What is the real impact of AI?
Artificial intelligence (AI) has become ubiquitous, with its potential seemingly limitless. Almost everyone has experience with it by now and has formed an opinion — whether at the kitchen table or at work. In just a few years, we’ve witnessed a full-blown AI revolution, one that shows no signs of slowing down. But what will remain of it in the future? How efficient are these calculation models and tools, really? What do they ultimately deliver? And most importantly, what is AI’s environmental impact?
Bubble or not?
A recent article on nos.nl (the only Dutch source in this article) questions the actual value of AI and suggests we may be on the verge of a bubble bursting and a new economic crisis:
“The growth of large tech companies continues unabated. Companies that have shifted their focus to AI in recent years are experiencing a golden age. Companies such as Microsoft, Google, and Nvidia are performing exceptionally well on the stock market.”
The AI market is booming
The figures are staggering: Nvidia’s market value has increased tenfold in less than three years, rising from over $400 billion to $4 trillion. The five largest American tech companies are now even bigger than the combined economies of the Netherlands, Germany, France, the UK, Italy, and Spain.
Massive investments in AI development are also transforming the global landscape. Gigantic data centers are being built worldwide — I drove past one near my home yesterday. Nature, of course, pays the price: entire sections of forest often disappear to make way for these facilities.
Jelle Zuidema, associate professor of explainable AI at the University of Amsterdam, notes in the same NOS article that more and more experts are questioning whether AI will quickly lead to increased productivity and significant cost savings.
But shouldn’t the discussion focus more on AI’s impact on the climate? Aren’t we repeating the same pattern as modern societies — prioritizing economic growth while the world literally burns and climatologists sound the alarm?
AI models have enormous emissions
Plan B Eco, a Polish startup focused on sustainable living and planetary protection, highlights the issue in an extensive article: global data centers account for 2.5% to 3.7% (source: 8 Billion Trees) of worldwide greenhouse gas emissions — more than the aviation industry. Training a single large AI model can emit around 300 tons of CO₂.
Each ChatGPT search consumes nearly 10 times the electricity of a typical Google search, often powered by fossil fuels like coal and gas — the primary drivers of climate change. That’s not what we signed up for.
And we haven’t even discussed water consumption. Dr. Camilla Pang has investigated this issue and discovered GreenPT, a Dutch AI tool aiming to compete with ChatGPT and Mistral in a sustainable way. Major media outlets, including The Guardian, have also covered this topic.
Beyond CO₂ emissions and water use, we must consider the rare earth metals mined for AI production — and the often low-paid workers who sustain these unsustainable practices.
How much water is actually consumed?
According to a study by the University of California, every time you run an artificial ChatGPT query, you consume an increasingly scarce resource: fresh water. For every 20 to 50 queries, approximately half a liter of water is lost from our overburdened reservoirs in the form of steam emissions. Data centers consume water by using electricity from steam power plants and by operating cooling systems to keep their servers at safe temperatures. In the U.S. alone, Google’s data centers are estimated to have used 12.7 billion liters of water in 2021 for cooling.
According to the article, a two-week training session for the GPT-3 AI program in Microsoft’s state-of-the-art U.S. data centers consumed approximately 700,000 liters of fresh water — almost as much as the production of about 370 BMW cars or 320 Tesla electric vehicles. Consumption would have tripled if the training had been carried out in Microsoft’s data centers in Asia, which are less efficient.
The university researchers therefore argue that large technology companies must take responsibility and set an example by reducing their water consumption. “It is truly crucial to expose and address the hidden water footprint of AI models, given the increasingly serious crisis of freshwater shortages, prolonged droughts, and rapidly aging public water infrastructure,” according to the article titled “Making AI Less Thirsty”.
GreenPT as a Sustainable Alternative
Fortunately, some companies recognize the importance of this issue and are committed to minimizing the environmental impact of their AI models. While ChatGPT lacks a proactive sustainability policy, Mistral AI takes measurable steps through life-cycle data, open reporting, efficiency guidelines, sustainable hardware, and strategic partnerships.
Then there’s GreenPT, a true game-changer, alongside its equally sustainable data center partner, Leafcloud. Unlike traditional providers, Leafcloud has designed a digital infrastructure that works with — rather than against — the environment. As Dr. Camilla Pang notes in her article, they offer a bold vision for powering future cities.
How do they do it? By reusing energy — capturing heat from data centers to warm communal swimming pools, showers, and buildings. Brilliant, right? We need more of this — and greater transparency — so consumers and businesses can make informed choices.
Quality of AI Tools
How do these chat tools compare? I tested all three for text-related advice, error correction, and topic-based queries. None are perfect: none caught the spelling error “financieën” in my native language, Dutch. When I asked for verification, ChatGPT and GreenPT confirmed it was indeed incorrect, while Mistral still claimed it was correct — so it didn’t recognize the word itself.
Beyond that, the differences are minor. Mistral excels at stylistic suggestions, while ChatGPT and GreenPT proactively offer follow-up recommendations and additional sources — without needing specific prompts.
GreenPT: The Most Sustainable Choice
Currently, GreenPT operates on a subscription model, with the cheapest plan at €4.50/month for 15 daily prompts — ideal for personal or workplace use. Their business subscriptions remain affordable, but paying for sustainability feels counterintuitive. Ideally, these companies should receive government and financial support to operate for free — as the world doesn’t have time to wait. For now, though, we must work with what we have.
Sources:
NOS article (in Dutch): https://nos.nl/artikel/2586573-is-de-euforie-rond-ai-een-bubbel-die-op-knappen-staat
Arxiv: https://arxiv.org/pdf/1906.02243
The Guardian: https://www.theguardian.com/commentisfree/article/2024/may/30/ugly-truth-ai-chatgpt-guzzling-resources-environmentgreenpt
Goldman Sachs: https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand
Dr Camilla Pang: https://camillapang.substack.com/p/my-experience-with-greenpt
University of California: https://news.ucr.edu/articles/2023/04/28/ai-programs-consume-large-volumes-scarce-water
Sustainability Content Agency writes, edits, and translates content on sustainability, social impact, and health in almost all European languages.