Source: Getty
The India AI Impact Summit offers a timely opportunity to experiment with and formalize new models of cooperation.
Across the Global South, governments are reshaping digital strategies to strengthen domestic AI foundations. Increasingly, AI is viewed not as an abstract frontier technology but as a practical tool to accelerate development, improve public service delivery, and strengthen economic competitiveness. From Nairobi to New Delhi, Rio de Janeiro to Jakarta, government administrations are testing AI to boost productivity and expand inclusion.
Yet entry costs remain high. Building AI capabilities requires access to scarce and expensive inputs, including compute infrastructure, high-quality datasets, and specialized technical talent, which remain unevenly distributed globally. Relying on foreign-supplied infrastructure, cloud platforms, and proprietary models presents a strategic concern as AI becomes more central to national economic strategy. These dependencies raise cost and access challenges as well as questions about contextual fit, long-term resilience, and economic autonomy. Some dependencies are inevitable given resource concentration and geopolitical realities. But South-South cooperation provides avenues for countries to extract greater dividends from existing capabilities by pooling resources, sharing implementation lessons, and building more context-appropriate AI systems.
South-South cooperation has long enabled collaboration and technical exchange among developing countries; however, this approach has yet to evolve to be fit for purpose in the AI era.
South-South cooperation has long enabled collaboration and technical exchange among developing countries; however, this approach has yet to evolve to be fit for purpose in the AI era. As the world convenes for the India AI Impact Summit 2026, this article argues for more nimble and targeted South-South AI cooperation. It proposes three practical pathways: anchoring collaboration in contextually fitted national initiatives, creating sustained cooperation “practice spaces,” and pooling and aligning financing mechanisms for shared infrastructure and collective outcomes.
Formal South-South cooperation channels have long provided technical exchange, capacity building, and policy coordination among developing countries. Originally conceived as a vehicle for collective self-reliance, these partnerships sought to help countries navigate shared structural constraints by pooling knowledge and resources. Yet historically, South-South cooperation has produced uneven outcomes. While it has facilitated valuable knowledge transfer, its effectiveness has often depended on sustained financing, institutional alignment, and the ability to adapt transferred models to local sociopolitical, economic, and administrative realities. It is rare that these needs are evenly met.
Brazil’s agricultural cooperation with African countries illustrates both the promise and limits of this model. Beginning in the 1970s and expanding through the 2000s, Brazil partnered with countries including Mozambique, Ghana, and Senegal to transfer tropical agriculture techniques developed during its transformation of the Cerrado savannah. Through bilateral agreements and technical missions led by Brazil’s agricultural research agency EMBRAPA, these partnerships supported the establishment of research centers and helped improve agricultural productivity in some regions.
Yet techniques developed in Brazil’s ecological and institutional context did not always work in African environments. In some cases, the original Cerrado model contributed to environmental strain, including soil degradation and water stress. In response, farmers and researchers in Mozambique and Zambia developed “second generation” methods that modified the Brazilian techniques to be more resource efficient and less water intensive, thereby serving the local context effectively. This experience underscores a central trade-off: while South-South cooperation can accelerate access to technical knowledge, meaningful adaptation often requires domestic capabilities and for resources to be recalibrated to local constraints.
While South-South cooperation can accelerate access to technical knowledge, meaningful adaptation often requires domestic capabilities and for resources to be recalibrated to local constraints.
India’s cooperation with African countries on information and communications technology during the 2000s and 2010s is a similar case. Through multilateral engagement with the African Union and initiatives such as the Pan-African e-Network Project, India supported the rollout of tele-education and telemedicine services linking Indian universities and hospitals with African institutions. These efforts were complemented by training programs, technical assistance, and lines of credit aimed at strengthening digital skills and public administration capacity. Some operational challenges like timeline delays and lack of clarity around roles of participating actors remained alongside success, because the top-down partnership structure caused overly bureaucratic ways of functioning. But ultimately, the partnership contributed to moderate outcomes of technology transfer and strengthened technical ties between India and participating African countries.
These lessons carry direct implications for South-South AI cooperation. AI systems rely on rapidly evolving infrastructure, specialized technical expertise, and continuous iteration. Institutional coordination failures, capacity gaps, and unclear operational ownership are likely to be even more consequential in the AI context.
These constraints do not diminish the relevance of South-South cooperation. Rather, they clarify where and how it can be most effective. South-South cooperation should therefore be understood as a strategic tool to advance peer learning of both successful approaches as well as practical challenges precisely because many countries in the Global South face similar structural limitations. Its success, however, depends on partnership design, institutional readiness, and clear implementation pathways.
The case for South-South cooperation on AI is reinforced by a growing convergence in how countries understand both the opportunities and constraints associated with AI adoption. Despite variation in political, economic, and institutional contexts, national AI strategies across the Global South tend to converge around two shared priorities: economic development and economic security. The first sees AI driving economic development, building on the digital transformation agendas launched in the early 2000s. The second seeks to strengthen economic resilience, expressed through an emphasis on domestic ownership of AI inputs, capability-building, and the reduction of asymmetric dependencies on foreign technology providers.
These priorities are shaping how AI is being developed across the Global South. For many countries, particularly in Africa, AI uptake is being driven by the urgency of job creation and wealth generation for a rapidly growing and politically salient youth population. This has prompted governments to respond through policies and investments that accelerate AI adoption, alongside a strategic push for economic self-determination. Developing select AI applications domestically—rather than relying solely on foreign-made systems—is becoming an important priority, as global models do not always reflect local languages or market contexts, and are vulnerable to disruptions. Exclusive reliance on externally supplied infrastructure and inputs can increase costs and create dependencies that limit long-term national interests.
Developing select AI applications domestically is becoming an important priority, as global models do not always reflect local languages or market contexts, and are vulnerable to disruptions.
In response, governments are emphasizing domestic AI agency, including greater control over data, infrastructure, and governance frameworks. Yet building these capabilities independently is resource-intensive and slow. South-South cooperation enables countries to exchange operational lessons grounded in comparable institutional and resource environments, thereby avoiding some of the costs. Many Global South countries face similar implementation challenges—complex and fragmented data ecosystems, compute gaps, and a need for specialized workforce development—which shape how AI systems can be developed and are deployed. As a result, policy and institutional models developed in peer contexts are often more adaptable than those imported from structurally different environments.
These shared constraints have also influenced the direction of AI development itself. Rather than focusing primarily on frontier model development, many Global South countries are prioritizing “Applied AI” solutions in sectors such as agriculture, healthcare, education, and public services. This emphasis on practical deployment creates opportunities for cooperation around shared use cases, implementation pathways, and governance approaches.
Against this backdrop, and ahead of the India AI Impact Summit, pre-summit dialogue on South-South cooperation has intensified through working groups. Both private and public stakeholders are exploring opportunities to collaborate across key layers of the AI stack particularly around linkages across compute provisioning, governance approaches, and the scaling of AI applications. To move from dialogue to practice, actors in this space must overcome structural limitations and develop clear cooperation pathways. Many cooperation channels remain shaped by historical relationships such as colonial ties, or anchored in diplomatic and development forums that are poorly suited to the iterative, technical, and implementation-driven nature of AI collaboration.
This is compounded by the paucity of sustained practice spaces, or operational environments where policymakers, technologists, researchers, and industry actors can exchange lessons, test models, and refine approaches in practice. Without such platforms, cooperation remains fragmented, with few durable mechanisms for translating shared ambition into coordinated implementation. These gaps expose important questions about how to operationalize South-South cooperation on AI. Do new institutions need to be created, or can existing regional and multilateral mechanisms be adapted to support more technical, delivery-oriented collaboration across regions?
The task ahead, therefore, is to identify clear areas for collaboration across the AI stack: linkages that anchor cooperation in concrete needs and, in doing so, shape the mechanisms, goals, and actors required to move South-South collaboration from dialogue to delivery.
A logical starting point is to identify priorities and bottlenecks across the AI stack, alongside operational models already emerging from the Global South. Examining how countries are financing, procuring, governing, and deploying AI infrastructure provides a foundation for designing cooperation that is both actionable and sustainable.
Access to high-performance compute remains one of the most significant structural constraints facing AI development across the Global South. Compute infrastructure requires substantial upfront capital investment, ongoing operational financing, specialized technical management, and geopolitical leverage—resources that remain highly concentrated among a small number of governments and private sector providers globally.
