Keywords
AI governance; Global South; Distributive justice; Digital sovereignty; AI infrastructure
Document Type
Article
English Abstract
Artificial intelligence (AI) is impacting economic and legal orders, yet its benefits and burdens remain unevenly distributed. This paper asks whether, and under what institutional and material conditions, the Global South can secure equitable participation in the AI economy. It advances a normative claim grounded in distributive justice, drawing on Rawls, the capabilities approach, and TWAIL critiques, arguing that bridging the AI divide is not a matter of charity but a duty of international cooperation under ICESCR Articles 2(1) and 15. Methodologically, the paper combines doctrinal analysis of international human rights law with comparative assessment across six constraint domains: energy, finance, connectivity and compute, governance, climate risk, and geopolitics. Existing literature privileges principle-level commitments such as fairness, rights, and human-centric AI, yet they underspecify how legal obligations translate into operational mechanisms, standards participation, and capacity-building for low- and middle-income states. The contribution is twofold. First, the paper develops an evaluative framework that operationalizes distributive justice into four tests—benefit allocation, cost-bearance, voice in standards, and capacity building—linked to human rights duties of participation, equity, and progressive realization. Second, it presents a policy toolbox, including renewable-aligned data-centre siting, blended finance, phased compute access, open-weight model strategies, and structured engagement in standards bodies. The result is a prioritized roadmap with legal, institutional, and measurable criteria that enable the Global South to translate justice claims into implementable AI capacity.
Citation Information
Effoduh, Jake Okechukwu; Eze, Kingsley; and Ufot, Victor. "Legal Infrastructure for Inclusive Tech Development: Artificial Intelligence in the Global South." The Transnational Human Rights Review 11. (2024) : 23-49. DOI: https://doi.org/10.60082/2563-4631.1110 https://digitalcommons.osgoode.yorku.ca/thr/vol11/iss1/3
References
1 Nour Dados & Raewyn Connell, "The Global South" (2012) 11:1 Contexts 12 https://doi.org/10.1177/1536504212436479
see also Bhupinder S Chimni, "Third World Approaches to International Law: A Manifesto" (2006) 8:1 Intl Community L Rev 3. https://doi.org/10.1163/187197306779173220
2 Danni Yu, Hannah Rosenfeld & Abhishek Gupta, "The 'AI divide' between the Global North and the Global South" World Economic Forum (16 January 2023), online: .
3 Anton Korinek & Joseph E Stiglitz, "Artificial intelligence, globalization, and strategies for economic development" (2021) National Bureau of Economic Research, Working Paper No 28453. https://doi.org/10.3386/w28453
4 International Covenant on Economic, Social and Cultural Rights, 16 December 1996, 993 UNTS 3 (entered into force 23 March 1976, accession by Canada 19 May 1976) [ICESC].
5 OECD, Recommendation of the Council on Artificial Intelligence, Doc No OECD/LEGAL/0049 (2019).
6 UNESCO, Recommendation on the Ethics of Artificial Intelligence, UN Doc SHS/BIO/PI/2021/1 (2021).
7 Universal Declaration of Human Rights, UNGA, 3rd Sess, UN Doc A/810 (1948) GA Res 217A (III).
8 Kinyua Gikunda, "Harnessing artificial intelligence for sustainable agricultural development in Africa: Opportunities, challenges, and impact" (2024) arXiv 1 ; Jeane M Tshimula et al, "Artificial Intelligence for Public Health Surveillance in Africa: Applications and Opportunities" (2024) arXiv 1.
9 Constantine Spandagos, Miguel AT Reaños & Muireann Á Lynch, "Energy poverty prediction and effective targeting for just transitions with machine learning" (2023) Economic & Social Research Institute, Working Paper No 762. https://doi.org/10.1016/j.eneco.2023.107131
10 Roba Alsaigh, Rashid Mehmood & Iyad Katib, "AI Explainability and Governance in Smart Energy Systems: A Review" (2022) arXiv 1. https://doi.org/10.3389/fenrg.2023.1071291
11 Huw Roberts et al, "Global AI Governance: Barriers and Pathways Forward. International Affairs" (2023) 100:3 Intl Affairs 1275. https://doi.org/10.1093/ia/iiae073
12 Hassan Qudrat-Ullah, "A Thematic Review of AI and ML in Sustainable Energy Policies for Developing Nations. Energies" (2025) 18:9 Energies 2239. https://doi.org/10.3390/en18092239
13 For a more detailed conceptual analysis of distributive justice in relation to AI and technological advancement, see John Rawls, A Theory of Justice, revised ed (Cambridge, MA: Harvard University Press, 1999); Virginia Eubanks, "Popular technology: Exploring inequality in the information economy" (2007) 34:2 Science & Public Pol'y 127; Josh Cowls & Luciano Floridi, "Prolegomena to a White Paper on an Ethical Framework for a Good AI Society" (2018) 28:4 Minds & Machines 689.
