In late September, Microsoft made a big announcement that they would be reopening the Three Mile Island nuclear plant to power their data centers. This move is just one example of the extraordinary lengths that tech companies are willing to go to meet the energy demands of artificial intelligence, cloud computing, and other technologies. As the world moves towards having nearly 30 billion Internet of Things devices by 2030, up from less than 10 billion in 2020, the need for energy and resources will only continue to grow.
While tech companies are focusing on low-carbon energy plans like nuclear power, they are neglecting to address their rapidly growing water needs. Data centers require massive amounts of water for cooling systems to manage the heat generated by servers. Studies have shown that just a few ChatGPT requests can consume as much as 500 milliliters of water, and this adds up quickly. Google and Microsoft have both seen significant increases in their water usage as they ramp up AI development, with projections showing that by 2027, the amount of water AI uses globally in one year will be equivalent to what a small European nation consumes.
Many data centers are located in water-stressed regions, exacerbating the issue. While some tech companies are investing in water recycling and exploring innovative solutions like transporting seawater inland, one of the simplest and most promising strategies remains largely untapped: rainwater harvesting. Rainwater harvesting has been practiced since ancient times, and it involves collecting rainwater from rooftops and using it for various purposes, including cooling systems in data centers.
Although rainwater harvesting has been restricted in some areas due to concerns about water quality, more states are authorizing the practice as evidence of its conservation benefits grows. Cities like Tucson and Austin are incentivizing rainwater collection, and companies like Apple, Ford, and Toyota have integrated rainwater harvesting systems into their facilities. Data centers, with their large, flat roofs, are prime candidates for rainwater harvesting. A 50,000-square-foot roof can collect about 31,000 gallons of water from just one inch of rain, and many data centers have roofs much larger than that.
The cost of setting up a rainwater harvesting system can be a barrier for some companies, as it typically ranges from $2 to $5 per square foot. Additionally, rainwater systems may not cover the total amount of water needed to cool a data center, especially for large facilities that consume millions of gallons of water daily. However, as the cost and uncertainty of water resources increase, rainwater harvesting is becoming a more appealing option, especially in regions with higher water costs.
Some industry leaders are already implementing rainwater harvesting in their data centers, showcasing the potential of this solution. Google has retention ponds for rainwater harvesting in South Carolina, while Microsoft has implemented a system in Sweden to reduce reliance on local water sources. Amazon Web Services also recognizes the benefits of rainwater harvesting in their water positive strategy.
On a policy level, green banks and federal investments could provide financing for rainwater projects, making it a more accessible option for companies looking to conserve water. Rainwater harvesting has the potential to garner bipartisan support and could even lead to tax credits similar to those for renewable energy projects. As public concern grows about the environmental impact of AI and other technologies, companies will need to consider the financial and reputational risks of not taking action on water conservation.
While it may seem counterintuitive to address a modern global challenge with an ancient technology like rainwater harvesting, sometimes the simplest solutions are the most effective. As the demand for energy and resources continues to rise, tech companies must look towards sustainable practices like rainwater harvesting to meet their needs while reducing their environmental impact.
