ChatGPT requires half a liter of water to have a conversation: The blue tech conflict
As droughts become increasingly widespread, there will be geopolitical escalation over a resource that is not only essential for life but also for technological competition
Artificial intelligence is indisputably going to become a part of our lives. Technology giants like Microsoft, Google and Amazon are heavily invested in it, integrating it into everything from search engines to digital tools. How can we understand this new reality’s impact on society? A recent study has drawn attention to the environmental cost of such models, especially their water footprint. Although that might seem like a secondary issue, it is not. It is well known that water will be the cause of the next global conflicts; in fact, there are already tensions over water resources within countries. As drought spreads, there will be a geopolitical escalation over a resource that is not only essential for life but also for technological competition. That must be factored into the equation.
When developing technology, a sustainable perspective is critically important. GPT-3 and GPT-4, the neural models behind popular chatbots, and other technologies like cryptocurrencies, have come under fire for their carbon emissions. But what about their water consumption? ChatGPT requires half a liter of water to have a 50-question conversation. Let’s crunch the numbers based on how many users there are.
Comparisons with other industries highlight the extent of this problem. State-of-the-art U.S. data centers have consumed 700,000 liters of clean fresh water for training artificial intelligence. That’s enough to produce 320 Tesla cars. And it would be worse if it were in the centers in Asia because consumption would triple there. Beef production and denim manufacturing also consume a lot of water, but their footprints factor in their entire life cycle and include a large share of non-potable water. Adding the water consumption associated with manufacturing and transporting AI servers, the overall water footprint could increase 10-fold.
But all is not lost. Once you are aware of this issue, you can optimize consumption by considering the system load and geography. Doing so is very similar to how we reduce our electricity bills by playing with prices at night and during off-peak hours. However, carbon reduction and water conservation can sometimes conflict. For example, in California, solar energy production is high around midday, resulting in peak carbon-saving hours. But the outdoor temperature is also high at that time, yielding the worst water efficiency. Thus, if we only consider carbon footprint reduction, we may end up with higher water consumption. Conversely, if we aim to reduce water consumption, we could increase the carbon footprint because there’s less solar energy available.
Legislation and large producers of artificial intelligence should immediately incorporate sustainability into the necessary conditions for technological development. That’s especially true for a resource as scarce as water. If technological warfare (the race between nations and companies for technological dominance) intersects with water wars (conflicts over scarce water resources), a new battlefield will emerge in a world that is already experiencing enough crisis without adding new conflicts between different actors.
Water is becoming a scarce commodity, and competition for access to this vital resource is likely to intensify as the effects of climate change worsen. If technology companies continue to expand their data centers around the globe, the site selection process must start to consider water usage in the selection process.
Placing data centers in regions where water is scarce can contribute to local water stress. Even if these centers are built with state-of-the-art water recycling systems, the large volume of water required to keep them operational could be detrimental in drought-affected areas. In addition, the water used in these data centers must be clean fresh water. In a country like Spain, for instance, which already suffers from water scarcity, diverting this valuable resource could lead to undesirable effects on different water uses, such as agriculture and consumption.
All of this underscores the interconnectedness of our global challenges. It forces us to think more holistically, recognizing that our pursuit of technological advances must not come at the expense of our planet’s resources. As we navigate this new battleground, success won’t just be measured by technological capabilities, but also by a commitment to sustainability and the equitable distribution of resources. Why not start a new race to reduce consumption and environmental impact? Nearly 3 billion people on the planet experience water stress; and that’s before the new generative artificial intelligence revolution. The future of technology must be sustainable as well as innovative. After all, what good is artificial intelligence if it endangers the very resources we need to survive?
Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition
