The Real Cost of Asking AI

The recent AI Summit in Delhi has dominated the news for many reasons, and that’s also what prompted me to join the conversation. But, instead of focusing on the technological marvels of artificial intelligence, I want to draw attention to a less discussed but equally important aspect, which is the sustainability cost of AI.

There is no debate that AI is a major disruptor in our lives. Everyone seems to be using it in some way or the other, at least one among its many free and paid versions. It functions like an ever-obedient assistant, efficient and fuss-free, ready to comply with endless patience. What concerns me, however, is how casually and unmindfully we are beginning to rely on it.

Recently, a friend mentioned that she wanted to make a simple soup at home because she had come down with a cold. She wasn’t talking about an exotic or extensive recipe. All she wanted was a basic soup, the recipe for which can easily be found anywhere on the Internet. Yet, her instinctive response was to ask ChatGPT for the recipe. I was struck by how effortless this reliance on AI has become.

There are countless examples of people using AI platforms for routine, trivial tasks that we once managed through memory, books, conversations with friends, teachers, or a quick Google search. What many don’t realise, however, is that AI does not live in “the cloud”. What it relies on is vast physical infrastructure: data centres that require enormous amounts of water, land, and electricity. Every 100-word AI prompt requires roughly half a litre of water, primarily for cooling systems. A single data centre can consume as much water as a city of 30,000 people. By 2030, global data centres are projected to require 350 billion litres of water annually. In a water-stressed country like India, where 163 million people still lack access to safe drinking water, this is very troubling.

AI also carries a carbon footprint that remains largely invisible and underdiscussed. Data is lucrative, and its environmental cost is conveniently overlooked. But this impact is very real. Researchers at the University of Massachusetts conducted a life-cycle assessment of training large AI models, particularly those designed to process and generate human language. The findings were sobering. Training one such model emitted 626,000 pounds of CO₂, which is nearly five times the lifetime emissions of an average American car. As models grow larger and more complex, the computational and environmental costs rise proportionately. In many ways, AI appears to be following the same trajectory as oil and gas. It has become difficult to live without it, but the cost to the environment is just as severe.

As someone who has long advocated reducing carbon footprints, especially in the way we eat, drink, and live, it’s time for me to factor AI into that conversation as well. I’ve written extensively about sustainable food practices, and I know how difficult it is to shift mindsets. Even recommending locally produced oils like mustard or sesame, both healthier for people and the planet, is often met with resistance when compared with olive oil.

AI platforms, particularly the free ones, surely have us hooked. But we know well enough that there are no free lunches. The costs may not be immediately visible, but they exist in the form of environmental degradation and carbon emissions. This goes far beyond simple data-harvesting strategies and raises serious questions about sustainability.

Moderation is a principle we apply to many aspects of life, and perhaps it’s time to extend that wisdom to AI. Using AI for complex challenges, such as protein-folding models, cancer drug research, or large-scale scientific problems, makes sense and can significantly advance human welfare. But using it indiscriminately for trivia may not.

This brings us to a more fundamental question: Do we need to place checks on how we use AI in our everyday lives? Just as we no longer remember phone numbers because they’re stored on our devices, are we gradually outsourcing our everyday thinking to machines? Emerging research suggests this may already be happening, with early evidence pointing to subtle declines in analytical and critical thinking when AI is overused.

And finally, let me return to the issue that prompted this piece in the first place—water. When we exhaust a finite resource without restraint, the consequences are inevitable. Preparing for an already depleting water future requires conscious choices. Perhaps the way forward is discernment: knowing when and how to put AI to work, and when not to.