Pennings, A.J. (2026, Sep 07) ICT4D and the Geography of AI Computation. apennings.com https://apennings.com/global-e-commerce/ict4d-and-the-geography-of-ai-computation/

Introduction

Artificial intelligence (AI) is often presented as a universally available technology. Connect to a model, enter a prompt, and receive an answer. But for much of the world, AI is not simply a software application. It is an infrastructure problem. Access to useful AI depends on data, networks, storage, computing power, cloud services, energy, technical skills, organizational capacity, and the ability to connect these resources into functioning systems. The geography of artificial intelligence is therefore also a geography of computation.

This raises a fundamental question for Information and Communication Technologies for Development (ICT4D), an important specialization in our Technological Systems Management.[1] What happens when the organizations and countries that need AI most lack the infrastructure on which advanced AI depends?

The International Telecommunication Union (ITU) has increasingly identified this problem as an emerging AI divide. Its recent analysis notes that the concentration of computing infrastructure, data centers, talent, and advanced AI capabilities is reinforcing existing geographical and economic inequalities. More than 150 countries, particularly in Africa and South America, reportedly lack major data-center facilities. At the same time, the computational resources necessary for frontier AI remain concentrated among a small number of countries and companies.

For ICT4D, this changes the educational problem. The objective should not simply be to teach students how to use ChatGPT or another frontier model. Students need to understand the complete computational system through which information becomes knowledge, computation becomes capability, and capability becomes organizational action.

This is particularly important for Technology and Systems Management (TSM) students with an ICT4D interest. A TSM graduate in a country with relatively limited AI infrastructure may never design a frontier model. That does not mean the graduate cannot become an effective AI systems manager. Quite the opposite. The student needs to learn how to assemble useful computational systems from whatever combination of local and external resources is available.

The starting point is the life of data. Data enters an organization through people, transactions, sensors, documents, mobile devices, government records, websites, machines, and other digital systems. It is then represented computationally, stored somewhere, retrieved when needed, processed by software or an AI model, connected to external tools, and potentially converted into an organizational action. The basic sequence can be expressed as:

Capture -> Represent -> Store -> Retrieve -> Compute -> Connect -> Coordinate -> Act -> Learn

Each step, however, has a geographical dimension. Consider a rural health organization. Patient information might begin on paper, move into a spreadsheet, enter a database, and eventually become part of an AI-assisted diagnostic or administrative system. But where is that database located? Where does the AI computation occur? Is the model running on a local computer, a regional data center, or a cloud service thousands of kilometers away? What happens when the network connection fails? Who owns the infrastructure? Who controls the data? What does it cost to transmit and process the information? And can the system continue functioning when external services are unavailable?

These are not peripheral technical questions. They are questions of development. The first capability, therefore, is data capability. A country cannot simply import AI and expect development to follow. Organizations need data that can be captured, represented, cleaned, stored, retrieved, and interpreted. Yet digitization itself involves choices about how local reality is represented. A community, patient, farm, business, household, or ecosystem has to be translated into categories, records, fields, coordinates, images, documents, or other computational objects before it can become part of an AI system.

This makes data representation an important ICT4D issue. The problem is not merely whether a country has data, but whether its institutions’ data structures adequately represent local circumstances. Systems designed elsewhere may poorly represent language, geography, administrative categories, cultural practices, informal economic activity, and locally important knowledge.

The second capability is network capability. AI is frequently described as a computational technology, but computation increasingly depends on telecommunications. The path from a user’s device to an AI system may involve Wi-Fi, a mobile network, an Internet service provider, national infrastructure, international fiber, cloud infrastructure, an application programming interface, and a remote model. Bandwidth, latency, reliability, data costs, and international connectivity therefore become part of AI capability.

This is one reason why ICT4D’s long-standing concern with broadband infrastructure remains relevant in the AI era. The digital divide has not disappeared. It has acquired another layer. A population may have smartphones and mobile connectivity but still lack access to substantial computational resources. The next divide is consequently not simply between those who are connected and those who are disconnected. It is between those who can meaningfully participate in computational systems and those who remain dependent on infrastructures controlled elsewhere.

