Introduction to the series
The report that follows is the first of a series on the deployment of institution-wide LLM agreements at universities in OECD, EU and other countries at similar levels of economic development (upper- and upper-middle-income according to World Bank).
The reports are of two types:
- Single-vendor global-oriented reports such as on OpenAI’s ChatGPT Edu or Microsoft’s Copilot Premium
- Country-specific reports such as on United Kingdom or the Republic of Ireland, with more to follow
In addition to the geographic countries the reports cover the pseudo-country domain .edu (US higher education institutions and some others, typically those wanting a global profile). An early decision was to extend the .edu report to several other countries by including the other countries which have similar subcomains – such as edu.au or ac.nz – thus getting away from an otherwise US-dominated tone.
The Global Academic Landscape for OpenAI’s ChatGPT
This report maps the global higher education landscape for ChatGPT Edu deployments across academic domains (excluding the separate UK higher education report), using evidence current to 4 May 2026. It covers institutional, system-wide, and national agreements across academic domains in high-income and upper-middle-income countries, drawing primarily on university and national-body sources, with OpenAI announcements and credible media used as secondary evidence where necessary.
The report comes in two versions: with footnotes or with endnotes. For technical reasons at this early phase it is more expeditious to use footnotes/endnotes ratther than citations.
The principal conclusion is that ChatGPT Edu has moved from isolated institutional experimentation to a recognisable global adoption pattern, but one that varies sharply by geography.
Outside the United States, adoption is increasingly shaped by national or system-level agreements. The clearest examples are Italy, where CRUI negotiated a nationwide agreement covering all public universities, and Estonia, where the AI Leap initiative has deployed ChatGPT Edu across public universities and secondary schools, reportedly reaching more than 30,000 users. This model was formalised further through OpenAI’s Education for Countries initiative, launched in January 2026, whose first cohort includes Estonia, Greece, Italy, Jordan, Kazakhstan, Slovakia, Trinidad and Tobago, and the United Arab Emirates.
By contrast, the United States remains a decentralised market in which institutions contract individually or through university systems. The largest confirmed agreement is the California State University system, with a $17 million deployment covering roughly 470,000 students and 63,000 staff and faculty. Other major confirmed US adopters include Arizona State University, Harvard University, Indiana University, the University of Southern California (USC), the University of Utah, Drexel University, Clemson University, and the University of California system. The report also suggests that pricing competition is intensifying in this market, citing UC Davis‘s 40 per cent reduction on renewal as evidence of increasing commercial pressure.
Several regions now appear to be entering a second phase of expansion.
India is one of the strongest growth markets, with OpenAI announcing a February 2026 cohort of six institutions, including IIT Delhi, IIM Ahmedabad, AIIMS New Delhi, Manipal Academy of Higher Education (MAHE), the University of Petroleum and Energy Studies (UPES), and Pearl Academy.
South Korea also appears strategically important, with Samsung SDS and LG CNS securing distribution rights and a major pilot at Korea National Open University (KNOU) involving around 90,000 users.
Additional confirmed deployments in Australia, Brazil, Japan, the Philippines, Turkey, the United Arab Emirates, and continental Europe indicate that the market is broadening beyond a small set of elite Anglophone institutions. (As the UK report will show, the University of Oxford has deployed ChatGPT Edu and a somewhat different arrangement with OpenAI applies at the University of Edinburgh.)
The report distinguishes carefully between confirmed agreements, pilots, and market signals. Some institutions, such as William & Mary and KNOU, remain in pilot phases, while others show active AI engagement without confirmed ChatGPT Edu adoption.
Overall, the evidence indicates a market that is expanding rapidly but unevenly: centralised procurement dominates outside the US, institution-level contracting dominates within it, and India and South Korea are emerging as especially significant next-stage markets.
The next steps
A multi-vendor report will soon be available on the Republic of Ireland, with the much larger UK report following soon after. Preliminary data gathering and “ranging shot” searches are under way for reports on Netherlands, Sweden, Norway, Denmark, New Zealand, Australia and Canada in the next phase. From then on we hope to be responsive to user demand.
Caveats
The paradigm driving these reports is to use the best and most integrated frontier LLM models to generate and moderate the reports, taking care to use hallucination-reducing and quality-check-friendly approaches to searching, with automated checks and a small amount of human intervention. Following a standard approach in large multi-draft projects, Markdown is used as the standard format for intermediate drafts and guidance instructions, with conversion to Word occurring only when final versions are required and detailed human editing and layout manipulations are done.
Thus no assurance can be given that all observations are correct or that every single URL works (and more so, will continue to work). In addition, we know that some deals are still “under the radar” and most likely will appear in the next few weeks; and one deal may even be unwound. University web subsites including those on AI often have an obscure and ever-evolving structure, which makes any tracking (human or automated) more difficult.
Our automated report methodology is evolving very fast; and it is likely that later reports will be both (a) more accurate, (b) more insightful and (c) faster to generate. It is intended that the suite of reports will be redone at least twice a year.
Already three other topics have been identifed as suitable for “the treatment”. Indeed one would have started already being released except that the LLM one was granted priority.
Opening up learning 