Access to high-performance compute remains one of the most significant structural constraints facing AI development across the Global South.
India’s IndiaAI Mission provides an incipient model for democratizing access to AI infrastructure, illustrating how strategic public coordination of scarce AI infrastructure can significantly lower entry barriers for innovators and startups while building domestic AI capacity. Under the IndiaAI Mission, the government of India funds access to a public compute poolof approximately 18,000 GPUs. Private cloud and infrastructure providers supply and manage the hardware through competitive tenders, while the government sets the terms of use, deciding who can access the compute and at what price. Capacity is allocated to domestic startups, researchers, and academic institutions at subsidized rates of up to 40 percent, with 100 percent subsidies for startups building foundational AI models. While the program significantly lowers entry barriers, beneficiary firms are required to offer a limited equity stake to the government (approximately 2–4 percent) as part of the participation terms.
This model separates infrastructure ownership from access control. Private domestic firms retain ownership and operational responsibility for the hardware, while the government finances access and manages allocation. This arrangement allows the Indian government to expand domestic compute access without bearing the full capital and operational burden of owning and managing the infrastructure directly. It also ensures that public subsidies generate long-term national returns through both equity participation and domestic capability development.
However, the transferability of this model is constrained by structural realities. India’s ability to implement such a program reflects its relatively large public budget, deep domestic technical talent pool, and greater bargaining leverage with global infrastructure providers. Many smaller or lower-income countries lack the fiscal space, institutional capacity, or market size to independently finance comparable compute subsidy programs.
In such contexts, and for countries with growing AI talent but limited access to high-performance computing, the relevant lesson is the governance structure: public financing used to shape access to privately owned infrastructure, and procurement frameworks designed to align private incentives with public capability-building goals. These elements can be adapted on a large scale (such as across multiple African countries) through regional pooling arrangements, multilateral financing, or shared infrastructure models.
Brazil and South Africa illustrate alternative approaches, where compute infrastructure is anchored more directly in publicly funded research institutions. In Brazil, the federal government finances and owns core national supercomputing infrastructure through public research institutions. Brazil’s AI strategy is built on its research institutions, notably the National Laboratory for Scientific Computing (LNCC), home to the Santos Dumont supercomputer. However, private sector vendors like NVIDIA and Eviden supply advanced chips and technical support. Access is primarily allocated to academic researchers and applied AI model development. This facility is central to Brazil’s sovereign compute capacity, but it also requires sustained public financing for procurement, maintenance, and upgrades, which are costs that can be difficult for smaller economies to sustain independently.
South Africa’s Centre for High Performance Computing (CHPC) operates under a similar public ownership model. Funded and overseen by the Department of Science and Innovation, the CHPC is publicly owned and operated, with access allocated to universities, researchers, and public-interest innovation initiatives. But importantly, South Africa’s model also extends access regionally, serving researchers and institutions across the Southern African Development Community (SADC). This illustrates one potential pathway for smaller economies that cannot independently sustain national supercomputing facilities. Rather than each country replicating infrastructure domestically, they can pool demand and share access through regionally hosted infrastructure.
Data governance is emerging as a critical area for South-South cooperation, particularly for countries seeking to develop AI systems that reflect local languages and institutional contexts. Latin America’s LatAm-GPT initiative illustrates how regional collaboration can expand access to relevant training data while maintaining institutional control over how that data is shared and used.
Coordinated by Chile’s National Center for Artificial Intelligence (CENIA), LatAm-GPT brings together an expanding set of partners of over thirty institutions across Latin America and the Caribbean, and it places shared data-rights priorities at the center of cross-border collaboration. Participating institutions contribute curated national and regional datasets under agreed licensing terms, which define how the data may be accessed and used for research and model development. These agreements typically do not transfer ownership of the underlying data. Instead, contributing institutions retain stewardship while granting limited rights for defined uses, such as training and evaluating language models. This approach allows countries to aggregate and harmonize data from multiple jurisdictions, generating a validated and quality dataset exceeding 8 terabytes that reflects regional linguistic and cultural diversity.
