Data centre surge strains resources: Can AI learn to consume with care?

Big Tech is pumping more than $60 billion into data centres in India to power a smart future. This will also put pressure on scarce power and water resources. Fortunately, there are solutions

The moment you type in a query and press enter on an artificial intelligence (AI) platform to get an image, infographic, text or video, the query goes to a football field-sized data centre where racks of power-hungry AI chips have a quick “think” before replying. This process, called “inference” — where a machine applies its knowledge (or learning, if you will) to answer the query — guzzles up power and water to generate an output. 

Most technology companies are reluctant to share exactly how much electricity and water these facilities consume, even as millions make AI their daily assistant. 

Sam Altman, chief executive officer (CEO), OpenAI, offered a rare glimpse in a blog- post titled The Gentle Singularity earlier this year, noting that an average ChatGPT query consumes about 0.34 watt-hours of electricity and 0.000085 gallons of water (roughly one-fifteenth of a teaspoon of water). Or, just a tiny drop of water, and about as much power as a microwave might consume in a little over one second — for a simple text query. Videos and images consume more. 

That’s insignificant at the level of a single prompt. But multiply that by billions of daily queries and tasks — from simple questions to complex tasks, like managing an intelligent robot factory or drug discovery—and you are talking about city-sized power and water consumption to make machines manufacture intelligence. 

As it happens, power-hungry and water-deficit India is ChatGPT’s second largest market by user. 

India is among the world’s fastest-growing AI markets. And this year global technology majors including Microsoft, Amazon and Google have announced plans to invest more than $60 billion in AI data centres, expanding businesses, imparting AI skills and so on. The investment boom promises jobs, cloud capacity and digital muscle for an economy hungry to move up the technology ladder. But can India power its AI future without overloading its electricity grid or draining scarce water resources? 

Data centres are nothing new for India. What is new is their scale and purpose. Traditional facilities supported email, banking systems, UPI payments, social media interactions, streaming platforms and so on. The new generation is built for AI — clusters of Graphic Processing Units or GPUs designed to train and run complex models. These GPU chips consume enormous amounts of power and generate heat. 

India’s installed data-centre power capacity is around 1.5 gigawatts (Gw), according to Bengaluru-based technology consultancy UnearthInsight. By 2030, that figure is expected to climb close to 9 Gw, enough to power a small city. 

Big Tech spends 

On December 9, Microsoft announced its largest investment in Asia—$17.5 billion over four years—to expand cloud and AI infrastructure in India. At the heart of the plan is a massive ‘hyperscale’ facility in Hyderabad, expected to go live in 2026. 

Google has committed $15 billion, among other things, on an AI data centre in Andhra Pradesh, its biggest bet in the country to date. Queries on India investments and environmental impact, sent to Amazon, Google and Microsoft went unanswered. Microsoft directed Business Standard to its previous announcements on sustainability commitments and using renewable energy, including preventing waste, and reusing and recovering materials at its data centres. Private-equity giants like Blackstone, KKR are also backing domestic data centre developers in anticipation of India’s explosion in data consumption spiking local computing power needs. 

Vinish Bawa, partner and leader, telecom, at consultancy PwC India, said, “A $50+ billion wave of data centre investment could add significant, round-the-clock power demand. While this improves grid utilisation, it will test transmission capacity and peak load management as well.” 

India’s power grid already faces stress during peak summer months. Adding multi-gigawatt, 24×7 loads could worsen shortages unless power generation and transmission keep pace. Anjal Prakash, clinical associate professor, Indian School of Business and a contributor to the UN’s Intergovernmental Panel on Climate Change (IPCC) report, said, “The data centre build-out could strain India’s grid which already faces peak shortages, potentially increasing blackouts.” 

Power demand might surge 10-15 per cent by 2030. “In the long term it risks higher emissions if it’s coal-based (currently half the power is coal-generated), but shifting to renewables like solar could cut CO2 emissions by half,” Prakash added. 

“If incremental demand is aligned with renewables,” said Bawa, “data centres could strengthen India’s clean energy transition, rather than materially worsen emissions.” 

Some experts believe India is well-positioned to support a higher grid load. Vibhuti Garg, director, South Asia, Institute for Energy Economics and Financial Analysts (IEEFA) said, “India has more than 200 Gw of installed renewable capacity and is adding over 30 Gw annually. Besides, solar and wind tariffs are among the lowest in the world here. Several states—Gujarat, Rajasthan, Tamil Nadu, Karnataka, Andhra Pradesh—have surplus renewable potential and are emerging as clean-energy hubs.” 

Garg added that technology firms have committed to net-zero targets at a global level. “It is essential to ensure that these commitments are upheld when such facilities are established in India.” Using a combination of technical improvements (low power chips) and renewables, global companies including Microsoft, Intel, Amazon, Meta and Google are committed to reaching net zero between 2030 and 2040. To be sure, that could be a challenge with rapid, AI-driven growth across industries. 

