Han Dieperink: Saving energy with AI
Han Dieperink: Saving energy with AI
This column was originally written in Dutch. This is an English translation.
Artificial intelligence consumes a lot of energy, but at the same time it can yield dramatic savings in the wider economy.
By Han Dieperink, written in a personal capacity
The figures speak for themselves: data centres currently consume 1.5% of global daily energy consumption, and that percentage is expected to double to 3% within a short period of time. For many people, this is proof that artificial intelligence (AI) is an energy monster that jeopardises our climate goals. This argument seems strong, but it overlooks an important point: AI may actually be the key to greater energy savings.
The context of consumption
Of course, the direct impact is visible. Each ChatGPT query emits more than 4 grams of CO₂: ten to twenty times more than a Google search. Multiply that by 400 million daily users and the carbon footprint begins to resemble that of small countries. The AI boom has sparked a race to build new gas and nuclear power plants, as the irregularity of renewable energy makes it difficult to make data centres completely green.
Furthermore, we now take into account the CO₂ intensity of AI that develops code, but we do not count how much energy a developer uses to write that same code. It is therefore important that we evaluate our energy consumption correctly.
Neutrality within reach
Recent research shows that if AI improves energy efficiency in the wider economy by just one-tenth of the rate of its own adoption, the net effect could be energy neutral or even slightly positive.
The real potential of AI lies in its ability to improve the efficiency of almost everything we do. The staggering inefficiency in parts of our energy system offers enormous opportunities. The materials production chain, which produces 60 billion tonnes of steel, glass, hydrogen, ammonia and copper annually, uses four to five times more energy than the absolute minimum required for the chemical reactions.
By 2030, companies could collectively save £2 trillion per year by leveraging existing digital energy tools. AI can automatically adjust electric vehicle charging times, manage heating and cooling, and optimise production schedules to reduce costs and emissions.
Become smarter, not bigger
The key lies in three factors that can curb AI's carbon footprint: teaching AI to be more sustainable, designing data centres more efficiently with renewable energy, and using AI to reduce emissions in other sectors. We are learning that most energy consumption is not in training models, but in their actual use. By teaching AI to ignore bad prompts and not waste energy on worthless tasks, we can achieve significant savings.
Sustainability as standard
In five years' time, we will probably no longer talk about “sustainable AI”. AI will be sustainable by definition. It will become the norm, not the exception. As “smart” devices and sensors become more widespread, AI will help reduce waste in energy production, transport and much more. The idea that energy will ultimately be the limiting factor for AI progress overlooks the transformative power of the technology itself. Instead of being an energy guzzler, AI can become the driving force behind our energy transition.