Inclusion for Whom? Frontier AI, Strategic Self-Interest and the New Geography of Research Advantage: A Discussion Paper

In July 2026, OpenAI launched ChatGPT for Academic Researchers, offering 10,000 researchers free access to frontier models and stating that the program would expand to 100,000 researchers during 2027. The initiative was presented as a response to the concentration of AI capability in a small number of companies and well-resourced laboratories. A manual count of the 505 institutions listed in the program’s country dropdown menu on July 30, 2026, shows that 346 institutions, or 68.5%, are in the United States. Adding the United Kingdom, Canada, Australia, and New Zealand raises the combined share to 395 institutions, or 78.2%, while Africa and South America have no eligible institutions. This paper asks what that initial geography reveals about the distribution of frontier AI capability in global higher education. Research excellence does not adequately explain the pattern, and the published criteria provide no transparent account of institutional selection. Individual access is also limited to specified scientific fields and researchers with recent papers on three English-dominant preprint repositories. Drawing on research on increasing returns, organizational learning, and cumulative advantage, the paper argues that early access contributes to institutional capability formation. Universities outside the initial network may miss the period in which practices, integrations, and provider relationships are being established. The same omissions may create space for competing technology ecosystems in underrepresented regions. The paper does not infer the motives behind individual selections or claim that divergence has already occurred. It shows that an access program can broaden participation while still creating conditions under which existing institutional differences compound.