This collaborative data strategy, underscored by shared agreements among public research centers, civil society, and universities, effectively functions as a regional data licensing ecosystem that preserves local language and context-rich content (Spanish, Portuguese, and Indigenous languages such as Quechua and Mapudungun) while ensuring data contributors retain appropriate governance through choices that favor explicit institutional consent, rather than web-scale scraping. LatAm-GPT, unlike global commercial models (developed under a highly competitive lens), seeks to strengthen regional capacity to develop and deploy AI systems suited to support practical AI deployment in domestic public-sector, academic, and sector-specific applications.
While the long-term commercial viability and the performance of regionally developed models remain open questions, these arrangements provide a foundation for building a regionally viable AI product that reflects local languages, knowledge systems, and institutional needs. Their significance lies in demonstrating how cooperative governance can support contextualized AI development, with outcomes to be assessed as the model is rolled out.
Similar frameworks are emerging in Africa, where at the national and regional level countries are developing licensing regimes and legal frameworks to govern the use of local language and knowledge datasets. Kenya has developed the NLP Nwulite Obodo Open Data License while the Masakhane African Languages Hub is a continental community driven initiative to advance the development of datasets to ensure African languages are represented in AI systems. African Voices has curated over 500 unique speech datasets to support automatic speech recognition for low-resource languages to help bridge the data gap.
South Africa’s InkubaLM illustrates how such datasets can support targeted model development. Designed as a compact language model optimized for low-resourced African languages, such as Swahili, isiXhosa, Yoruba, Hausa, and Zulu, InkubaLM prioritizes performance on specific linguistic tasks rather than general-purpose capability. Like other models developed outside major commercial labs, InkubaLM demonstrates that high-impact AI can be developed without requiring the massive compute and data budgets typical of global giants. Its design intentionally prioritizes efficiency and relevance to local needs, enabling natural language tasks such as translation and sentiment analysis in contexts long underserved by mainstream AI. InkubaLM relies on edge computing to perform many of its tasks, thereby offsetting the usual resource burden of larger AI models. This approach signals to Global South governments that context-relevant and resource efficient models can be co-created and deployed within Global South infrastructure settings.
Digital public infrastructure (DPI) can provide a practical foundation for deploying AI in public services, particularly where interoperable digital systems are already in place. At its core, DPI refers to shared digital systems such as digital identity platforms, digital payment services, and data exchange frameworks that enable governments and private actors to deliver services more efficiently. India’s DPI stack, built around digital identity (Aadhaar), real-time payments (UPI), consent-based data exchange, and interoperable APIs, has created common infrastructure that both public agencies and private firms can use.
Because these systems were designed as shared platforms rather than isolated programs, they provide a cohesive base where AI tools can be added for defined tasks such as eligibility checks, service routing, or multilingual interfaces in sectors like health, agriculture, and education. The practical lessons lie in how interoperable standards, clear consent rules, and APIs can lower the cost of building localized AI applications without each agency starting from scratch.
India’s DPI ecosystem reflects over a decade of political and policy ambition, which was necessary for uptake. Several countries across Asia, Africa, and Latin America are now deploying this approach, often with support from multilateral institutions and technical partners, illustrating a practical South-South cooperation pathway for AI deployment.
South-South cooperation can expand access, accelerate learning, and strengthen domestic AI capability, but it should not be regarded as a panacea for countries seeking to buttress their AI ecosystems.
South-South cooperation can expand access, accelerate learning, and strengthen domestic AI capability, but it should not be regarded as a panacea for countries seeking to buttress their AI ecosystems. Its effectiveness depends on how partnerships are designed, financed, and operationalized. The task ahead is to identify the appropriate mechanisms through which countries can operationalize South-South cooperation in the AI value chain.
As countries convene at the India AI Impact Summit 2026, there is a narrow but consequential window to translate this convergence into meaningful action. Three practical approaches stand out.
Where feasible, South-South cooperation should build on existing national and regional initiatives rather than creating new frameworks from scratch. Anchoring collaboration in operational systems allows countries to exchange implementation lessons, adapt governance models, and expand access to infrastructure without replicating the full scale of investment.
A near-term opportunity lies in linking India’s AI Mission with Africa’s efforts, including the Africa Declaration on AI and the emerging work of the Africa AI Council. India’s experience offers practical lessons in structuring public-private compute provisioning, procurement design, and institutional coordination. Collaboration could focus on targeted technical exchanges between IndiaAI implementers and Africa AI Council on compute, enabling African startups and research institutions to access shared or regionally hosted compute.
