6 Conclusion

We have argued that the AI Urban Exclusion Cycle (Figure 1) provides a useful conceptual framework for understanding how AI-driven smart city initiatives in the Global South may reproduce and intensify urban inequality through the interaction of data inequality, uneven smart infrastructure, and AI governance. These interconnected and mutually reinforcing mechanisms translate pre-existing urban inequalities into spatial exclusion, which in turn reinforces urban inequality and shapes subsequent patterns of urban data, infrastructure investment, and AI-enabled governance.

Importantly, these challenges do not suggest that AI should be rejected within urban development practice. Instead, they highlight the importance of more inclusive and human-centered approaches to urban AI governance. Participatory planning, representative datasets, ethical governance frameworks, and stronger institutional accountability will be essential for ensuring that AI supports equitable urban development without deepening existing socio-spatial divides.

Ultimately, we argue that cities in the Global South cannot be considered truly "smart" if technological innovation is accompanied by the continued exclusion of vulnerable populations. The future of urban AI should therefore be evaluated not only by its capacity to improve efficiency and optimize urban systems, but also by its ability to advance spatial justice, strengthen social inclusion, and support more equitable urban futures. We hope that the AI Urban Exclusion Cycle will stimulate further debate and research on how AI-enabled urban development can be designed, governed, and evaluated in more inclusive and equitable ways across cities in the Global South.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Miguel Valdez, The Open University, United Kingdom

Reviewed by: Nikos Kapitsinis, University of Copenhagen, Denmark