The third capability is computational capability. Here the geography becomes particularly visible. Advanced AI requires substantial computing resources, specialized processors, data centers, electricity, cooling, networks, and technical expertise. Countries with limited access to these resources must make strategic decisions about what to develop locally and what to obtain through external providers.

That does not necessarily mean that every country needs its own frontier model. A more realistic strategy may combine locally controlled data and applications with externally provided models and cloud computation. The key is understanding the architecture well enough to determine where dependence is acceptable and where local capacity matters strategically.

Retrieval-augmented generation provides a useful example. A local organization can maintain its own documents, databases, policies, records, and other knowledge resources while using an external model for computation. The organization can index, retrieve, and supply information to the model as needed. In this architecture, the organization does not need to train a frontier model to make its own knowledge available to AI. The critical capability becomes building and managing the local knowledge infrastructure around the model.

This distinction is fundamental to ICT4D. AI capability does not necessarily mean owning the most advanced model. It can mean possessing the organizational and technical capacity to combine data, retrieval, networks, models, tools, and people effectively.

The fourth capability is organizational capability. AI becomes economically and socially consequential when it connects to existing systems. An AI model that can answer questions is one thing. An AI system that can retrieve organizational data, query a database, access an enterprise system, communicate with another application, prepare a transaction, and request human authorization is something quite different.

This is where APIs, databases, cloud services, tool interfaces, retrieval systems, and emerging protocols such as the Model Context Protocol become important. They provide mechanisms for AI systems to connect computation to the wider digital environment.

The TSM graduate therefore needs to understand not only models but the infrastructure around them. The student needs to understand how data moves, where it resides, how systems authenticate users and services, how permissions are established, how tools are invoked, and how computational outputs enter organizational workflows.

The fifth capability is human and institutional capacity. This may ultimately be the most important ICT4D issue. A country can acquire cloud access without developing the skills to use it effectively. It can buy AI services without building organizations that can integrate them into productive workflows. It can establish data centers without creating the institutional knowledge necessary to govern them.

This is why the emerging international discussion around the AI divide increasingly emphasizes skills, infrastructure, data, policy, and institutional capacity together. The ITU’s 2026 AI for Good Lab initiative, for example, explicitly combines AI readiness, skills development, locally relevant datasets, compute infrastructure, and open-weight models in its work with developing and emerging economies.

The TSM graduate occupies precisely this intersection. The student does not need to become a specialist in every component of the AI stack. Rather, the graduate needs sufficient technical understanding to manage the relationships among the components. A TSM graduate should be able to look at an organizational problem and trace its computational pathway: where the data originates, how it is represented, where it is stored, how it is retrieved, where computation occurs, which tools are available, how AI agents coordinate, where humans intervene, and how the final output becomes an organizational action.

This perspective also changes how we understand AI agents. In a high-resource environment, an organization might deploy multiple specialized AI systems: research agents, data agents, financial agents, compliance agents, logistics agents, and managerial agents. But the ICT4D question is more fundamental: can agentic systems amplify scarce human expertise?

A small public-health organization, for example, may not have the staff to continuously monitor scientific literature, analyze local data, translate information into several languages, track supplies, prepare reports, and coordinate with government agencies. Well-designed AI systems could extend the productive capacity of a small human organization by letting computational agents perform parts of these tasks under human supervision. The objective is not simply automation. It is capacity multiplication.

This is also where governance becomes essential. The more an AI system moves from producing information toward taking action, the more important questions of authorization, security, accountability, and institutional control become. NIST’s AI Risk Management Framework emphasizes that AI risk management must extend across the lifecycle of AI systems, with governance operating across the processes of mapping, measuring, and managing risk.

For an ICT4D-oriented TSM graduate, this means asking a deceptively simple question “What should the system be allowed to know, decide, and do?”

That question becomes especially important when computational infrastructure is geographically external. If an organization depends on a foreign cloud provider, an overseas model, an external API, and internationally controlled data infrastructure, then the system may provide valuable capabilities while simultaneously creating new forms of infrastructural dependency.

