Generative AI and climate change are on a collision course

In 2025, AI and climate change, two of the biggest social disruptors we face, will collide.

The summer of 2024 broke the record for the hottest day on Earth since data collection began, sparking widespread media coverage and public debate. This is also the year that both Microsoft and Google, two major big tech companies investing heavily in AI research and development, missed their climate goals. While this also made headlines and sparked outrage, the environmental impacts of AI are still far from common knowledge.

In reality, the current “bigger is better” AI paradigm, epitomized by technology companies’ pursuit of ever larger and more powerful language models presented as the solution to every problem, comes with costs. very significant for the environment. These range from generating colossal amounts of energy to power the data centers that run tools like ChatGPT and Midjourney to the millions of gallons of fresh water that are pumped through these data centers to ensure they don’t overheat and the tons of metals of rare earths. necessary to build the hardware they contain.

Data centers already use 2 percent of global electricity. In countries like Ireland, that figure amounts to a fifth of the electricity generated, prompting the Irish government to declare an effective moratorium on new data centers until 2028. While much of the energy used to power data centers Data is officially “carbon-neutral”, this is based on mechanisms such as renewable energy credits, which technically offset the emissions incurred when generating this electricity, but do not change the way it is generated.

places like Data Center Alley‘ in Virginia run primarily on non-renewable energy sources such as natural gasand energy providers are delaying the retirement of coal-fired power plants to keep up with the greater demands of technologies such as AI. Data centers are sucking up huge amounts of freshwater from scarce aquifers, pitting local communities against data center providers in locations ranging from Arizona to Spain. In TaiwanThe government chose to allocate precious water resources to chip manufacturing facilities to get ahead of growing demands rather than allowing local farmers to use them to irrigate their crops amid the worst drought the country has seen in more than a year. century.

My latest research shows that switching from older standard AI models (trained to perform a single task, such as answering questions) to the new generative models can consume up to 30 times more energy just by answering the exact same set of questions. Tech companies that are increasingly adding generative AI models to everything from search engines to word processing software also fail to disclose the carbon cost of these changes; We still don’t know how much power is used during a conversation with ChatGPT or when generating an image with Google Gemini.

Much of the discourse from big tech companies about the environmental impacts of AI has followed two trajectories: either it’s not really a problem (according to Bill Gates), or an energetic breakthrough will come and magically fix things (according to Sam Altman). What we really need is more transparency around the environmental impacts of AI, through voluntary initiatives such as the AI Energy Star project I am leading, which would help users compare the energy efficiency of AI models to make informed decisions. I predict that by 2025, voluntary initiatives like these will begin to be implemented through legislation, from national governments to intergovernmental organizations like the United Nations. In 2025, with more research, public awareness and regulation, we will finally begin to understand The environmental footprint of AI and take the necessary actions to reduce it.