14 In practical terms, this means the Global South should enjoy a meaningful share of the economic benefits of AI, access to its life-saving applications, and influence over its governance (see Robert Nozick, "Distributive Justice" (1973) 3:1 Philosophy & Public Affairs 45).
15 By one recent assessment, Africa (home to about 18 percent of the world's people) accounts for less than three percent of the global AI market and under five percent of global AI research output since 2014 (see Alexander Tsado, Ahura Al & Celina Lee, "Only Five Percent of Africa's AI Talent Has The Compute Power It Needs" (12 November 2024), online: ).
16 Indeed, some scholars have dubbed the current situation a form of "digital colonialism" or "digital extractivism": where data and resources from the Global South are mined by Global North firms with minimal local benefit. This dynamic echoes past injustices rooted in colonial exploitation and uneven development, suggesting that today's AI divide is not merely a coincidental outcome of market forces, but part of a broader historical pattern of inequity (see Michael Kwet, "Digital Colonialism: US Empire and the New Imperialism in the Global South" (2019) 60:4 Race & Class 3). https://doi.org/10.1177/0306396818823172
17 As of 2022, roughly 600 million Africans lacked access to electricity, representing about 43 percent of the continent's population. This crisis is especially acute in rural areas: only about 30.4 percent of the rural population in Sub-Saharan Africa has electricity access, compared to 98.3 percent in rural South Asia and 96.5 percent in rural Latin America. Even in urban areas, where connectivity is higher (around 80.7 percent in Sub-Saharan Africa's cities), it lags far behind the near-universal urban electrification of the Global North. This electric divide is mirrored by a digital connectivity gap - for example, Africa's internet penetration was just 36 percent in 2021 (up from eight percent in 2011) compared to a global average of over 60 percent. The linkage is direct: inadequate electricity supply stymies internet infrastructure expansion (cell towers, fiber networks, data centers), since these technologies all depend on power. Thus, energy poverty creates a vicious cycle, limiting the reach of both basic connectivity and advanced computing capacity in many Global South regions. (See Jake O Effoduh, "Africa's Energy Poverty in An Artificial Intelligence (AI) World: Struggle for Sustainable Development Goal 7" (2024) 15:3 J Sustainable Development L & Pol'y 32). https://doi.org/10.4314/jsdlp.v15i3.2
18 International Energy Agency et al, Tracking SDG7: The Energy Progress Report (Washington, DC: International Bank for Reconstruction and Development, 2025) at 10-31.
19 Ibid; Euan Graham, Nicolas Fulghum & Katye Altieri, Global Electricity Review (Ember, 2025).
20 Joseph Rand et al, Queued Up: 2024 Edition, Characteristics of Power Plants Seeking Transmission Interconnection As of the End of 2023 (Berkeley Lab, 2024); Reuters. (2025, June 18); Forrest Crellin, "Poor grid planning could shift Europe's data centre geography, report says" (18 June 2025), online: ; Elisabeth Cremona & Pawel Czyzak, Grids for data centres: ambitious grid planning can win Europe's AI race (Ember, 2025).
21 See Beyond Fossil Fuels et al, How Europe's grid are preparing for energy transition (Beyond Fossil Fuels, 2025) and IEA, "Lack of ambition and attention risks making electricity grids the weak link in clean energy transitions" (17 October 2023), online: .
22 See Catherine Clifford, "Why America's outdated energy grid is a climate problem" (17 February 2023), online: .
23 International Energy Agency, Africa Energy Outlook 2022 (International Energy Agency, 2022).
24 Ibid; Simon Mundy, "What Chinese customs data tells you about Africa's energy future" (29 August 2025), online: .
25 We see early signs already: South Africa's largest data center operator is building a 120MW solar farm to power its facilities; Kenya touts its 90% renewable electricity grid as a lure for global cloud providers. These examples show how clean energy can become a selling point to "channel" AI infrastructure to the Global South, serving both development and climate goals. The African Union and organizations like the World Bank have promoted concepts such as "geographical load balancing", where computing tasks (like AI model training) could be scheduled in regions where renewable power is currently available (sunny daytime in the Sahara, windy night in Patagonia, etc.), thereby smoothing demand and reducing costs. While ambitious, such ideas underscore the potential synergy between solving energy poverty and advancing AI capabilities. (See Marianna Budaragina et al, "Powering AI in the Global South" (24 May 2023), online: ).
