Amid AI Demand, Data Centres Face Green Power Challenge

The infamous Three Mile Island in Pennsylvania is known as the site of the most significant nuclear meltdown in US history. While the disaster struck 45 years ago, the functioning second reactor was only shut down in 2019. Now, nuclear power is coming back to the island.

Microsoft has cut a deal with the plant to buy the facility's entire electricity generating capacity over the next 20 years. Amazon too signed three deals to set up small modular nuclear reactors.

So, what is fuelling Big Tech’s demand for power? The insatiable hunger for technology, but more specifically, the data centres that enable it.

The Cloud’s Earthly Cost

Contrary to the literal meaning, saving something on “the cloud” has nothing to do with the sky, but everything to do with the ground. Googling how to use ChatGPT, asking ChatGPT to draft emails, and even venting on social media about chatbots’ shortcomings, all have a tangible impact on the world — not just on people’s feelings.

Data centres, the backbone of 21st-century digitisation, enable everything from watching Netflix to training artificial intelligence (AI) models. The virtual world isn’t as virtual as we think.

And they have significant environmental consequences. Lined with racks of servers, these behemoth buildings consume vast amounts of energy and water, leaving a substantial ecological footprint to store our data and keep us online.

'Watt' Is The Future Of Indian Data Centres?

As the digital landscape evolves, recent trends indicate India is poised to become a significant player in the global data economy. The question now is how this growth is measured. Unlike real estate, a data centre isn’t measured in bigha, marla or square metres, but by its capacity for electricity consumption.

Last week, ICRA (Investment Information and Credit Rating Agency of India) estimated that India’s data centre capacity is set to double over the next 2.5 years, increasing from 950 MW (Megawatts) in FY24 to around 2,000 MW by FY27. 

One MW is a million watts. For context, Noida's peak power demand during the hot summer months is around 1,500 MW in a day. The capacity of Noida's 16 existing data centres are in the range 1.5-200 MW. To put it simply, powering India’s projected data centres would be akin to supplying electricity to an entire city.

Now, bring into the mix AI, machine learning, cryptocurrency and blockchain, which need more computing power in the form of graphics processing units (GPUs) housed in the data centres. 

These processors are complex, expensive, and also condensed. But the hot, new technology is also literally hot, and needs air conditioning or water-cooling. Although these processors are smaller in size, they consume more energy and water. 

A Call For Incentives At A Time For Regulation?

As per the ICRA report, an optimistic outlook for India’s data infrastructure was echoed by Reliance Jio Chairman Akash Ambani, during his address at the World Telecommunication Standardisation Assembly (WTSA) 2024, earlier this month.

He talked about the transformative potential of AI for India's future, stressing the urgent need to revise India’s draft Data Centre Policy, 2020. He also highlighted the importance of keeping Indian data within the country.

“Indian data should remain in India’s data centres. Therefore, Indian companies that are ready to set up AI and ML data centres should get all necessary incentives, including for power consumption," he said.

If the government indeed does subsidise power for data centres to capitalise on the AI boom, the case of Northern Virginia could act as a cautionary tale.

Success Story Or Cautionary Tale?

With about 300 data centres and 2,550 MW capacity, Northern Virginia on the east coast of the US has the largest concentration of data centres anywhere in the world. It is going through far-reaching economic, environmental and technological changes. And not all of them are good.

Apart from some financial gains to a few localities in the state, the long-term costs for the community and the environment is fast outweighing the benefits. The area is rife with transmission lines and big concrete structures encroaching upon residential areas.

It is a state of affairs largely brought about by Virginia’s subsidies for the infrastructure needs of the Amazons, Googles, Metas and Microsofts of the world.

With no comprehensive policy looking at the overall impact, Northern Virginia has data centres mushrooming across the state, pumping water, syphoning power and increasing energy demands.

To meet the state's data centre needs over the next 15 years, Dominion Energy Virginia, which provides power in the state, forecasts the production of power to almost double.

Footprints Or Scars?

The Secretariat spoke to David Gyulnazaryan, an independent consultant based in Nice, France, who specialises in decarbonisation and heat-reuse of data centres. “The use of servers has come into daily use,” he says, adding, “They only used to be set up in places like universities. Now, there is a large business demand for supercomputers.”

The time to think about the long-term impact of these centres is now, since they could derail efforts to meet the country’s decarbonisation climate goals.

The sustainability of data centres is measured using efficiency metrics. It’s all about the PUE, WUE and CUE. This isn’t a list of sounds while trying to solve a Wordle, but the data centre's power-, water- and carbon- usage effectiveness.

PUE, the most widely used, calculates the ratio of total power supplied to a data centre, against the power used by its IT equipment. Lower values indicate greater efficiency, since a smaller share is used to run secondary functions like cooling.

However, in the AI era, PUE as a measure falls short, explains Gyulnazaryan. “It only measures total energy consumption, not the actual useful output (read: computing power) through fewer servers,” he says, reiterating that these smaller servers actually consume more power.

Sustainability Push A Start, But Not Enough

In a bid to enhance the computing capacity for start-ups, researchers and public sector agencies, the Indian government has finalised a tender to procure over 10,000 GPUs, as part of its Rs 10,370 crore IndiaAI Mission.

A recent amendment by the Ministry of Electronics and Information Technology (MeitY) requires bidders to ensure that the PUE of their data centres does not exceed 1.35. As a result, the data centre industry is worried.

The Secretariat talked to multiple data centre operators, and their response to this sustainability push was tepid, if not cold. “It came as a surprise, and not a good one,” said an executive at a Noida-based data centre. “A PUE of 1.5-1.6 was considered good in a tropical climate like India's,” they said. 

With the projected expansion of data centres in India, it is unlikely that the target can be achieved without tremendous reliance on fossil fuel and coal-based energy. Although data centres are worried about meeting this standard, it is a step in the right direction. 

Meanwhile, with India also exploring nuclear as a decarbonisation pathway — small nuclear plants for private use may soon be managed by the NPCIL (Nuclear Power Corporation of India) — the need to reimagine power grids for sustainability could be closer to us than we think.

But, while electricity fuels our online world, it’s only part of the picture. 

(This is Part I of a two-part series on the environmental footprint of data centres in India. Part II will dive into the water requirements of data centres, and possible policy initiatives for a sustainable future)