At the start of the year, Ejay Hernandez posted a video saying: “You need to use ChatGPT for your grocery lists.” “It is a lifesaver,” Hernandez said in the video, which now has almost ten million views.

Only five days later, he took it all back. “I didn’t know about the environmental implications,” Hernandez reasoned in a second video. The comments were then flooded with people claiming the same thing, writing that they had “no idea” that training one AI model produces the same amount of carbon dioxide as five cars in their lifetime. So, as more people turn to ChatGPT to plan their weeks, create fitness plans and even set their New Year resolutions, what is the true cost of our growing dependence on AI chatbots?

January was a breakthrough month for those who, like Hernandez, hadn’t considered AI’s environmental impact. The wildfires across Los Angeles sparked a conversation online about how AI technology contributes to global warming – which then turned into an exaggerated rumour that ChatGPT “caused” the fires. There’s also the release of DeepSeek, a free AI-powered chatbot created by a Chinese startup that looks and works similar to ChatGPT, which has spun out the entire AI industry: it not only costs significantly less to build, but also claims to use far less energy than its competitors (which could slash energy use). 

During what could be considered a new peak in AI environmental panic, artist and activist Matt Bernstein posted a quote that went viral on Instagram: “Somewhere, the men who build AI chatbots are selecting the interiors for the rocketships they will use to leave earth and all of us burning with it”. It’s a grim picture, considering that Elon Musk and his Department of Government Efficiency are proposing to overhaul the US government using AI. However, Shaolei Ren, associate professor of electrical and computer engineering at the University of California, Riverside, cautions against making sweeping statements linking ChatGPT to any wildfires. “I think that connection is really remote,” he says. “When we increase energy usage, we generate more carbon emissions, and this has a long-term impact on climate change, but the current carbon emission contribution from AI is less than one per cent.”

As with all things climate crisis-related, AI’s role is part of a deeply scientific picture that remains hard to focus our attention on. Sure, “ChatGPT alone ruined LA” makes for a compelling (but false) story, but the impact AI has on the environment is far more nuanced. For example, Ren calls the claim that one ChatGPT uses ten times the energy as a Google search “anecdotal” as they are still figuring out ways to measure these buzzy examples. “There’s no official data to confirm this number, but I think the general order is pretty reasonable,” he says. “It just highlights the much, much heavier resource usage of AI computing than other types of general workloads.” After all, all energy consumption has a negative environmental impact since the US still relies heavily on fossil fuels.

I think that, nowadays, young people have gotten used to using ChatGPT as a constant companion. Instead of doing a Wikipedia search for the capital of the country, they’ll ask ChatGPT, and all of this adds up

While most of the focus remains on AI’s impact on climate change, Ren has also been exploring how “thirsty” AI is. “This is also a water consumption problem because this energy will be converted into heat, and the most efficient way to get rid of the heat is actually using water evaporation,” he says. Then there’s the even lesser-discussed issue of public health. A 2024 study that Ren co-authored found that training an AI model can produce air pollutants equivalent to more than 10,000 round trips by car between Los Angeles and New York City. “These particles can get into people’s lungs and create asthma, coughing, heart attacks and other bad health outcomes,” says Ren. The study found that air pollution stemming from AI disproportionally affects certain low-income communities (especially those close to power plants or backup generators at the data processing centres). 

Dr Sasha Luccioni, AI and climate leader at Hugging Face, says she started out initially researching ways that AI could tackle climate change when she realised it was part of the problem. “There are really cool ways it can be used: everything from detecting deforestation from space to proposing new combinations of molecules for batteries and solar panels to tracking a specific species in a rainforest in real-time,” she says. When asked if the pros of AI outweigh the negative environmental impacts, she says it’s “not a fair comparison”. “AI is just a blanket term,” she says. “It’s like asking if transportation is good or bad.” It’s true that not all AI is the same, and using ChatGPT for a recipe or targeted advertising is not the same as harnessing the technology for potential environmental solutions. 

For Luccioni, one of the most concerning factors when it comes to AI’s environmental impact is the increased usage of AI in superfluous ways. “I think that, nowadays, young people have gotten used to using ChatGPT as a constant companion,” she says. “Instead of doing a Wikipedia search for the capital of the country, they’ll ask ChatGPT, and all of this adds up.” This is a growing problem, with the share of teens using ChatGPT for their schoolwork doubling since 2023. Ultimately, each incrementally small action usage uses a relatively small amount of energy, but this is no longer the case once everyone uses ChatGPT for everything – all day, every day. This is especially true when you factor in that generative AI models (like ChatGPT) are a lot more power-hungry than extractive ones (a form of natural language processing that uses machine learning to identify patterns).

Perhaps Hernandez’s hunch was correct when he immediately stopped using ChatGPT to create weekly grocery lists. While a comprehensive understanding of the impact of AI on the planet is still being compiled as the technology develops, one thing seems to be abundantly clear: you probably shouldn’t be using it for every question, query or mundane task. “We really don’t need to be using ChatGPT for absolutely everything, the way people are now,” says Luccioni, who is currently developing a first-of-its-kind energy star rating for AI. Once people have more specific data (and an energy rating), researchers like Ren are hoping that people will start running a cost-benefit analysis in their minds. “I think if we’re given the knowledge, given the water consumption, given the energy usage and maybe even the health impact, some people may not use AI for day planning,” he says. “They will, hopefully, be using that for more meaningful purposes.”