Energy Consumption of AI in Schools: Know What You’re Doing!

Data Hunger Is Growing

Around the world AI is driving the addition of ever more servers. In 2024 the International Energy Agency (IEA) estimated the global electricity consumption of data centres at 415 terawatt‑hours per year, roughly 1.5 % of total electricity use. That may sound modest, but the figure jumped a striking 12 % in just five years. It becomes a serious problem when energy sources are scarce and the planet’s capacity to absorb CO₂ is dwindling (see NOS).No new wind farm can offset that. A recent article in Nature titled Light bulbs have energy ratings – so why can’t AI chatbots underlines the issue.

What Works Partly: More Efficient Programming

A seemingly simple solution that is not yet widely applied is energy-efficient programming. Bernard says: “We certainly need to do that. At NOLAI, we are calculating the energy consumption of all projects in which educational AI prototypes are being developed. As always, measuring is knowing! We do this in our Software Energy Lab. But that's not enough. We have known about the Jevons paradox since 1865: efficiency lowers costs, increasing demand and consequently increasing total consumption. More efficient programming is necessary, but it’s not the solution. More needs to be done, and schools can also play their part.” Bernard’s research also shows that opportunities lie on the organizational side. By setting lower requirements – for example, only agreeing that systems are always available during school hours – or by looking at green hosting and using hardware for longer, improvements can be made.

What Really Works: Conscious Use in Schools

Bernard gives schools 8 tips to be more aware of energy consumption. Because schools want to do this, but often don’t know how or where to start. Schools considering new ICT or AI applications can get to work with this concretely. Also see the whitepaper for an easy-to-print checklist:

1. Question your suppliers
   Providers of ICT and AI are still too rarely asked about the consumption of their products. Not only the costs of purchase and management are important, but also the electricity consumption – there can be significant differences. Also good for the wallet, a win-win.
2. Stick to the purpose
   People tend to apply new technology because it's new, not because it necessarily adds something. Remember that AI and ICT are ‘just’ supporting tools, if the support doesn’t directly connect to learning and teaching, something is wrong.
3. Follow the data
   Know what is being consumed at school and by what. Create an overview in the way that suits you, even if it’s on the back of a napkin. Start by mapping all data streams. What kind of data is there, where is it stored, and how is this data exchanged? We do this for our projects too, and it turns out that there are often blind spots that are quickly overlooked. But that’s where opportunities for sustainability (or privacy risks) can lie.
4. Pay attention to the time
   Especially in schools, not everything needs to be available 24/7. Letting go of that requirement already saves a lot of energy. Pause school systems, such as the internet connection, in the evenings, weekends or school holidays.
5. Dare to let go
   Where bulging archive cabinets used to automatically ensure regular clean-up, all that online information is now out of sight. Agree on clear maximum storage periods for data. It can be very generous, as long as it's gone when it's no longer useful. Not cleaning up should be difficult, but extending it should be easy. Otherwise, we will suffer from data obesity.
6. Remain a free spirit
   Large technology providers know how to trap you so that it becomes almost impossible to switch. This is called vendor lock-in, an increasing problem. So ask about cancellation terms and switching options, even if you are not (and perhaps never) going to do so. That way, you can also switch to green products in the future.
7. Extend use as long as possible
   For ‘end-user’ devices like smartphones and laptops, it is especially true that production costs are burdensome, not so much their use. Then it is important to extend the use for as long as possible. Maybe that laptop can last another year, with or without minor repairs…
8. Keep it small
   General AI, such as ChatGPT or Microsoft Copilot, is designed for large, complex calculations. Often, not all that capacity is needed for the solution you were looking for. Aside from privacy and sovereignty issues, this is a very good reason to choose custom AI, or educational AI. More specific models deliver higher quality for that specific purpose, and can also be better researched for responsible use in schools. And these more specific models use less energy because they are smaller.