South-South cooperation requires sustained institutional engagement, and one-off agreements. In many cases, the necessary linkages between institutions do not yet exist and must be established. Policymakers will need to identify which domestic actors are best positioned to engage their counterparts abroad, whether ministries, regulators, publicly funded research institutions, or industry consortia are appropriate, and determine the cadence of engagement. Cooperation may be most effective when anchored at the institutional or technical level, rather than solely through national-level diplomatic channels. Defining clear roles, ownership, and operational mandates will be critical to ensuring partnerships deliver practical outcomes.
One effective approach is to establish issue-specific South-South working groups focused on specific bottlenecks such as compute access, multilingual data, AI literacy, and governance experimentation. These groups should be time-bound, draw participation from countries actively implementing solutions, and be mandated to produce operational outputs. For example, a compute-focused working group could link IndiaAI, Brazil’s public computing institutions, and African regional hosting initiatives, including SADC-linked facilities, to develop shared procurement guidance and access frameworks. A data and language track could connect LatAm-GPT, Bhashini, Kenyan NLP initiatives, and African language-model developers to align licensing approaches and crowd-sourcing methodologies. A governance track could enable regulators from the Global South to jointly test proportionate, risk-based approaches in sandboxed settings, moving beyond consensus governance toward practical interoperability.
Financing will determine whether South-South cooperation can move from isolated pilot programs to sustained AI capability-building that strengthens strategic sovereignty and reduces overreliance. While multilateral development banks and external partners can play a catalytic role, cooperation must first align with the strategic and economic interests of participating countries themselves. As the cases of IndiaAI, Brazil’s supercomputing infrastructure, and South Africa’s CHPC demonstrate, governments are already financing core AI infrastructure to strengthen domestic innovation ecosystems, reduce entry barriers for local researchers and firms, and retain greater control over how AI systems are developed and deployed.
Countries must come to cooperation with defined investment priorities, institutional ownership, and governance frameworks that external partners can reinforce, not substitute for.
For Global South countries, maintaining technological agency will require deliberate national financing strategies. Countries must come to cooperation with defined investment priorities, institutional ownership, and governance frameworks that external partners can reinforce, not substitute for. Without domestic financing commitments, cooperation risks reinforcing structural dependencies rather than reducing them. Ensuring that local private sector actors are integrated as co-investors and implementation partners is equally critical.
Regional cooperation can extend the value of national investments while lowering entry costs for participating countries. Shared compute infrastructure, pooled data resources, and joint research initiatives allow countries to distribute capital costs, reduce duplication, and strengthen bargaining power with global technology providers and development partners. Emerging mechanisms such as the Africa AI Fund provide a vehicle to pool resources, finance shared infrastructure, and anchor AI capability-building at continental scale.
These three recommendations provide practical pathways to operationalize South-South cooperation for AI. Effective collaboration requires sustained resourcing, clear institutional ownership, and fit-for-purpose linkages aligned with shared priorities. The India AI Impact Summit offers a timely opportunity to experiment with and formalize these models, not just through new commitments, but through deliberate coordination that enhances what is already working and imagines creative new pathways where critical linkages are scant.
Carnegie does not take institutional positions on public policy issues; the views represented herein are those of the author(s) and do not necessarily reflect the views of Carnegie, its staff, or its trustees.
Despite the strategic importance of relations between the EU and the African Union, deep divisions remain between the blocs. At their upcoming summit, both partners should strive to build a mutually beneficial cooperation.
Marta Martinelli
Tech giants are increasingly able to wield significant geopolitical influence. To ensure digital sovereignty, governments must insist on transparency and accountability.
Raluca Csernatoni
In competition with China’s Belt and Road Initiative, the Global Gateway strategy needs to find an edge. To better promote its interests through investment, the EU’s offer must become more coherent, transparent, and accountable.
Ceren Ergenc, Chaofan Yu
The second Trump administration has shifted the cornerstones of the liberal international order. How the EU responds will determine not only its global standing but also the very integrity of the European project.
Rym Momtaz
The EU’s pursuit of tech sovereignty has often sidelined the role of democracy in the digital sphere. The union should adopt a tech citizenship strategy that promotes citizen engagement, democratic innovation, and accountability.
Richard Youngs
