Technology in education has long followed a familiar pattern: excitement about a new tool, followed by disappointment when it fails to live up to its promise. Generative AI, unleashed on the world with the release of ChatGPT in November 2022, felt different, and for many education systems, it still does. But as ministries of education in low- and middle-income countries move from curiosity to planning, one question has risen to the top: How much does it cost? A new paper, Costing AI Use in Education in Low- and Middle-Income Countries, explores and unpacks this question.
Historically, estimating costs associated with uses of new technologies in education has been difficult and complex for policymakers and planners. Budgeting for ‘edtech’ has often focused mainly on visible, upfront capital expenditures while ignoring much larger, ongoing total costs of ownership and operation. Edtech + AI introduces new challenges for education planners, especially in countries that face severe resource constraints. To plan effectively for the future of digital education in an age of AI, education systems will benefit from understanding how related business models and technologies are evolving.
The edtech sector has transitioned through three distinct but overlapping paradigms:
- Initially, procurement focused on one-off purchases of hardware and software.
- Over time, the introduction of subscription models for software and education content (and, in some places, including leased hardware as part of bundled services) resulted in an evolution away from what were largely capital expenditures to increasingly include various types of recurrent operating expenditures.
- Now, a new phase is emerging, introducing new, variable costs related to the use of generative AI models.
This evolution disrupts traditional cost considerations related to edtech, altering some of the defining questions a ministry must ask and introducing new fiscal vulnerabilities.
| The Evolution of EdTech Costs | |||||
|---|---|---|---|---|---|
| Phase | Core Activity | Examples | Defining Question | Cost Type | Fiscal Vulnerability |
| Ownership | Buy physical assets or permanent licenses | Devices, laptops, desktop software | “Do we have it?” Inventory | CapEx Capital expenditure | Assets become outdated within a few years Obsolescence |
| Access | Subscribe to cloud platforms or lease hardware | Digital content portals, personalized learning platforms, SaaS, leased computers | “Are we using it?” Utilization | OpEx Operating expenditure | Flat fees apply regardless of actual use Underutilization |
| Consumption | Consume variable computing units in real time | Generative AI, API processing, metered large language models | “Are we using it efficiently?” Resource optimization | Utility cost Variable operating expense | Usage-driven costs can produce unpredictable bills Budget volatility |
How AI Shifts Traditional EdTech Expenditures
Most ministries of education utilize established frameworks to categorize and budget for spending on educational technology. The introduction of AI shifts cost dynamics within this framework. While the general taxonomy remains unchanged, the internal financial plumbing does change, introducing highly variable, recurring expenditures that now run alongside costs that were historically fixed and predictable. For ministries in resource-constrained settings, this exposure is particularly acute: unlike higher-income systems with fiscal buffers, LMICs have limited room to absorb unpredictable utility cost spikes mid-budget cycle.
| Cost Category | EdTech Baseline | EdTech + AI Shift |
|---|---|---|
| Infrastructure | Purchasing physical hardware (devices, servers) and installing school-wide broadband and electrical networks. | Shifts from one-time hardware asset purchases to include ongoing fees for cloud compute hosting and high-bandwidth data transfers. |
| Software and Digital Content | Paying predictable, flat annual “per-seat” or “per-student” subscription licenses for software and static digital tools. | Shifts increasingly to reflect a new variable “token economy”, where costs are metered dynamically by usage volume, prompt frequency, and data processing. |
| Training and Professional Development | Teaching basic digital literacy and how to support the use of technologies in support of sound pedagogical practices. | Shifts to include new pedagogical literacies related to working in an environment characterized by dynamic change as a result of AI. |
| Implementation and Integration Services | Setting up local networks, Learning Management Systems (LMS), and School Information Systems (SIS). | Shifts to fund new activities, such as building data pipelines (e.g., RAG systems) to securely connect AI models to core national education databases (EMIS, SIS). |
| Data and Legal | Reviewing basic software terms of service, privacy protections and ensuring standard password protection for student registry data. | Shifts to fund algorithmic governance, including continuous testing for bias, copyright, and data privacy compliance. |
| Maintenance and Support | Fixing physical hardware, managing local network outages, and installing software updates. | Shifts to include ongoing technical audits to monitor for “model drift” (declining AI performance) and maintaining guardrails to ensure safe use. |
| Project Management | Overseeing hardware distribution, tracking device and content usage, and managing vendor contracting. | Shifts to support continuous monitoring of the impact of AI, tracking cost-effectiveness, and managing the equity risks across the school system. |
| Miscellaneous | Budgeting backup funds for installation delays, replacing broken equipment, paying for security and e-waste disposal. | Shifts to introduce flexible contingency funds to handle changes in token costs and utility cost spikes. |
For ministries of education operating with tight budgets and limited fiscal flexibility, token-based AI pricing – where costs are charged as usage occurs rather than fixed in advance – is not just a technical accounting matter. It is a strategic planning challenge. Understanding this new cost equation early, before large-scale AI procurement decisions are made, is important for education planners. Those who master this shift will transition from merely managing budgets to supporting more flexible, dynamic, and equitable approaches to AI that support the needs of teachers and learners.
Source : World Bank
