In November 2022, querying a model with performance equivalent to GPT-3.5 cost about 20 dollars per million tokens. Less than two years later, in October 2024, that price had fallen to 0.07 dollars. A reduction of around 280 times.
This figure illustrates the extraordinary speed at which the economics of artificial intelligence are changing. Models that until very recently required enormous computing capacity can now deliver similar performance at a fraction of the price. And everything suggests that this trend will continue, although, as some organizations warn, the cheaper it is to use AI, the more we can use it and, therefore, total spending may even increase.
For education, and especially for the systems with the fewest resources, this drop in prices seems to be a good opportunity. Could technologies that until recently were reserved for the countries and schools with the greatest financial capacity be put within reach of schools around the world?
The working paper Costing AI Use in Education in Low- and Middle-Income Countries, published by the World Bank in July 2026, attempts to answer that question from an unusual angle: cost. Its aim is not to determine whether AI improves learning (a question on which the authors themselves acknowledge that the evidence remains limited) but to find out what it means, economically, to incorporate it into an education system.
And when that full calculation is made, the price of the model turns out to be only a small part of the problem.
The paradox: where it can contribute the most, it is hardest to deploy
The possibilities offered by artificial intelligence can be especially attractive in low-resource educational settings. The report mentions tools capable of personalizing teaching, expanding access to materials, reducing some of teachers’ administrative tasks or helping education systems identify students at risk of dropping out. These are potentially relevant applications where teachers are scarce, classrooms are overcrowded or information systems are weak.
However, one only has to look at the material conditions of many schools to discover the other side of that possibility.
According to data provided by the UNESCO Institute for Statistics (UIS) and the International Telecommunication Union (ITU), in low-income countries only 38% of primary and secondary schools have electricity. Barely 27% have computers for teaching purposes and 23% have internet access for that same purpose. Among the general population, only 27% use the internet and fixed broadband penetration is around 1%. In high-income countries, by contrast, most of these indicators are close to or above 90%.
The starting point matters because it determines what has to be paid for even before contracting an artificial intelligence tool. The World Bank explains this through four conditions, which it calls the four Cs: connectivity, compute, context and competency. That is, connectivity; computing capacity; data and content suited to the context; and skills to use the technology. When any of those foundations fails, the bill grows. And when several fail, introducing AI requires investments that far exceed the cost of the tool itself.
Here an essential issue arises for vulnerable environments. It is much simpler and cheaper to deploy an application in an urban school that already has electricity, wifi, computers and teachers with digital experience than to bring it to a rural school without a stable connection. If, in addition, the technology is expected to work in different languages, respond to local curricula and reach teachers and students with very different levels of digital literacy, the needs increase once again.
Equity, in other words, is part of the economic calculation.
The price of AI begins before the AI
To understand why, the report proposes looking at how the economics of educational technology have changed.
In a first stage, incorporating technology meant mainly buying products: computers, servers or software. The investment was considerable, but relatively easy to budget. You bought a hundred computers and you knew what they cost. The big risk came later: that equipment aged and had to be replaced.
The expansion of the internet brought a second logic: services. Educational platforms, digital content and cloud-based programs began to be contracted through subscriptions. Spending thus went from being concentrated in an initial purchase to recurring every month or every year.
Artificial intelligence adds a third economy: that of consumption. Every time someone asks a model a question, an inference takes place: the system uses the capacity acquired during its training to generate a new response. That operation has a cost. A single query may cost very little. Thousands of teachers and millions of students using it repeatedly generate ongoing spending that depends on how much and how it is used.
That is why the World Bank describes the shift from an economy of products and services to an economy of tokens. Spending is no longer entirely predictable: much greater use than anticipated can also translate into a bigger bill.
And tokens are still only one part. There are also cloud infrastructure, data transfer and storage, integration with education systems, cybersecurity, privacy protection, bias monitoring, maintenance and technical support to pay for. AI also introduces new recurring expenses, such as monitoring how the models perform and updating the safeguards needed to use them safely.
On top of all this comes one of the most important components: teachers. Access to a tool does not guarantee its use. The report cites an upcoming study carried out in Peru that shows this with particular clarity: teachers who were given AI licenses free of charge, training in using prompts to prepare lessons and periodic nudges to do so barely reached 15% weekly use. In Latin America, moreover, estimates cited in the document indicate that 73% of teachers lack the basic digital skills needed to teach.