These data centres are much needed as enterprises and users lean on machines to automate tasks, predict trends, solve complex medical, climate change, agriculture problems and so on. 

Gaurav Vasu, CEO, UnearthInsight, said, “There has to be a well thought-out location and grid strategy—concentrate the most power-hungry AI clusters in zones with easy access to large renewable energy bases, like the Kutch region of Gujarat, and coastal belts of Maharashtra and Andhra Pradesh—rather than pushing everything into already stressed urban cores.” 

Water problems 

Apart from electricity there’s a water challenge as well. Prakash said hyperscale data centres can consume up to five million litres of water a day for cooling. According to the World Bank, India’s data centre water consumption is expected to more than double from about 150 billion litres in 2025 to nearly 360 billion liters by 2030. 

“The risk is greatest in data campuses located in water-stressed urban regions, especially those using evaporative cooling systems. Air-cooled or recycled water systems could mitigate the risks. Rain water harvesting and such alternative means to water conversation will also help,” said Prakash. 

Air cooling, liquid immersion cooling and closed-loop systems (recirculating water in a sealed system) reduce freshwater use. Treated wastewater and rainwater harvesting can further limit impact. Smaller, modular data centres—expected to account for around half of the new investment — are far less water-intensive than hyperscale campuses. 

As data centre investment gathers pace, experts also insist on environmental levies to mitigate risks. Prakash advocates clear benchmarks — such as Power Usage Effectiveness (PUE) below 1.3 and a path to 100 per cent renewable power by 2030—aligned with global best practices. 

PUE is a key metric for data centre efficiency, showing the ratio of total energy consumed by the facility to the energy used by IT equipment (servers, storage, network gear). A PUE of 1.0 is perfect (all power goes to IT). But real-world data centres always use more for cooling, lighting, etc, so lower PUEs (1.3 or less) signify better efficiency. 

Learning from best practices 

Garg of IEEFA said, “There’s a strong case for establishing India-specific clean energy obligations for data centres, including clean power sourcing or mandates for hybrid renewable energy storage contracts.” 

India could look at models from other countries as well. 

“Nordic countries (Sweden/Finland) power data centres almost entirely with renewables and reuse waste heat for district heating. Spain and Portugal leverage coastal locations, high renewable penetration, and water-efficient cooling systems,” said Vasu of UnearthInsight. 

In the US and Europe, hyperscalers including Google, Microsoft, Amazon Web Services (AWS) are experimenting with 24×7 carbon-free (CFE) matching and unprecedented transparency on energy use. CFE matching is a method for companies to decarbonise by matching every kilowatt-hour (kWh) of their electricity use with carbon-free sources (like wind, solar, nuclear) on the same local grid, every hour of every day, ensuring clean power is available when needed. 

One of the more intriguing ideas gaining traction is repurposing decommissioned power plants into AI-ready data centres. Vinit Mishra, partner, technology consulting, EY India said, “The rapid acceleration of AI has triggered an unprecedented demand for hyperscale, AI-ready data centres, while the power sector is simultaneously navigating aging infrastructure and evolving sustainability priorities. This convergence presents a unique opportunity: repurposing decommissioned power plants into next-generation AI-centric data centres.” 

This strategic alignment, Mishra believes, could not only accelerate India’s AI readiness but also unlock value, promote circular economy principles, and set the stage for a more resilient and future-ready power ecosystem. 

For utilities, this model offers new revenue streams from idle assets and improved grid stability through predictable, high-quality demand. For developers, it aligns with circular economy principles, strengthening environmental credentials. 

Physician heal thyself 

The irony of the AI boom is that the technology driving demand may also help manage its consequences. Smarter chips are already becoming more energy-efficient. AI bellwether Nvidia’s new GPUs are 2X to 3X more power efficient than the earlier generation chips. However, each new generation of AI model is trained on around three to four times the computing power of the previous one, which could negate the power efficiency of newer chips. 

If AI can help design chips that consume less power, cooling systems that use less water, and grids that balance demand intelligently, it may yet prove capable of solving the very problems it creates. 

So the moment you send your query to AI, it may then land not just in a football-field-sized data centre, but in a carefully engineered ecosystem designed to think sustainably. 

Focus on data 

Power and water demand

• Data centre power demand to double from 0.5-1% to 1-2% of total power demand by 2030
• Data centre water consumption to increase from 150 billion litres in 2025 to 358 billion litres by 2030

Potential solutions

• Locate centres at places where renewable power is easily available.
• Look at alternatives to water for cooling data centres
• Check out global best practices: Nordic countries run data centres on 100% renewable power and use waste heat to power homes
• Split demand between hyperscalers and mini-data centres
• Establish clean-energy obligations for data centres
• Repurpose decommissioned power plants into next-gen AI-centric data centres

Sources: World Bank, IEEFA, EY & UnearthInsight

 

The writer is a New Delhi-based independent journalist