Artificial intelligence arrived decades ago to change our lives, but it only really started to do so in 2022. The launch of ChatGPT was a before and after, but things seem to have changed. What is the relationship between AI and energy? Well, every time we use one, we exceed the predictions of the other (for worse, although projects like this one in Iceland could be a game changer).
AI and energy, two sides of a coin we don’t yet understand
Artificial intelligence has become the technological Holy Grail, but its rapid development raises a crucial question: at what energy cost? A Cornell University study finds that training models like ChatGPT in 2021 has already consumed as much water as Microsoft’s US data centers in a year.
This process is essential for generative AI like OpenAI’s, whose training can consume more than 400 megawatt hours (MWh), or enough to power 40 average American homes for a year. Doesn’t that seem unacceptable to you? In fact, it is.
AI’s energy footprint doesn’t stop there. Every time a tool like ChatGPT generates data, it also uses a significant amount of computing power and therefore energy. An analysis shows that its operation could cost 564 MWh of electricity per day.
ChatGPT consumes as much as a country: the reality
Data from French company Shneider Electric reveals that AI already consumes 4.3 GW of energy globally, equivalent to the consumption of some small countries. The rapid adoption of artificial intelligence, which prioritizes economic profitability over ecological impact, is worrying.
It is estimated that by 2028, consumption could reach 20 GW, representing a growth of 36%. Data centers, responsible for fueling the development of AI, are the main cause of the shortage that, although we do not yet perceive it, is increasingly imminent.
A ChatGPT query costs three times more than a Google search, according to several specialized consultancies. In fact, it costs more than leaving the computer on for a whole hour (if we add the cost of training this generative AI tool).
The cost of energy in this sense is higher than in many countries, especially medium-sized European countries and most countries in South Asia, Africa or Central and South America. Obviously, it is also much higher than in the island states of Oceania.
Nuclear power plants to continue using AI? Here’s the proposal
The dilemma is getting worse, and some proposals suggest extreme solutions. According to Sergey Edunov of Meta, two nuclear reactors would be enough to meet the demand generated by AI applications by next year. Nuclear power plants emit clean and safe energy, but doubts remain about their performance.
In fact, the problem is not the cost of AI today, but what it will cost in the medium term (or even the short term, if we are realistic). One issue that humanity has not yet been able to resolve is that of dealing with nuclear waste, which pollutes for some 24,000 years.
Companies such as Meta and Microsoft are already considering nuclear energy as an alternative to fuel the voracity of AI. Although it is a controversial option, it could be a necessary measure to ensure sufficient energy supply. Are we ready to accept this compromise?
As you can see, times are tough for AI and energy, which have not yet learned to coexist in our way of life. Overuse of the former will result in consumption of the latter that we can still afford. But what will happen when it increases exponentially? Perhaps it will be necessary to resort to new solutions, like this wooden wind turbine.