Training, accompanying and supporting teachers is therefore also part of the cost of AI. And that expenditure must fit into extremely tight budgets. In low-income countries, annual education spending is around 55 dollars per student, and teacher salaries and other recurring expenses absorb a large share of the available resources.
The pilot mirage: 500 students are not five million
Let us now imagine an artificial intelligence project for 500 students. A technology company provides free access to the model. Computing costs are subsidized. An international organization funds the project. A small group of teachers receives training and has specialists on hand who can quickly fix any problem. The experience takes place in schools with good connectivity and in a language for which the tool works properly.
The pilot gets good results and, what is more, seems cheap. Now let us try to take it to five million students.
This is one of the most interesting warnings in the World Bank report: the cost of an AI pilot can be a very poor indicator of what it will cost to take that same intervention to scale. Many of the projects currently under way in low- and middle-income countries are heavily subsidized by companies, international bodies or non-profit organizations.
As they grow, those exceptional conditions begin to disappear. The company that gives away tokens for 500 students is unlikely to take on the consumption of five million. A standalone application may need to connect with the national education information system, which requires developing secure data-exchange systems, cloud infrastructure and cybersecurity mechanisms.
And equity appears once again.A pilot can select urban schools, work with especially motivated teachers and have technicians dedicated to resolving any incident. A national rollout must also reach rural schools, provide infrastructure where it is lacking, train and support many more teachers and, in certain contexts, adapt the technology to minority languages and regional dialects.
That is why the report proposes another way of understanding pilots. Besides checking whether a tool works, they should serve to learn how much it costs to use it: to observe how teachers and students actually interact with it, how much consumption they generate, what difficulties arise and what capabilities the education administration needs to develop. A pilot can thus become a laboratory for discovering the necessary conditions before committing millions to a national rollout.
– Start where AI can be viable
What, then, can an education system that wants to take advantage of the possibilities of artificial intelligence but has very limited resources do? The World Bank proposes moving forward in stages and offers some pointers.
– Start with teachers
Providing teachers with access to AI requires far fewer users and devices than doing so directly with all students, and therefore reduces both the initial investment and the recurring costs. Starting with secondary or higher education can make the process even easier, because these institutions tend to have better electricity, connectivity and infrastructure.
– Think of a smaller AI
While large language models try to answer an enormous variety of questions and require significant computational resources, Small Language Models (SLMs) can specialize in specific tasks. The report points to several potential advantages for systems with fewer resources: lower operating costs, the ability to run locally and without a permanent connection, adaptation to local curricula and languages, and greater control over data. In exchange, they tend to have more limited capabilities for complex reasoning and less general knowledge.
– Gradual rollouts that allow you to test, measure, learn and recalculate
In a market where models, their capabilities and their prices change constantly, deciding today how much it will cost to use a given technology five years from now is extraordinarily difficult.
Because the most suitable model for a vulnerable school or education system does not have to be the largest or the most advanced. It can be the one that meets a specific educational need well enough, with an infrastructure, capabilities and cost that can be sustained over time.
The cost of reaching the last one
The logic of economic efficiency invites us to start where it is easiest: connected schools, digitally competent teachers, available infrastructure. Each new user costs less and results arrive sooner. But that same logic can also determine who gets access first to the opportunities offered by a new technology and who will have to wait.
That is why measuring the cost of AI solely per user can be misleading. Reaching a remote school will probably be more expensive than adding another already-connected urban school. Serving a language with few speakers may require an investment that is hard to justify in terms of scale. Training and supporting those who start out with fewer skills will require more resources. Economic efficiency will tend to favor those who are already better prepared to adopt the technology.
Bringing equity into the calculation means accepting that reaching some students will cost more than reaching others. And that difference is not necessarily an inefficiency: it may be precisely the price of building an education system in which a new technology does not reproduce previous inequalities.
How much does it cost to ensure that those for whom access is hardest can also benefit from it: that is the measure missing from many of the calculations we make today about the price of artificial intelligence.
ProFuturo Observatory Team 
