A search on ChatGPT uses 10 times more energy than a comparable Google search. 

Few topics stir our emotions more these days than what artificial intelligence, or AI, will mean for humanity’s future. Many believe AI will lift the human condition and bring accelerated learning and innovation to countless communities and people. Many others fear that AI’s emergence is really the beginning of humanity’s decline. A few even suggest it will eventually lead to our extermination.

What everyone does agree upon is that the last few years have witnessed a dramatic expansion of AI’s role in nearly every sector of the economy and every part of our lives. AI’s development is the top priority of all major tech companies, from Microsoft and Alphabet (Google’s parent company) to Amazon and Meta, as they compete with one another for technological and marketplace advantage. Amazon, for example, recently announced it would spend more than $100 billion for capital investments in AI infrastructure in 2025 alone.

What hasn’t received as much attention is what goes along with AI’s rapid growth: a breathtaking increase in demand for both energy and water. If the possible applications for AI are nearly unlimited, then our thirst for more energy appears unlimited as well—at least in terms of current energy investments and technologies. And those communities and countries that aren’t rapidly pursuing new energy supplies and sources are likely to fall even further behind. 

Data centers form the irreplaceable infrastructure AI technologies rely upon, and they consume massive amounts of water and electricity to cool and run their facilities. To train ChatGPT-3, Microsoft’s data center used 700,000 liters of water, or more than double the amount of water needed to produce all the beef that the average American eats in a year. Meta used almost 30 times that amount—22 million liters—to train its LLaMA-3 open-source AI model. In terms of electricity, training ChatGPT-3 used about as much energy as it takes to power 130 US homes for a year. 

Even simple queries generated by the average AI user have an outsized impact on energy and water usage. One search on ChatGPT uses 10 times the amount of energy in a Google search. Asking ChatGPT to generate a single 100-word email requires approximately 16 ounces of water and enough electricity to power 14 LED light bulbs for an hour. 

Since 2021, the number of data centers around the world has increased by almost 30% from approximately 8,000 to 10,655. More than half are in the United States, and they use approximately 4% of all the energy generated in the US. That’s the equivalent of the energy use of 14 million households. In the next five years, that figure is expected to jump to more than 9%. 

China and the EU have already seen massive investments in their data infrastructure, and all signs suggest that accompanying energy needs will skyrocket in the years ahead. Europe currently hosts about 15% of the world’s data centers; by 2030, its data center-related power needs will require an amount of power equal to the current energy consumption of Portugal, Greece, and the Netherlands combined. In Africa, countries across the continent are rushing to build hyperscale data centers to meet their growing AI and data needs. The energy demands projected to accompany these data centers will soon exceed supply by 300%. 

The good news is that AI companies and host nations are implementing a range of initiatives to try to address increasing global demand. Analysts at Goldman Sachs expect significant investments by tech firms in new renewables like nuclear energy. In Europe, renewable energy projects like wind and solar will receive up to 850 billion euros in investment. Additional good news comes in the form of technological innovations that are projected to help meet demand while minimizing energy use. Today’s AI computer chips use 99% less power than 2008 models when performing the same tasks. 

It’s hard not to see the promise of AI as anything less than a historic revolution that will impact lives everywhere. And yet, nothing is guaranteed. Without meaningful adjustments to the energy supply chain or AI models themselves, the revolution could be more modest than we hoped for—and need. Or more damaging than we feared. We’re already facing a future where the world is yearning for more energy. What are we willing to do to get there?

This blog was researched and drafted by Chelsea Acheampong and Caitlyn Shrewsbury.