The objective of ICT4D should therefore not be framed simply as bringing AI to developing countries. It should be framed as developing the capacity to participate in, adapt, govern, and benefit from computational systems.

This marks an important shift in how we understand the digital divide. Earlier ICT4D debates focused heavily on access to telephones, computers, Internet connections, and broadband. The AI era adds another layer that includes access to data, compute, models, tools, technical expertise, and organizational integration. The divide is becoming a systems divide. The geography of computation consequently matters as much as the geography of connectivity.

A TSM curriculum designed around this problem would follow data from beginning to end. Students would learn how data is captured and represented, how databases and storage systems preserve it, how retrieval systems make it available to AI, how models process it, how APIs and tools connect AI to external systems, how agents coordinate computational work, and how outputs become organizational decisions and actions.

But the student would learn to ask one additional question at every stage. “What can we do with the resources we actually have?” That question is quintessentially ICT4D. It shifts the emphasis from importing technology to building capability. It encourages architectures that are affordable, resilient, locally relevant, interoperable, secure, and able to function under conditions of constraint.

The result is a different conception of the TSM graduate. The graduate is not simply an AI user, a programmer, or a manager who happens to have access to AI. The graduate becomes a manager of networked computational capacity.

That role may become increasingly important precisely because AI infrastructure is unevenly distributed. Countries with abundant compute will need managers who can integrate enormous computational resources into organizations. Countries with less access will need managers who can accomplish useful things with limited compute, intermittent connectivity, external models, local data, open technologies, and scarce technical expertise.

In both cases, the fundamental problem is connecting computation to human purposes. ICT4D therefore remains relevant in the age of AI not because AI is another digital technology that needs to be distributed. Its deeper contribution is to ask who controls the infrastructures through which computation occurs, who can participate in them, whose knowledge they represent, and whether computational capability actually expands human and organizational capabilities.

The geography of computation is ultimately a geography of opportunity. The central development challenge is no longer simply getting people online. It is ensuring that they can capture data, represent their own realities, retrieve knowledge, access computation, connect tools, coordinate intelligent systems, and convert computational capacity into meaningful action.

That is where ICT4D, AI, and Technology and Systems Management meet.

References

International Telecommunication Union. (2025). AI for Good Impact Report. ITU.
International Telecommunication Union. (2026, July 8). AI ambition to action: The AI for Good Lab initiative. ITU.
International Telecommunication Union. (2026, July 10). Bridging the AI infrastructure divide. AI for Good Global Summit.
National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0) (NIST AI 100-1). US Department of Commerce.
Tabassi, E. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1.
United Nations. (2024). Mind the AI divide: Shaping a global AI agenda for inclusive development. United Nations Development Programme/related UN system work.
World Bank. (2023). Digital Economy for Africa (DE4A) Initiative. World Bank.

Notes

[1] Our BS degree in Technological Systems management has a specialization in ICT4D, as well as options in Computer Science and Applied Math and Statistics.
AI Prompt(s) Compress into a blog post called ICT4D and the Geography of Computation. Below is a compressed version that keeps the ICT4D perspective central and treats the TSM graduate as a manager of computational systems under unequal conditions of infrastructure, data, connectivity, and compute.

© ALL RIGHTS RESERVED

Not to be considered financial advice. AI is often used, and results are thoroughly interrogated. Links are used for some citations. Views are my own and do not express the stances of my employers, past or present.

---

Anthony J. Pennings, PhD is a Professor at the Department of Technology and Society, State University of New York, Korea and a Research Professor for Stony Brook University. He teaches AI and broadband policy. From 2002-2012 he taught digital economics and information systems management at New York University. He also taught in the Digital Media MBA at St. Edwards University in Austin, Texas, where he lives when not in Korea.

Related

Category: artificial intelligence, creative cities, democratic political economies, digital coordination, digital geography, enabling frameworks, food policy and security, geopolitical risk, global communications, global e-commerce, mobile technologies, public administration, science and technology studies, spreadsheet logic, sustainable development, technologies of meaning, technology management
Tags: ICT > ICT4D > The International Telecommunication Union (ITU)

Comments

Comments are closed.