Can AI be Sustainable, Really?

Let’s begin with some facts. In 2024, it was found that training a single Artificial Intelligence (AI) model i.e, Chat-GPT-3 led to the consumption of 1,287 MWh of electricity- which could power 120 American households. In fact, it was deduced in 2019 that the carbon emissions while training a large language model were as much as five times the lifetime of an average car.

These statistics are limited to just the testing phase, while in terms of application, a Chat-GPT request is estimated to consume 100 times more energy than a google search. Data centres currently account for 1-1.5% of global electricity consumption, with AI driving this demand further up. In fact, a 2025 report by the International Energy Agency (IEA) established that AI would eventually drive up this consumption to 945 terawatt-hours by 2030, which is slightly more than the entirety of Japan’s consumption.

Naturally, the severity of this impact extends to the usage of water as well. AI models are water-guzzlers and as many countries race to develop an indigenous AI bot system, the water consumption of AI further contributes to uneven distribution of accessible water. AI requires water not just to function, but also to cool down, leading to concerns about its “water footprint”. This rampant usage could lead the water demands of AI to be more than 4-6 Denmarks by 2027. This would come at a time when almost six billion people are already exposed to water scarcity.

On the other hand, AI has also proved to be monumental in transforming the route to sustainability, providing insights and mechanisms to cut down costs, and finding efficient alternatives. Google has been harnessing AI for sustainable initiatives like fuel efficient routing and better optimization of traffic lights. Initiatives like Climate Change AI are also playing a crucial role in integrating machine learning to combat climate crises globally. These instances indicate the double-sworded nature of AI, making the debate about its progress a rather gray subject than plain black or white.

The Paradigm of Equity and Resource Efficiency:

The essential elements of sustainability are recognised to be protection of the environment, social equity and economic prosperity.While AI’s environmental footprint often comes under the radar, the other principles of sustainability are often under-discussed. AI’s socio-economic architecture is equally important as its energy-profile and there is a considerable need for research in this arena.

As a subset of the debate on social equity, access to the internet has gained prominence in recent years. However this debate has also highlighted the conflicting nature of AI and its effect on digital equity, which warrants careful consideration. While the sects of people who do not have access to the internet also lose out on access to AI, many who previously faced barriers related to tech fluency or language proficiency today overcome them with the use of AI.

In 2025, it is established that if used efficiently, AI can break significant barriers of inaccessibility and inequity. Initiatives like Google’s AI Edge gallery are reredefining the contours of the market, as it would enable users to access AI as an offline-available resource. This is a milestone for a demographic like India, where a whopping number of 1.2  billion people are smartphone users but a lot of them are still digitally disconnected. Furthermore, apps like Lookout are making the world safer and more accessible for people lacking eyesight. But to limit the dimensions of equity to just accessibility would be ignorant of other realities. Equity also extends to who makes these systems, who decides the algorithms, and who gets to take part when these algorithms are being set. By avoiding these facets, the goal of access to all would continue to remain a facade, with the same old dichotomies persisting between tech-rich and resource-poor, rural-urban, those who are preferred by the datasets and those whose voices get trampled upon by the same datasets.

When it comes to the economics of AI usage, the paradox of resource efficiency is a key aspect. AI is not a neutral tool. Its development and deployment demand significant proportions of energy, talent, data, and public attention. If these elements are drawn disproportionately from vulnerable ecosystems or marginalised communities, then the economics of AI cease to be sustainable. However, AI has also proven to optimize the demand and supply curve of various business models, by forecasting demand accurately and planning the supply chains efficiently. Thus, while AI itself requires heavy resources, it in its own multifaceted way acts as a significant step towards reducing resource wastage as well.

Segueing to water used for AI, the purpose of which is largely to provide energy and to cool systems as discussed earlier, is currently freshwater. However, the growing progress in terms of AI leads the author of the blog to ruminate over whether the water in data-centres could be renewed and restored. While data-centres have shifted to using non-potable water,  other alternatives in the form of harvested water or wastewater, which includes the water used domestically can be utilised in the AI cycle. In fact, initiatives like Wastewater AI are trying to use AI itself to maximise wastewater treatment efficiency, which could be used at AI centres as well. 

Conclusion:

Ultimately, the reality of whether AI becomes a tool that drains or replenishes our resources, will depend not only on its algorithms, but also on our ambitions. If we treat AI as a public good—designed inclusively, governed equitably, and run on renewable energy and recycled water—it has the potential to be a game changer for our sustainability goals. But at its worst, it can deepen extractive practices, monopolise infrastructure, and draw resources away from communities that need it most. The question, therefore, is not whether AI can be sustainable- but whether we will choose to build and deploy it in ways that align with our shared, long-term survival. AI is here, and it is going to stay. How we- policy makers, law agencies, technology enthusiasts, climate change warriors, national defense and security, growth advocates- decide to engage with it determines how it engages with the users and therefore, its impact on the world.