The Energy Consumption Dilemma of Artificial Intelligence: Astonishing Insights



by SEAD DEDOVIC

The Energy Consumption Dilemma of Artificial Intelligence: Astonishing Insights
© Andrea Verdelli / Getty Images News

While artificial intelligence (AI) holds promise for addressing climate change, it's simultaneously a significant energy consumer, leading experts to raise concerns. Ralf Herbrih, director of the Haso-Platner Institute in Potsdam, highlights the surging energy demands of AI tools.

AI is rapidly advancing and finding applications in various sectors, from autonomous vehicles to healthcare and economics. Scientists all over the world want to use the potential of AI and make big changes in the world that would go in a positive direction.

A scientific team from Amsterdam found that AI's electricity consumption is already on par with that of an entire country. Data processing centers, including AI, currently account for 4-5% of global energy consumption, which increases to 8% when you include standard hardware like computers and phones.

Projections suggest this share could reach 30% in the near future. Training an AI model, for instance, demands the constant operation of numerous graphics cards, each using around 1,000 watts, a process taking several weeks.

It is obvious that using an AI system requires a lot of energy. However, it is very possible that adequate solutions will be found in the future and this problem will be solved.

AI requirements

Efforts are underway to reduce AI's computational requirements to lower energy consumption, but substantial progress isn't expected in the immediate future.

It will take a lot of work and effort to find the right solution for this problem. Recent findings by Dutch scientist reveal that AI consumes substantial energy not only during training but also during its ongoing operations.

For example, systems like ChatGPT consume electricity each time they generate text, potentially adding to energy demands. We have to admit that AI has saved many, and that it will only reach real expansion in the years ahead.

While the extreme scenario of AI consuming energy equivalent to an entire country is currently improbable due to server costs and technical limitations, it underscores the need for addressing AI's energy footprint as it continues to advance. We will see what the future holds and what solutions will be found.