26 Orrick, "Historic Renewable Energy Power Purchase Agreement: Microsoft and Brookfield Agree on Framework to Deliver Over 10.5 GW Capacity Globally" (2 May 2024), online: .
27 The International Energy Agency estimates that achieving universal electricity access in Africa by 2030 would require connecting 90 million people a year (tripling the current rate) and an annual investment of around USD 25 billion. For context, that is just about 1 percent of global energy investment today (see IEA, "Africa Energy Outlook 2022: Key Findings" (last visited 17 November 2025), online: ).
28 Crellin supra note 20; African Development Bank Group, "Mission 300 Africa Energy Summit: Continent to connect 300 million to electricity by 2030 in new ambitious and collaborative initiative" (27 January 2025), online: .
29 For instance, a tech company might partner with a government to build a solar plant that powers both a data center and the local community, sharing costs and benefits.
30 International Finance Corporation, "Scaling Solar: Two PV plants bring clean energy to more than 500,000 in Senegal" (1 June 2021), online: ; World Bank Group, "Unlocking Low-cost, Large-scale Solar Power in Zambia" (14 May 2019), online: ; U.S. Department of Energy, "Best Practices Guide for Energy-Efficient Data Center Design" (last edited July 2024), online: .
31 In 2024 alone, private investment in AI worldwide exceeded $252 billion, with the lion's share concentrated in the United States and China (see Nestor Maslej et al, Artificial Intelligence Index Report 2025 (Standford CA, Standford University Human Centered Artificial Intelligence, 2025).
32 For instance, one analysis noted that excluding China, only about $1.7 trillion of the projected $15.7 trillion for the AI global economic impact by 2030 will accrue to the entire developing world (see Chinasa T Okolo, "AI in the Global South: Opportunities and Challenges Towards More Inclusive Governance" (1 November 2023), online: ).
33 Ben Mkalama & Stefan Ouma, "Making sense of funding inequalities in the venture capital space: a state of the art review paper with views from Africa" (2025) 23:2 Socio-Economic Rev 1013 https://doi.org/10.1093/ser/mwaf007 ; see also Fei Qin, Tomasz Mickiewicz & Saul Estrin, "Homophily and peer influence in early-stage new venture informal investment" (2022) 59 Small Business Economics 93 https://doi.org/10.1007/s11187-021-00523-3 ; Giancarlo Giudici, Massimiliano Guerini & Cristina Rossi-Lamastra, "Elective affinities: exploring the matching between entrepreneurs and investors in equity crowdfunding" (2020) 15:2 Baltic J Management 183. https://doi.org/10.1108/BJM-08-2019-0287
34 The term "computing desert" has even been used to describe parts of the Global South, highlighting not just the lack of hardware but also the scarcity of investment fueling the AI ecosystem (see Access Partnership, "AI beyond data centers: On-device, on-demand, everywhere" (2025), online: ).
35 We already see large tech companies increasing their footprint in developing regions: IBM has established research labs in Kenya, South Africa, and Brazil; Microsoft has opened an Africa Research Institute in Nairobi and has AI development teams in Nigeria; Google set up i ts irst A frican A I research center i n Accra, Ghana. (See Albert L Botchway, "Google AI Research Center Launches Community Engagement Hub to Drive Local Innovation" (4 September 2025), online: ; Darryl K Taft, "IBM Labs Open in Brazil Kenya" (20 August 2012), online:
36 For example, when Google opened its Ghana AI lab, it spurred local universities to update their curricula and inspired a few Ghanaian venture funds to consider AI startups for the first time (see Yossi Matias, "Google Research enhances its AI growth in Africa" (25 May 2022), online: ). Policymakers should actively court such investments by creating enabling environments, e.g. offering tax incentives for tech R&D centers, ensuring strong internet connectivity in tech parks, and protecting intellectual property.
