Artificial Thirst

The secret depletion of already perilously low reserves of drinkable water

AI poses a serious environmental risk, one that also so far endures virtually no regulation whatsoever. Industrial thirst invokes obvious thoughts about consumption—that of factories and manufacturing processes—but conceals a lesser known but imminently more nefarious glutton: Artificial Intelligence. Data centers supporting AI operations withdraw enormous amounts of water from the local community, and in areas of minimal infrastructure the shortage of drinkable water is significantly exacerbated.

Freshwater scarcity has become one of the most pressing challenges shared by all of us in the wake of the rapidly growing population, depleting water resources, and aging water infrastructures.

This statement, made in a recent paper on the consumption of freshwater by AI systems, aptly describes an imminent—but under-discussed—environmental problem. The United Nations has identified several key causes behind the worsening problem of water scarcity: finite supply under growing demand; climate change and melting ice; and insufficient water resource management. Demand upon a dwindling supply derives from economic expansion, which is accompanied by urban development, industrial consumption, and increased population.

To date, the largest consumer of water by sector remains agriculture, distantly followed by industry. Mark Giordano of Georgetown University states that scarcity typically has less to do with the actual presence of potable water and more to do with infrastructural access to and management of it. Because agriculture consumes nearly 70% of business-consumed freshwater—and agriculture, climate issues, and poor water management go hand-in-hand in lower income areas of the world—the poor routinely suffer from water crises the most.

But another threat to the world’s drinkable water is emerging—consumption by the systems powering artificial intelligence.

Global water crisis

Currently, around 2.2 billion people face water issues, including over 450 million children. The issue is most severe in the Middle East, North Africa, and India. Water allocated to industry leads to inequality because many advanced countries import it from developing countries that already suffer shortages due to lack of infrastructure. The UK, for example, is the 6th “largest importer of virtual water in the world… over half of the UK’s water footprint comes from countries facing water scarcity.”

Source: Council on Foreign Relations

As the Himalayan glaciers melt more swiftly than they have over the last several hundred millennia, the water crisis in its downstream regions will exponentially worsen. Nearly 2 billion people rely on Himalayan ice and snow for drinkable water. Like the Himalayan glaciers, polar ice has been declining at an alarming rate. Since the 1990s, the ice caps have been melting 6x times faster than previously, when the ice sheet existed near a state of balance.

Between 1992 and 2019, for example, the Greenland ice sheet lost around 3,900 billion tons of ice. The Antarctic ice sheet is losing ice cover at nearly the same rate. Melting ice sheets affect the cyclical flow of rivers, which will negatively impact downstream communities and ecosystems. Another concern beside the impact on volume is that the various pollutants deposited into the ice sheets over the last century or two will re-enter the freshwater stream.

Vincent St. Louis, professor in the Department of Biological Sciences at the University at Alberta, is currently studying the affect of glacial melt on lakes and rivers in the Canadian Rockies. Describing some of the concerns of his team, he gave this example: “DDT that was used back in the 1950s has been deposited in these glaciers and locked in the ice. Now that they're melting, there is the potential that DDT will be released into the drinking water.”

Technology’s artificial thirst

New technologies that purport to introduce some sort of societal advancement often come with dark sides. An example I have written on before is cryptocurrency. There, I noted that “Just to mine Bitcoin—a single one among many cryptocurrencies—miners consume about 110 Terawatt Hours [of electricity] per year, or ‘0.55% of global electricity production, or roughly equivalent to the annual energy draw of small countries like Malaysia or Sweden’.”  That electricity is in many cases supported through coal production, one of the filthiest forms of energy still in use. Proponents of cryptocurrency espouse its ostensibly only benefit as follows:

The biggest advantage of cryptocurrency is that it's not owned by a single financial or government entity. This eradicates the monopoly of money and ensures cryptocurrency value isn't dictated by a central bank or authority.

This is not compelling given its instability that renders it a generally losing bet. Moreover, consuming over 1% of global electricity (across all crypto mining) hardly seems an adequate tradeoff for a platform as amenable to scammers as it is to legitimate investors.

Another technology slated to “make everything we care about better,” according to venture capitalist Marc Andreessen, is Artificial Intelligence. Andreessen is one among many who have fought tooth-and-nail against the regulation of AI to protect its sweeping theft of intellectual content from the internet, among other reasons, thereby preserving Andreessen’s billions in profits earned now and later.

In the paper from which this article opened, the authors noted the following disconcerting fact: “[T]he global AI demand may be accountable for 4.2 – 6.6 billion cubic meters of water withdrawal in 2027, which is more than the total annual water withdrawal of [] Denmark or half of the United Kingdom.” Furthermore, “training GPT-3 in Microsoft’s state-of-the-art U.S. data centers can directly evaporate 700,000 liters of clean freshwater, but such information has been kept a secret” [emphasis mine].

