The New Colonial Infrastructure Argument

The “digital colonialism” argument is made with increasing frequency by African academics, South Asian development economists, and a growing number of technology researchers who observe that the AI tools being deployed in developing countries are built by companies in the developed world, run on infrastructure controlled by those companies, and generate value that flows disproportionately to their shareholders.

The argument deserves serious engagement, which it rarely receives. It is usually either embraced uncritically by critics of technology companies or dismissed impatiently by technology optimists who prefer the “democratization” narrative. Both responses miss what is valid in the critique and what is not.

Start with what the critique gets right. The structural feature it identifies — that technology infrastructure deployed in developing countries tends to be controlled by entities outside those countries, generating data and value that compounds at the controller rather than the user — is real and documented. The history of telecommunications infrastructure in Africa is partly a history of African governments and consumers generating revenue for companies headquartered in London, Stockholm, and Paris. The mobile money infrastructure on which hundreds of millions of Africans depend runs on networks whose ultimate ownership and control sits in Dublin, Amsterdam, and (increasingly) Beijing.

AI adds dimensions to this dynamic that are specifically concerning. Data is more directly valuable as an input to future AI improvement than telephone usage data was to future telecommunications improvement. When an AI company deploys a tool in Nigeria, collects usage data, and uses that data to improve the model for users globally, it is extracting a resource — training data — that has economic value and that the Nigerian users are providing without compensation or control. Whether this constitutes exploitation or simply a terms-of-service relationship that users accept depends on norms that are still being negotiated.

The governance dimension is also real. When critical government services run on AI systems controlled by foreign companies, those companies have leverage that is qualitatively different from what a commercial vendor relationship normally implies. An AI system embedded in a country’s judicial process, healthcare triage, or immigration system is not just a tool that can be switched off — it is organizational capacity that has been partially outsourced to a foreign entity. The dependency that creates is not hypothetical.

Now the critique’s limits. The colonial analogy is historically specific in ways that the “digital colonialism” framing ignores. Colonial relationships were characterized by political control — the ability to impose legal systems, extract resources by force, prohibit competing relationships, and ultimately govern the colonized population. Digital technology relationships are commercial. The companies deploying AI in developing countries cannot imprison users who choose alternative tools, cannot extract resources beyond what users choose to provide, and operate in legal environments that are governed by the host country’s law (however imperfectly enforced).

This distinction matters because it implies different remedies. Colonial domination required anti-colonial politics — independence movements, sovereignty assertion, physical expulsion of colonial powers. Technology dependency requires different responses: governance frameworks, regulatory capacity, capability development, negotiating sophisticated terms of service. The rhetorical similarity between colonialism and technology dependency obscures the practical differences in how each can be addressed.

The second limit is the counterfactual. The colonial critique implies that deploying American or Chinese AI in a developing country is worse than the alternative. But the alternative is not “locally built AI that serves the population’s needs” — that alternative does not exist for most developing countries on any near-term timeline. The actual alternative is usually “no AI for this application” or “AI that is technically inferior because it is built with much smaller resources.” The choice between “AI controlled by someone in San Francisco” and “no AI” is not obviously resolved by anti-dependency reasoning.

The most instructive case for evaluating the colonial critique is the one where it is most clearly correct: the content moderation decisions of social media platforms in developing countries. Facebook’s content moderation in Ethiopia, Tigrinya, and Amharic has been documented as significantly worse than its English-language moderation — contributing, according to multiple documented analyses, to the spread of hate speech and violence incitement during the 2020-2022 Tigray conflict. The platform prioritized content moderation capacity in languages with large advertising markets. Tigrinya had a small advertising market. The result was a failure of the platform’s nominal commitment to user safety that had direct consequences for a violent conflict.

This is not merely a commercial failure. It is a case where the application of a technology in a developing country produced harms that the technology company’s governance structures were not designed to prevent, because the governance structures were built around the interests of users in profitable markets rather than users in vulnerable markets. The “colonial” element here is not extraction of resources. It is the application of norms and governance structures that serve the developer’s interests rather than the user’s.

The remedy the colonial critique points to is governance — requiring that content moderation, AI safety, and platform accountability apply equally to users regardless of their market value. That is a legitimate demand. It does not require accepting the full colonial analogy.

The data governance framing, which is less rhetorically loaded than “digital colonialism,” is probably more productive for policy purposes. The question is: who controls data generated in developing countries, on what terms, with what compensation and rights for the communities that generate it?

Several African countries have enacted data protection legislation since 2018, following the GDPR’s example but adapting it for their specific contexts. Kenya’s Data Protection Act (2019), Nigeria’s Data Protection Regulation (2019 and updated 2023), South Africa’s POPIA (effective 2021), and Rwanda’s Data Protection Law (2021) all create frameworks for data rights that, in principle, apply to AI companies operating in those countries. The enforcement capacity to make these frameworks effective is a different and harder problem.

The development of data governance expertise within African institutions — regulators, civil society organizations, academic researchers who can evaluate compliance and advocate for stronger enforcement — is probably more impactful than the legislative frameworks themselves, which exist on paper and matter primarily when someone can evaluate compliance and consequences follow from violations. Funding that expertise is a development investment that deserves more attention than it receives from the bilateral and multilateral development institutions that are thinking seriously about the AI development agenda.

The digital colonialism critique is worth taking seriously not because the historical analogy is precise but because the underlying concern is real: that AI deployment in developing countries can produce outcomes that primarily benefit the technology suppliers rather than the communities being served, that the governance frameworks preventing this are weak, and that the communities most exposed to poor AI outcomes have the least voice in the governance conversations that determine those outcomes.

That concern can be addressed with governance tools, capability investment, and accountability mechanisms. It does not require treating every American or Chinese AI deployment in a developing country as an act of domination. Most of these deployments provide real value to real users under imperfect conditions. The goal of development policy should be to make the conditions better — more equitable data governance, stronger local negotiating capacity, more investment in domestic alternatives — rather than to reject the value because the conditions are not yet right.

The colonial critique, applied precisely, points at real problems. Applied imprecisely, it forecloses solutions.