Who gets heard in the AI rumble?
The conversation on AI is strongly shaped by the leading countries in the field, usually higher-income countries and scientific powerhouses.1
Let's take, for instance, the mentions of ChatGPT in the global public discourse since its release until the beginning of 2025:2
- Around 141,500 news articles were observed, with reporting hotspots in New York, London, San Francisco, Berlin, and Tokyo.
- 43% of the pieces were written in English.
- The most frequent organizations starring in the news were Microsoft, Google, and Twitter (now X).
In contrast:
- Despite Spanish being the second language with the most native speakers in the world,3 just 11% of the news was posted in that language.
- India was the only developing country among the top ten most referenced countries. However, its repeated mentions concern its relevance as an emerging market for AI applications.
With the Global North receiving so much attention for AI progress, what kind and magnitude of developments are arising in the Global South?
Where is AI research actually happening in the Global South?
I'd like to approach this issue from the angle of metascience, a field of study with its intellectual seeds in Eastern Europe over a century ago, and now at the epicenter of science policymaking worldwide.4,5,6
With so many definitions, I settled on defining metascience:
As the scientific approach to understanding and assessing how knowledge is generated and applied, its impact, and the structure of the actors involved, from individual researchers to (inter)national science systems.
The Global South — a concept coined by the "Global North" that will not die, even though it explains little, as pointed out by historian Stephen Kotkin7 — is multilaterally acknowledged as a set of developing countries in continents such as Africa, Latin America and the Caribbean, Asia (excluding Israel, Japan and South Korea) and Oceania (excluding Australia and New Zealand).
In this piece, I'm interested in understanding more about AI in the Global South by using open access sources from the metascience infrastructure, with a focus on its endogenous research capabilities (a less politically charged term than research sovereignty).
Among the multiple dimensions I can use to identify local research, I will focus on the Producers dimension.8
It refers to research resulting from knowledge integrated locally by a local community of practice, or to a team of scientists affiliated with institutions geographically based within the same country.
For example, a research paper on Neural Networks and Applications published by scientists from Universidad del Rosario and Universidad Nacional, both Colombian universities, would fall under the dimension I'm interested in.
I looked at research articles in the AI subfield9 with at least one author in the Global South over the last decade (2016–2025) using OpenAlex,10 an open access catalog of over 463 million work records. I excluded China from the search since it now follows the United States in AI development.11
I initially found 134,600 records.12 After a standard cleaning process of removing publications without a unique identifier for scientific papers (DOI) and duplicate records, I identified which publications may help us unveil country-level endogenous research capabilities.
Therefore, I identified publications whose authors were affiliated with institutions in the same country, and, for the aim of this piece, I excluded multinational collaborations, even those South-South collaborations such as between institutions in Niger and Algeria that worked together in research articles.
I ended up with 93,458 research articles on AI reflecting endogenous research capacities.
Feel free to explore the data using the interactive map below.
It is possible to draw many conclusions based on this visualization. The first point is that, in developing scientific systems, vast differences are clear both between and within continents. Note that these are raw counts, not adjusted for country size or overall research output, and that OpenAlex's coverage of local journals varies from country to country.
Take India and Indonesia, which produced over 28,000 and 22,000 articles respectively, compared to Iran, which generated only a fourth of that amount.
In Africa, similar cases include Egypt and Algeria, the latter ranked second with half of Egypt's output, while in South and Central America and the Caribbean, Brazil produces seven times more than Colombia, the second-highest producer in the region.
Moreover, India's production exceeds the total of Egypt and Brazil combined by more than threefold.
A noticeable absence from the Latin American side of the map is Mexico, which registered around 600 publications on AI-related topics. It is important to note, however, that my definition of local research excludes international collaborations. So Mexico's absence from the map does not indicate a lack of research activity, but rather an intensive international research agenda involving collaboration with other countries.
Not all AI is created equal: the Global South's research priorities
Now, let's move into the research portfolio and specialization of Global South geography and its endogenous AI research capabilities.
Feel free to explore the data using the interactive tree map below. It can be filtered by continent.
As might be expected, each region has been focusing on different topics. First, Asia accounts for the largest volume of publications, mostly focused on sentiment analysis, optimization algorithms, AI applications in cancer detection, educational technologies, and cybersecurity.
Africa, for its part, shows a substantial accumulation in solar energy and photovoltaics, optimization algorithms, geochemistry, and natural language processing.
South and Central America and the Caribbean, the region where I'm from, displays a balanced distribution across solar energy, optimization, educational robotics, quantum information, and semantic web technologies.
Beyond their differing national policies, priority agendas, funding, and capacities, the three regions share interests in computational methods, AI applications, and emerging digital technologies.
The Global South is researching AI, but who's paying?
The Global South may be quieter in the global AI conversation, but it is far from silent in AI research.
Behind the headlines lies a diverse landscape of countries developing their own research capabilities, although at remarkably different scales and with distinct priorities.
Again, this highlights how little a "one size fits all" concept of the Global South explains.
While locally produced research does not mean scientific independence, it offers a starting point for understanding where these capabilities are emerging.
Now, another issue might arise. Focusing on academic publications and topics is only part of the narrative. Researchers, universities, and policymakers should look beyond publication counts to identify and strengthen the people, institutions, and resources behind local AI research.
This brings us to a question that publication data alone cannot answer: Who is funding this research?
In the second part of this series, I'll follow the money to examine who funds AI research in the Global South, how funding sources vary across countries, and whose priorities may be shaping the region's scientific agendas. After all, understanding who produces AI knowledge is only half the story. Understanding who pays for it may reveal who gets to shape its future.
Notes
- hai.stanford.edu/ai-index/global-vibrancy-tool ↩
- doi.org/10.1007/s44163-026-01039-z ↩
- ethnologue.com/insights/most-spoken-language ↩
- researchonresearch.org/project/meta-moment ↩
- cfrr.net ↩
- whitehouse.gov/science ↩
- youtube.com/watch?v=OyWMtTHdH7Y ↩
- doi.org/10.1093/reseval/rvag045 ↩
- openalex.org/subfields/1702. This subfield is defined as computer science and engineering practices for intelligence demonstrated by machines and intelligent agents. It is composed of over 70 topics, such as topic modeling, metaheuristic optimization algorithms research, logic, programming, and type systems, natural language processing techniques, speech recognition and synthesis, semantic web and ontologies, cryptography and data security, neural networks and applications, multi-agent systems and negotiation, among others. ↩
- openalex.org ↩
- hai.stanford.edu/ai-index/global-vibrancy-tool ↩
- The replicable OpenAlex query can be found here. ↩