Data centers powering AI consume water both for electricity generation and on-site cooling. Before even accounting for actual water use, the mere evaporation of water by these data centers’ various processes consumes 0.18 billion cubic meters, more than the total annual water withdrawal of Liberia (a country of about 5 million people in West Africa).

For cooling, these centers require clean, freshwater to avoid clogging their systems with mineral deposits or bacterial growth, removing viable drinking water from the population that needs it. Cooling is necessary because of the intensive energy requirements of these facilities, which produce massive amounts of heat. Generation of these enormous quantities of power also consumes equally large amounts of water.

Neither Google nor Microsoft disclose their water usage toward this purpose, but Meta reported 3.58 L/kWh for its operations. The United States national average, by comparison, is 3.1 L/kWh. One LLM training run, “BLOOM,” “consumed enough energy to power the average American home for 41 years” (page 120). **A training run is simply one cycle of data-sweeping followed by its application to the AI process to “train” it to provide better outputs. For more on this, see here.

Sample of AI’s water usage. Source: Making AI Less ‘Thirsty’: Uncovering and Addressing the
Secret Water Footprint of AI. Models

Year-upon-year, companies engaged in operations related to AI see extraordinary increases in their water consumption. Just in 2022, the water intake by Microsoft and Google grew by 34% and 20%, respectively, compared to the previous year. For Microsoft, that 34% uptick meant the consumption of nearly 1.7 billion gallons of freshwater. Shaolei Ren, a researcher at the University of California, Riverside, calculates that for every 5 to 50 prompts or questions, ChatGPT imbibes the equivalent of a 16 oz. bottle of water.

The study referenced herein relies on estimations because most tech companies do not bother to disclose specific numbers. It sums up this transparency problem this way:

To exploit the spatial-temporal diversity of water efficiency, it is crucial to have better visibility of the run-time water efficiency and increase transparency by keeping the AI model developers as well as end-users informed. Nonetheless, such data is often lacking. For example, even scope-1 water usage (either withdrawal or consumption) is not included in today’s AI model cards [citation omitted], not to mention the scope-2 water usage. Additionally, there is very limited data available for embodied water usage by chip making, which adds challenges to a holistic lifecycle view of AI’s water footprint.

We recommend AI model developers and data center operators be more transparent. For example, what are the runtime (say, hourly) on-site scope-1 WUE and off-site scope-2 WUE? What about the water footprint of AI models trained and/or deployed in third-party colocation data centers? Such information will be of great value to the research community and the general public. As the first step, we recommend that the scope-1 and scope-2 water usage information be included in AI’s model cards.

Professor Ren, cited above, notes that “If you’re not aware of the resource usage, then there’s no way that we can help conserve the resources.” OpenAI, on the other hand, merely says “We recognize training large models can be energy and water-intensive” and that it will give “considerable thought” to water conservation. Google claims, “Wherever we use water, we are committed to doing so responsibly.”

Many articles focus on using AI to make existing water systems more efficient. This seems pointless if the efficiency improvement reaches only a fraction of what AI actually consumes. Without transparent reporting by AI companies, we will never know. This survey of AI’s apparent exorbitant consumption of water—brief as it necessarily is because of the dearth of information—reinforces the point I have harped upon from early on: Artificial Intelligence’s development and deployment must be regulated.

We cannot rely on fraudsters like Elon Musk or arrogant wealth-hoarders like Marc Andreessen to determine how society coexists with this technology. Nor can we depend upon AI companies’ self-serving statements about their “responsible” consumption practices. AI is here to stay and will inevitably grow by leaps and bounds as various industries continue to incorporate it. This means that its consumption of drinkable water will likewise increase year-to-year.

Billions of people already struggle to find sufficient water merely for survival. All the numerous other problems with the uncritical push to commercialize AI pale in comparison to this one. Governments failing to address these issues are nothing less than grossly negligent. Back in March of 2023, I summed up the argument in a way that I believe remains prescient.

The advent of increasingly advanced AI… necessarily demands a reimagination of what it means to be successful in society. It requires a reevaluation of how technology is implemented, who controls it and for what purpose. Right now, a tiny sliver of the world’s population makes these choices. And their record sucks.

* * *

I am the executive director of the EALS Global Foundation. You can follow me at the Evidence Files Medium page for essays on law, politics, and history, like the Evidence Files Facebook for regular updates, or Buy me A Coffee if you wish to support my work.