Why Your Generative AI Tools Are Power-Hungry AI's massive growth drives soaring power and water use in data centres, the surge will push energy consumption from 415 TWh in 2024 to 945 TWh annually by 2030

Tech companies are hogging massive amounts of power to deliver you generative AI.

Think about it -- one ChatGPT request requires about ten times more electricity than a Google search, according to the International Energy Agency (IEA).

Therefore, we're not surprised when OpenAI CEO Sam Altman admitted that writing "thank you" and "please," in ChatGPT were costing "tens of millions of dollars."

Multiple sources estimate that each ChatGPT message produces approximately 4.32 grams of CO₂, while the average query costs about USD 0.36 cents [nearly INR 37]. Note that ChatGPT currently has nearly 800 million users, and set to conclude the calendar year at a whopping 1 billion users.

Just to put things into perspective, a single cigarette has a carbon footprint of 14 grams of CO₂ equivalent.

Why so power hungry?

At the core of everything is the data centers where servers are kept. As you may or may not know these facilities require enormous amounts of electricity to keep the machines running. They also consume vast quantities of water for cooling to ensure smooth operation.

Beyond cooling and electricity generation, semiconductor manufacturing itself is highly water-intensive. The process of producing the specialised chips that power AI requires ultrapure water in massive volumes. According to a recent Morgan Stanley report, semiconductor facilities can consume as much as five million gallons of ultrapure water per day, which is about 19 million litres.

To put this in perspective, in 2020 the Indian Ministry of Housing and Urban Affairs set 135 litres per person per day as the benchmark for urban water supply. Based on this standard, the daily water used by a single semiconductor facility could instead meet the needs of around 140,000 people, roughly equal to the population of a mid-sized Indian city.

What makes this even more concerning is that semiconductor facilities use ultrapure water, which requires extensive treatment. This adds further energy and environmental costs. The same report also projects that annual water consumption will reach around 1,068 billion litres by 2028, representing an eleven-fold increase from 2024.

Another major factor is the sheer volume of user requests. In 2025 alone, there are expected to be about 378 million AI users, up from 116 million just five years earlier, according to KPMG.

The rapid growth in demand is also driving an unprecedented expansion of data centres. Today, there are about eight million data centres worldwide, compared to just 500,000 in 2012. This surge will push energy consumption from 415 TWh in 2024 to 945 TWh annually by 2030, according to the IEA's Energy and AI report.

For context, India generated a total of 1,949 TWh of electricity in FY 2023–24. That means by 2030, the projected energy use from AI and digital technologies alone would amount to more than half of India's current total power generation.

Average data center power consumption

According to a blog post by Enconnex, small data centres consume between 1 and 5 megawatts (MW) of power. To put it simply, 1 megawatt equals 1,000,000 watts. These smaller facilities are often run by banks, hospitals, or telecom companies to handle sensitive workloads on their own premises.

In contrast, large hyperscale data centres, the kind operated by major tech companies to power cloud computing and AI, consume anywhere from 20 MW to more than 100 MW, depending on their size and complexity. For example, China Telecom's Inner Mongolia Information Park is one of the largest facilities in the world, with a capacity of 150 MW when running at full load. It supports up to 50,000 servers and serves major clients such as Alibaba, Tencent, and Baidu.

To understand the scale, a 150 MW data centre's daily electricity consumption could power around 414,000 average Indian homes for a day.

Other tech giants in this space include Microsoft Azure, which has 240 data centres (124 of them operational), Google Cloud, which runs 110 data centres (66 currently operational), Amazon Web Services (AWS), with around 100 data centres worldwide, and Meta Platforms, which manages 24 data centre campuses across the globe.

India's data centre landscape

As of August 2025, India hosts around 190 data centres, with Maharashtra (35) and Tamil Nadu (17) leading the way. In 2024, India's data centre capacity stood at 1.4 GW. For context, one gigawatt equals 1,000 megawatts or 1 billion watts. This capacity is projected to rise to 9 GW by 2030, which would account for about 3 per cent of the country's electricity consumption, up from less than 1 per cent today.

But risks loom large. "Bengaluru, for instance, faces a daily water shortfall while hosting 16 data centres with a combined capacity of 205.6 MW, highlighting the sustainability risks in water-stressed regions," said Viral Thakker, Partner at Deloitte India.

The number of data centres could increase sharply as global tech giants are eyeing India. OpenAI recently launched its first office in New Delhi and, according to Bloomberg, is scouting partners for a massive new 1 GW data centre. Google is planning an eight-storey, 381,000-square-foot data centre in Navi Mumbai by 2027, with an investment of INR 1,144 crore spread over ten years. Microsoft is setting up its fourth Indian data centre in Hyderabad, expected to be operational by 2025.

Rajesh Chhabra, General Manager for India and South Asia at Acronis, believes it will be difficult to ignore the impact on power and water when multinational IT companies establish large data centres in India. "One data centre can use millions of litres of water and 30 to 50 MW of electricity every day," he noted.

What's the way forward?

Experts say the demand can only be met sustainably through renewable energy, smarter cooling systems, and innovative technologies. "Solar power is especially viable in India, with over 300 sunny days annually. Wind energy is strong in coastal states such as Tamil Nadu and Gujarat, while bioenergy offers potential in agricultural regions," Thakker pointed out.

Battery storage technologies are also critical to address the intermittency of renewable energy. Tamil Nadu, for example, has mandated that data centres source at least 30 per cent of their energy from renewables.

Chhabra added that small nuclear reactors, reusing waste heat, and designing longer-lasting equipment could also play a role. "The aim is to enable AI to grow more sustainably and cleanly for everyone."

Michael Sell, Senior VP at the Global Association of Risk Professionals (GARP), emphasised that companies that act early on sustainability will not only reduce environmental impact but also gain a competitive edge.

At the individual level, "getting rid of unnecessary questions" can help reduce energy use, said Chhabra. Thakker echoed this, adding, "Thoughtful prompting, avoiding repetitive or unnecessary queries, lowers both energy and water consumption."

At the organisational level, training, awareness, and responsible AI use policies can encourage efficient prompting and help reduce energy demands across industries.