If AI wants to develop sustainably, it must first secure this key resource
The development and deployment of AI systems requires large amounts of computing power, which in turn generates large amounts of heat energy that must be dissipated to avoid overheating, and most major technology companies rely heavily on water for cooling. As AI continues to develop at an alarming rate, the demand for water has surged, resulting in a surge in water consumption by AI companies and raising concerns about sustainability, highlighting the urgent need for the AI industry to significantly reduce its water consumption.
For example, OpenAI’s GPT-3 model uses 45 TB of text data for training. To process such a large amount of data, the company relies on tens of thousands of computer resources that run around the clock. This huge computational intensity translates into extremely high energy requirements, sometimes requiring as much as tens of thousands of homes to train a natural language model. Such systems dissipate more heat than state-of-the-art air-cooling methods can handle, so water-cooling systems have become standard practice for temperature regulation in AI data centers.
Recent analysis estimates that OpenAI’s GPT-3 training system may consume more than 700,000 liters of water. With the exponential growth of AI chat tools and other emerging AI applications, water costs can quickly accumulate for even a seemingly insignificant single AI application.
In 2022 alone, Microsoft’s global water consumption will increase by 34%. This is largely due to the expansion of the computing power of the OpenAI model. From 2019 to 2023, Google’s water demand increased by 60%, with more than 70% of the water dedicated to supporting the cooling of AI data centers. As generative AI applications proliferate in the consumer and enterprise space, these numbers will continue to grow at an even faster rate.
To alleviate these pressures, data centers implement alternative evaporative cooling methods that do not result in loss of potable water, and some companies also recycle used water in the data center. However, the amount of recycling pales in comparison to the depletion of lakes, rivers, reservoirs, aquifers, etc. Even if cooling water is switched from drinking water sources to non-potable water sources, it cannot fundamentally solve the industry’s huge water demand.
Realize true AI water neutralization and move toward sustainable development
About half of the world’s population already faces seasonal or year-round water shortages, and persistent drought in major breadbasket regions threatens food supplies. As climate change intensifies pressure on global water resources, meeting basic human needs and sensitive ecosystems takes priority over discretionary uses such as cooling data centers, but conflicts and disputes are inevitable.
While AI holds the promise of solving some sustainability challenges, its water needs present a paradox as to whether addressing other environmental issues such as energy efficiency and carbon emissions takes precedence over water risks. Direct water measurement and reporting is still sorely lacking, and most AI models trained for environmental impact reporting are narrowly focused on easy-to-track metrics such as server energy use and related greenhouse gas emissions. They rarely include water consumption. , concerns related to water as a basic human right are taken into account.
The awakening of public awareness also has a significant impact on responsible water management. People who interact with AI systems often ignore the large power, cooling, and freshwater needs behind them. How to make potential water consumption visible so users can consider limiting unnecessary AI application usage, similar to how drivers can be prompted to make more efficient route choices and gasoline usage.
Achieving true AI water neutralization will require scaling up alternative cooling methods, promoting rainwater harvesting and wastewater recycling, ensuring supply chain efficiency, and exploring other technologies. At the same time, the most feasible way for the sustainable development of AI is to limit the random use of AI and improve AI efficiency.
Thanks for your reading. Share your thoughts, and suggestions, and help shape a better experience. If you find it inspiring, share it with your friends give it a ‘clap’ and follow. Let’s build something great together — drop your comments below!