37 Holger Görg & David Greenaway, "Much Ado About Nothing? Do Domestic Firms Really Benefit from Foreign Direct Investment?" (2004) 19:2 World Bank Research Observer 171. See also Binyam A Demena & Syed M Murshed,"Transmission channels matter: Identifying spillovers from FDI" (2018) 27:7 J Intl Trade & Econ Development 701. https://doi.org/10.1093/wbro/lkh019
38 A salient example is the recent breakthrough by a Chinese startup with an AI model called DeepSeek. In late 2024, DeepSeek announced it had trained its cutting-edge large language model for a cost of only about $6 million, using innovative techniques and a fraction of the computing power that traditionally giant models require. By contrast, Meta's latest large model (LlaMA 3) reportedly cost around $80 million to train, using thousands of top-tier GPUs. In fact, DeepSeek managed with one-eighth the number of advanced GPU chips yet achieved comparable performance on certain tasks. If these claims hold true (and early independent assessments in Nature magazine suggest DeepSeek's performance is indeed impressive for its cost), it heralds a future where the entry barrier for developing competitive AI systems is dramatically lower. (See Elizabeth Gibney, "China's Cheap, Open AI Model DeepSeek Thrills Scientists" (23 January 2025), online: ).
39 Jordan Hoffmann et al, "Training compute-optimal large language models" (2022) arXiv 1.
40 Edward E Hu et al, "LoRA: Low-Rank Adaptation of Large Language Models" (2021) arXiv 1.
41 For instance, Tunisia and Senegal have discussed creating sovereign data centers for government and research data (see Aubra Aubra, Jane Munga & Sharmista Appaya, "From the Margins to the Center: Africa's Role in Shaping AI Governance" (8 November 2024), online: ).
42 André B Luiz, Urs Hölzle & Parthasarathy Ranganathan, The Datacenter as a Computer: Designing Warehouse-Scale Machines, 3rd ed (Morgan & Claypool, 2019).
43 Emma Mawdsley, "From billions to trillions': Financing the SDGs in a world 'beyond aid'" (2018) 8:2 Dialogues in Human Geography 191. https://doi.org/10.1177/2043820618780789
44 Ibid.
45 Mariana Mazzucato, & Caetano CR Penna, "Beyond market failures: the market creating and shaping roles of state investment banks" (2016) 19:4 J Econ Pol'y Reform 305; see also Görg & Greenaway, supra note 37. https://doi.org/10.1080/17487870.2016.1216416
46 The former pertains to the physical and digital infrastructure: high-speed internet networks, modern computing hardware, data centers, and devices. The latter refers to the human capacity: the pool of AI researchers, engineers, and digitally literate citizens.
47 The United States alone has well over 2,500 data centers, including many "hyperscale" facilities, while the entirety of Africa has around 150 data centers. In terms of capacity, Africa accounts for less than one percent of global data center capacity despite having 18 percent of the world's population. Analysts estimate the continent would need at least 700 new data centers to meet demand over the medium term. (See Landry Signé, "Fixing the Global Digital Divide and Digital Access Gap" (5 July 2023), online: ).
48 The Tony Blair Institute noted that the US and China dominate access to the world's most advanced AI chips (GPUs), hosting nearly half of these critical resources between them (see Bridget Boakye et al, "How Leaders in the Global South Can Devise AI Regulation that Enables Innovation" (28 March 2024), online: ).
49 Jeffrey Dean & Luiz A Barroso, "The tail at scale" (2013) 56:2 Communications ACM 74. https://doi.org/10.1145/2408776.2408794
50 For instance, renting GPU time from a European data center from West Africa might be not only slow due to undersea cable bottlenecks (see Intelligensis, "Sinking Signals: A Look at West Africa's Data Blackout" (15 March 2024), online: ), but also several times more expensive relative to local income levels (see GSM Association, "The Mobile Economy Sub-Saharan Africa 2024" (April 2025), online: ).
51 As mentioned earlier, internet access in regions like sub-Saharan Africa, South Asia, and parts of Latin America lags behind global averages. Even where mobile networks exist, usage can be low due to cost barriers - the usage gap (people who live under a mobile internet signal but are not using it) is around 60 percent in Africa, much higher than the global average. The reasons range from high data costs (African consumers pay some of the world's highest prices for data as a proportion of income) to lack of digital literacy. (See Matt Shanahan, "Despite improvements, Sub-Saharan Africa has the widest usage and coverage gaps worldwide" (28 October 2025), online: ; Anne Delaporte, "The state of mobile internet connectivity in Sub-Saharan Africa: why addressing the barriers to mobile internet use matters now more than ever" (28 October 2025), online: ; Daniel G Mahler, Jose Montes & David Newhouse, Internet Access in Sub-Saharan Africa, No. 13 (Washington, DC: World Bank Group, 2019).
52 Jonas Hjort & Jonas Poulsen, "The Arrival of Fast Internet and Employment in Africa" (2019) 109:3 American Econ Rev 1032. https://doi.org/10.1257/aer.20161385
53 For example, the 2Africa subsea cable, funded by a consortium that includes big tech firms, is under deployment to encircle the African continent with high-capacity internet, which should increase Africa's international bandwidth dramatically and reduce costs.
54 In Latin America, rural connectivity is being tackled through innovative means like community networks and satellite broadband (e.g., via SpaceX's Starlink, which has begun servicing parts of Brazil, see Reuters, "Brazil regulator authorizes fresh 7,500 Starlink satellites to operate locally" (8 April 2025), online: ).
55 Mahadev Satyanarayanan, "The emergence of edge computing" (2017) 50:1 Computer 30 https://doi.org/10.1109/MC.2017.9 ; see also Jiasi Chen & Xukan Ran, "Deep Learning With Edge Computing: A Review" (2019)_107:8 Proceedings of the IEEE 1655. https://doi.org/10.1109/JPROC.2019.2921977
56 The United States is home to an estimated 50-60 percent of the world's top AI researchers, with large shares also in Europe and East Asia. Africa, by contrast, has produced a tiny fraction of AI research papers (under 5 percent of global output) and faces a dearth of PhD-level AI professionals. (See Hélène Draux, "Research on artificial intelligence - the global divides" (4 January 2024), online (blog): ).
57 Aubra Anthony, Jane Munga & Sharmista Appaya, "We can't regulate what we don't know: Building government capacity for AI" (8 November 2024), online (blog): .
58 An example is how Rwanda became one of the first countries to use drone delivery at scale for medical supplies, bypassing legacy logistics; similarly, a country could become a pioneer in AI-on-the-edge (deploying AI on solar-powered devices in villages) instead of waiting to build mega data centers.
59 Yuyi Mao et al, "A Survey on Mobile Edge Computing: The Communication Perspective" (2017) 19:4 IEEE Communications Surveys & Tutorials 2322; Chen & Ran supra note 55. https://doi.org/10.1109/COMST.2017.2745201
60 There are promising developments: multinational operators (e.g., Africa Data Centres, Huawei, Orange) have been launching data center campuses in cities like Lagos, Nairobi, Johannesburg, and São Paulo to meet regional demand. Governments can encourage more of this by easing regulations, ensuring reliable electricity for these facilities, and perhaps taking equity stakes to ensure public interests (for instance, mandating that a portion of capacity be reserved for educational or governmental use at discounted rates).
61 Satyanarayanan supra note 55.
62 One initiative is MASakhane, an open research collective of African AI researchers working on natural language processing for African languages. Lacking big funding, they share datasets and models openly and crowdsource contributions - a model of bottom-up innovation. Such community-driven projects, while not a substitute for industrial-scale investment, help bridge gaps in areas the big players ignore (e.g., creating speech recognition for Zulu or Yoruba, which big tech has little commercial incentive to perfect).
63 For example, Canada's Vector Institute and some African universities have joint programs to train AI students; and pan-Latin American collaborations are forming for AI in Spanish and Portuguese contexts.
64 Karim R Lakhani & Eric von Hippel, "How open-source software works: "Free" user-to-user assistance" (2003) 32:6 Research Pol'y 923. https://doi.org/10.1016/S0048-7333(02)00095-1
65 Some African countries have recognized this and started integrating digital skills and even coding/AI basics into their school curricula (Kenya's Digital Economy Blueprint is an example, aiming to teach coding and digital citizenship in K-12). (See Kenya Ministry of ICT, Kenya Digital Economy Blueprint (2019)).
66 Here, the private sector and civil society can help: Google's programs have reportedly trained nearly 8 million people in Latin America in digital skills since 2017, and initiatives like AI Saturdays - a volunteer-led global program of free AI workshops on weekends - have chapters in many developing cities (see Mundy supra note 24).
67 There are success stories to emulate: Canada's AI capacity was significantly bolstered by government-funded scholarships in the 2000s, which attracted and retained talent, eventually turning cities like Montreal into AI hubs (see Yvonne Lau, "Canada AI by numbers - how much money is being spent and who is spending it" (20 August 2025), online: ). If similarly, say, India or Brazil created a national "AI fellowship" program sending hundreds of students to study abroad or funding local PhDs, the long-term payoff could be substantial in building a talent base.
68 For example, Nigeria's Tech in Diaspora conferences encourage knowledge exchange (See Light of Africa, "Connect & Ignite 2025 - Nigeria's Tech Diaspora in the UK Gathers Again to Drive Impact Across Borders" (10 July 2025), online: ).
69 We see encouraging moves like the Partnership on AI including developing country voices, UNESCO's capacity-building workshops on AI ethics for policymakers in Africa, and the ITU's programs to extend broadband (see UNESCO, "AI in Action: UNESCO Empowers Education Policymakers to Harness Artificial Intelligence" (2 October 2025), online: ; Makhtar Diop, "Broadband for All: A digital infrastructure moonshot for Africa" (11 September 2020), online: ).
70 Some companies have launched initiatives - for instance, Microsoft's Airband initiative invests in rural broadband for underserved communities (see Shelley McKinley, "Microsoft Airband: An annual update on connecting rural America" (5 March 2020), online: ; see also Facebook (Meta) has supported open maps and translation projects for low-resource languages (see Leigh McGowran, "Meta says its AI model is the first that can translate 200 languages" (7 June 2022), online: ). These efforts, however, need to be scaled and aligned with local priorities, avoiding a top-down "tech saviour" attitude. Empowerment and co-creation should be guiding principles: local engineers and policymakers must be in the driver's seat, with international partners in supporting roles.
71 Jonas Hjort & Jonas Poulsen, "The Arrival of Fast Internet and Employment in Africa" (2019) 109:3 Am Econ Rev 1032; Chen & Ran supra note 55. https://doi.org/10.1257/aer.20161385
72 Matt Andrews, Lant Pritchett & Michael Woolcock, Building state capability: Evidence, analysis, action (Oxford: Oxford University Press, 2017); see also World Bank Group, World development report 2017: Governance and the Law, (Washington, DC: World Bank, 2017). https://doi.org/10.1093/acprof:oso/9780198747482.001.0001
73 As of 2024, fewer than ten countries in all of sub-Saharan Africa had adopted a robust national AI policy or strategy (see Melody Musoni, "Envisioning Africa's AI governance landscape in 2024" (January 2024), online: ). This leaves the majority of nations without a clear roadmap on AI, and without dedicated governance mechanisms to oversee issues like data protection, algorithmic bias, or AI safety. In contrast, over 30 countries in Europe had some form of AI strategy by that time, illustrating the governance gap.
74 For example, several African governments and cities have begun using facial recognition systems and intelligent surveillance, often through foreign vendors or loans, without clear legal frameworks to prevent misuse. This raises serious civil liberties and human rights concerns (see Bhagyashree Nanda, "Facial Recognition in Africa: Balancing Innovation, Privacy, and Inclusion" (9 June 2025), online: ). Without data privacy laws or surveillance limits, there is a risk of AI tools being used for political oppression or discrimination. In one reported instance, a city implemented a smart CCTV network that was later criticized for targeting certain ethnic neighbourhoods, illustrating how bias in algorithms can translate to bias on the ground. Another instance saw predictive policing tools, imported with little transparency, raising fears of reinforcing systemic biases against marginalized communities. These examples mirror problems seen in the North, but the checks and balances are often weaker in the South, heightening the risks. Where a mature democracy might have civil society organizations and strong courts to challenge wrongful AI use, many lower-income countries may lack such robust recourse.
75 Graham Greenleaf, "Global Data Privacy Laws 2023: 162 National Laws and 20 Bills" (2023) 181 Privacy Laws & Business Intl Report 1. https://doi.org/10.2139/ssrn.4426146
76 For instance, consider the scenario of a foreign health tech company running a free disease diagnosis app in a developing country - it might gather millions of medical images or health records for its AI training, then commercialize the resulting product globally, with no obligation to share profits or even insights with the source country. Such dynamics have been described as "digital extractivism," where data from the South is extracted by the North, much like raw minerals, leaving the source country with little value added (see Mark Latonero, Governing Artificial Intelligence: Upholding Human Rights & Dignity (New York: Data & Society Research Institute, 2018).
77 Anu Bradford, "The Brussels effect" (2012) 107: 1 Nw UL Rev 1 ; Henry Farrell & Abraham L Newman, "Weaponized interdependence: How Global Economic Networks Shape State Coercion" (2019) 44:1 Intl Security 42. https://doi.org/10.1162/isec_a_00351
78 Stephane Couture & Sophie Toupin, "What does the notion of "sovereignty" mean when referring to the digital?" (2019) 21:10 New Media & Society 2305 https://doi.org/10.1177/1461444819865984 ; Julia Pohle & Thorsten Thiel, "Digital sovereignty" (2020) 9:4 Internet Pol'y Rev. https://doi.org/10.14763/2020.4.1532
79 For instance, a few countries in Africa have begun multi-stakeholder consultations to shape their AI strategies - Rwanda's AI policy involved input from academia, industry, and civil society to cover issues like gender equity in AI and using AI for social good (see Republic of Rwanda, Ministry of ICT and Innovation, "The National AI Policy" (2023), online: ). This inclusive approach is a strength that some Global South nations are leveraging, arguably better than some early movers did.
80 Mike Ananny & Kate Crawford, "Seeing Without Knowing: Limitations of the Transparency Ideal and Its Application to Algorithmic Accountability." (2018) 20:3 New Media & Society 973. https://doi.org/10.1177/1461444816676645
81 This could occur, for example, by providing template regulations for member states or pooling resources for capacity-building. Similarly, in Latin America, groups like the OECD LAC AI Initiative are encouraging countries to adopt coherent AI principles aligned with human rights and democratic values (see OECD, "AI, data governance and privacy: Synergies and areas of international co-operation" (2024) 22 OECD Artificial Intelligence Papers, 1).
82 For instance, if procurement of AI systems is non-transparent, it may lead to overpriced contracts that benefit cronies rather than effective solutions.
83 Nick Couldry & Ulises A Mejias, The Costs of Connection: How Data is Colonizing Human Life and Appropriating it for Capitalism (Standford, CA: Stanford University Press, 2019). https://doi.org/10.1515/9781503609754
84 Eric Masanet et al, "Recalibrating global data center energy-use estimates." (2020) 367:6481 Science 984 https://doi.org/10.1126/science.aba3758 ; David Mytton, "Data centre water consumption" (2021) 4:11 npj Clean Water 1. https://doi.org/10.1038/s41545-021-00101-w
85 A case in point is Lagos, Nigeria, a megacity already struggling to provide clean water. In recent years, Lagos has seen a cluster of new data centers built to serve West Africa's growing digital needs. Local reports have highlighted that these data centers may be significantly straining the city's limited water resources. Lagos' groundwater levels are precariously low, and the presence of a dozen data centers requiring cooling has raised concerns among experts and residents. One study found that in a high-end district of Lagos, water shortages (from saltwater intrusion and overextraction) have been aggravated by the power and cooling demands of nearby data centers, which often rely on water-intensive backup generators and cooling towers (see Abdallah Taha & Alfred Olufemi, "Data Centers 'Straining Water Resources' as AI Swells" (16 November 2023), online: ). Similar stories are emerging elsewhere: In South Africa's Gauteng province, officials questioned data center operators on their water sourcing during drought periods (Alsaigh, Mehmood & Katib, supra note 10). In Kenya's Nairobi, plans for a major new data hub prompted calls for an environmental impact assessment focusing on water and energy use. These instances show the necessity for sustainable planning; data centers should ideally use advanced cooling technologies (like air cooling or recycling greywater) and be situated in locations where they will not deprive communities of vital resources.
86 Roy Schwartz et all, "Green AI" (2020) 63:12 Communications ACM 54 https://doi.org/10.1145/3381831 ; David Patterson et al, "Carbon emissions and Large Neural Network Training" (2023) arXiv 1.
87 Emma Strubell, Ananya Ganesh & Andrew McCallum, "Energy and Policy Considerations for Deep Learning in NLP" (2019) arXiv 1; Patterson et al, supra note 86. https://doi.org/10.18653/v1/P19-1355
88 Alexandra S Luccioni, Sylvain Viguier & Anne-Laure Ligozat, "Estimating the carbon footprint of BLOOM, a 176B parameter language model" (2023) 24 J Machine Learning Research 240.
89 Brian R Copeland & Michael S Taylor, "Trade, growth, and the environment" (2003) J Economic Literature 7. https://doi.org/10.3386/w9823
90 Countries like Uruguay and Costa Rica, which have near-100% renewable electricity, have an inherent advantage and can market themselves as green data hosts. In Africa, Kenya's predominantly renewable grid (powered by geothermal and hydro) means that any data centers located there would have a lower carbon footprint than if they were in a coal-dependent country. (See Nick Hedley, "These countries are leading the way to 100% renewable electricity" (14 October 2024), online: ).
91 For example, Facebook's data center in Luleå, Sweden uses ambient Arctic air for cooling year-round, eliminating water use (see Taha & Olufemi, supra note 85). While the tropics cannot replicate Arctic conditions, techniques like evaporative cooling with reclaimed water, or liquid cooling that allows recycling of water in closed loops, can be implemented.
92 ASHRAE, Thermal Guidelines for Data Processing Environments, 5th ed (Peachtree Corners, GA: ASHRAE, 2021); Khosrow Ebrahimi, Gerard F Jones & Amy S Fleischer, "A review of data center cooling technology, operating conditions and the corresponding low-grade waste heat recovery opportunities" (2014) 31 Renewable & Sustainable Energy Rev, 622; Mytton, supra note 84. https://doi.org/10.1016/j.rser.2013.12.007
93 See Edward Oughton et al, "Global vulnerability assessment of mobile telecommunications infrastructure to climate hazards using crowdsourced open data" (2023) arXiv 1 ; Jasper Verschuur et al, "Quantifying climate risks to infrastructure systems: A comparative review of developments across infrastructure sectors" (2024) 3:4 PLOS Climate e0000331. https://doi.org/10.1371/journal.pclm.0000331
94 An additional concept is the use of microgrids and distributed energy to keep services running during climate events. If an AI application is providing, say, early cylone warning or coordinating emergency response, it is vital that it remain online during disasters. Microgrids with solar and battery backup can ensure that key communication towers or data nodes remain powered when the main grid fails (such as after a storm). Countries can identify critical digital infrastructure and provide them with climate-resilient power supplies (perhaps subsidized by climate adaptation funds, since this clearly intersects with adaptation needs). (Sabita Maharjan, "Providing Microgrid Resilience during Emergencies Using Distributed Energy Resources" (2015) IEEE Globecom Workshops 1.
95 Satyanarayanan supra note 55; Intergovernmental Panel on Climate Change, Climate change 2022: Impacts, adaptation and vulnerability (Cambridge, UK: Cambridge University Press, 2022). https://doi.org/10.1017/9781009325844
96 Ibid.
97 The EU's "Climate Neutral Data Centre Pact" could serve as a model for other regions - essentially a commitment by data center operators to achieve 100% carbon-free power by a target date, to recycle heat waste, etc. If similar commitments are encouraged in the Global South, possibly with international support, these commitments could pre-emptively ensure a greener trajectory.
98 Andrews, Pritchett & Woolcock, supra note 72.
99 David Rolnick et al, "Tackling Climate Change with Machine Learning" (2019) 55:2 ACM Computing Surveys 1.https://doi.org/10.1145/3485128
100 Farrell & Newman, supra note 77.
101 Tim Büthe, &Walter Mattli, The New Global Rulers: The Privatization of Regulation in the World Economy (Princeton, NJ, Princeton University Press, 2011). https://doi.org/10.23943/princeton/9780691144795.001.0001
102 Jean-Christophe Plantin et al, "Infrastructure studies meet platform studies in the age of Google and Facebook" (2018) 20:! New Media & Society 293. https://doi.org/10.1177/1461444816661553
103 Anu Bradford, "The Brussels Effect" (2015) 107:1 Nw UL Rev 1.
104 AI is projected to contribute an additional 14 per to North America's GDP by 2030, for instance (see PWC, Sizing the prize: What's the real value of AI for your business and how can you capitalize? (PwC, 2025)).
105 AI is key to cybersecurity, autonomous weapons, and surveillance.
106 Farrell & Newman, supra note 77.
107 Laura DeNardis, The Global ar for Internet Governance (New Haven CT: Yale University Press, 2014). https://doi.org/10.12987/yale/9780300181357.001.0001
108 For instance, when some African countries considered requiring that user data collected by foreign firms be stored in-country (a data localization measure), big tech lobbies argued it would raise costs and stifle innovation.
109 Pepper D Culpepper & Kathleen Thelen, "Are We All Amazon Primed? Consumers and the Politics of Platform Power" (2020) 53:2 Comparative Political Studies 288; Couldry & Mejias, supra note 83. https://doi.org/10.1177/0010414019852687
110 Tim Büthe & Walter Mattli, The New Global Rulers: The Privatization of Regulation in the World Economy (Princeton, NJ: Princeton University Press, 2011). https://doi.org/10.23943/princeton/9780691144795.001.0001
111 Culpepper & Thelen, supra note 109.
112 Couldry & Mejias, supra note 83.
113 Farrell & Newman, supra note 77; Büthe & Mattli, supra note 110.