The next time you type a prompt into an AI chatbot, consider this: that single interaction consumes roughly 29 milliliters of water. Two tablespoons. It sounds trivial until you multiply it by 2.5 billion, which is the number of AI prompts processed globally every single day.
The numbers stop being trivial very quickly.
A landmark report published this month by the United Nations University Institute for Water, Environment and Health has put a precise and alarming figure on what the AI boom is costing the planet. By 2030, global data centers powering artificial intelligence are projected to consume 945 terawatt-hours of electricity, nearly triple the combined annual use of Pakistan, Bangladesh, and Nigeria combined. Their associated water footprint will equal the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa.
That is not a future scenario. It is a trajectory already in motion.
The Thirst Behind Every Query
Training GPT-4 alone required approximately 600 million liters of water, equivalent to filling 237 Olympic-sized swimming pools. Future systems of similar scale could require close to one billion liters just for training. And that is before a single user types a single prompt.
The more complex the task, the worse it gets. Generating a single AI image requires 1,450 times the energy of a simple text classification task. A short AI-generated video can consume as much electricity as processing 200,000 spam emails. A highly complex video can use energy equivalent to powering an LED bulb for 42 hours.
The water problem is structural. Most data centers rely on evaporative cooling towers that draw enormous volumes of freshwater to keep servers from overheating. Larger data centers can require up to 5 million gallons of water per day, equivalent to the needs of a city of 50,000 people. Roughly two-thirds of data centers built since 2022 have been located in water-stressed regions.
Nvidia’s Answer: Stop Using Water Entirely
On Monday, Nvidia moved to position itself as the solution to a crisis it has, in large part, helped create.
The company announced that its newest AI servers will use liquid cooling exclusively, eliminating the need for air-cooling fans that rely on water. Heat will instead be dissipated by a liquid coolant made of water and propylene glycol, recirculated in a closed loop that does not need to draw in new water.
In favourable climates, Nvidia’s 45-degree liquid cooling architecture can reduce facility cooling water consumption from roughly 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero, a potential 100% reduction in water use. A 50-megawatt hyperscale facility could save over $4 million annually in cooling-related energy and water costs by shifting to this infrastructure.
“The water consumption challenge for data centers is largely solved,” said Josh Parker, Nvidia’s chief sustainability officer.
A Step Toward Sustainability, But Not a Silver Bullet
The transition to liquid cooling solves two critical crises for the tech industry: it enables the continued scaling of frontier models without melting the hardware, and it provides a pathway for tech giants to chase their ambitious 2030 water-positive climate pledges. However, environmental researchers and a recent UN report caution against viewing this hardware upgrade as an absolute cure for AI’s ecological footprint.
While the theoretical savings are massive, the real-world application deserves intense scrutiny. Upgrading global infrastructure to support direct-to-chip liquid cooling will take years, meaning legacy data centers will continue evaporating millions of gallons of water using older HVAC technologies in the interim. Furthermore, the highest efficiency gains—where water use drops to near zero—are largely restricted to cooler climates. In hotter regions, data centers will still have to rely on water-consuming backup chillers to regulate ambient temperatures during peak summer months.
The most profound concern raised by the UN is an economic one. Historically, as technology becomes more efficient, it becomes cheaper. Cheaper, highly efficient AI will inevitably drive a massive surge in global adoption.
This creates a paradox, as Nvidia’s chips become better at conserving water per prompt, the sheer volume of users, prompts, and newly constructed data centers will multiply. Ultimately, the aggregate environmental footprint of the AI industry risks growing far larger, easily outpacing the localized efficiency gains brought on by liquid cooling. While hardware innovation is a necessary step, the broader challenge of AI’s insatiable resource consumption remains far from solved.
