Global AI race: CEEW founder on how India can scale data centres sustainably and its climate impact

We see, for instance, the role that AI can play in early warning systems and disaster response. It improves the prediction of floods, cyclones, forest fires and heatwaves. An AI-powered flood forecasting tool can provide information up to seven days in advance. So, it has measurable impacts in reducing the costs of livelihoods or lives impacted. An AI-generated system like climate radar can help translate alerts into actionable personalised recommendations.

Second, it can also help with better climate and weather forecasting because it can increase the forecasting accuracy of extreme weather by up to 15%, which is much better than traditional meteorological methods. Advanced aerosol forecasting with AI delivers an operational five-day forecast, which helps you know air quality warnings, health assessments, etc. Similarly, other forecasts can help with dealing with a decline in crop yields or identify better production sites for agricultural businesses, up to 10% of the usual cost.

The third area is to do better climate risk assessment at a hyper-local level, which helps to build the right infrastructure and build insurance mechanisms to protect people and their economic activities. For instance, at CEEW, we’ve been using AI and machine learning tools to replicate various models for heat risk assessments because we can then develop very granular heat action plans at a neighbourhood-by-neighbourhood level… on the mitigation, adaptation, and resilience side, we see clear use cases. In some ways, some of it is already being deployed but a lot more can happen.

Even with several use cases, challenges persist. How can India scale AI infra use in a sustainable and resource-efficient manner?

Different things need to happen. Since we did a paper on scaling India’s data centre ecosystem, we found that about 15-odd states have some kind of AI or data centre policy as part of their industrial policy, but only 5 of those 15 states have embedded sustainability features consciously in those policies.

We think that scaling up data centre infrastructure and leveraging AI as a technology for impact innovations must go hand in hand with sustainability goals, and therefore, at least three urgent things need to happen.

One is to enable a more firm clean power at scale… Without this, AI will continue to rely on fossil fuel-heavy grids. Second, we’ve got to institutionalise the performance transparency of this infrastructure… right now we don’t have a power use efficiency or a water use efficiency or a carbon intensity benchmark…

Third, we have to integrate AI into the grid and spatial planning. It shouldn’t be just a race to deploy thoughtlessly. AI should be deployed optimally, meaning which is the location where the land-water-energy nexus is most optimally used so that data centres also have sustainability as a long-term benchmark that they can pursue and how they can use grid assets etc…

Finally I would say that we certainly need a lot more incentives that reward this transparency and optimal performance. Will those incentives come from carbon markets or water scarcity markets? Will the subsidies that will be given or the tax holidays that will be given be greater, for instance, if you are a more optimally designed data centre — those types of signals to the industry will also nudge them towards better efficiency.

What are the risks at the city level as data centre expansion accelerates in urban hubs within Delhi-NCR and other metropolitan cities, given that we also don’t have a national policy framework yet on governing data centre development?

As per our latest (CEEW) report, we already see that about two-thirds of data centres are located in primarily six regions like Maharashtra, Karnataka, Tamil Nadu, West Bengal, Telangana and Delhi. Many of these are also water-stressed regions. They add to the power demand, of course.

The challenge here is the very rapid compounding of the problem that might occur within five years. For instance, today, we have about half a gigawatt demand. By 2030, this might be anywhere between 4.5 and 6.5, and some estimates even say up to 8 gigawatts.

The same applies to the water footprint… Water use today is about 0.02 per cent, which is 150 billion litres, but imagine if this doubles, one might think it’s okay, it’s only 0.04%, but that means someone else gets less, and then suddenly it becomes a sensitive issue.

So the issue here is intention… Shift from opportunistic siting to planned siting for infrastructure development. Shift from efficiency as a nice-to-have to a non-negotiable. Shift from just data centres as a resource gazer to a resource optimiser.

Effectively, we call it a double helix. There’s a decarbonisation imperative, and there’s a digitalisation imperative. If they run separate tracks, eventually those tracks will hit each other, but as perpendicular lines, and that would create a conflict. If you optimise the resources correctly, then both revolutions can feed off each other.

What does the Indian approach to AI resemble globally?

There are different paradigms here. One is a paradigm that the only race worth running is a compute power race, but I think the purpose of the Summit… is for a different race. To run that race is a compute power plus plus, which is — what those use cases are, how do we optimally use resources and then how do we democratise this so that the lesson creates a very different, inclusive architecture of what that future could look like.

I would argue that we should facilitate more open data, shared infrastructure…

Second, we should use more partnerships to co-develop AI solutions that cater to challenges in a localised way. I would even extend it to create a portal for jobs and skills if there is going to be a disruption.

So how do you match jobs and thousands of different types of jobs and skills to the geographies where they are; how can we have shared innovation which is contextually determined; and how do we then champion more inclusive AI governance. What would a global AI platform look like and what would multilateralism mean in a world increasingly shaped by AI?

I think these are larger meta questions. The users are here, and now we have a good point. The efficiency issues will emerge not as pain points but choke points in the next three to five years. We must be anticipating and doing, but the meta questions are both philosophical and real because they ask us what the race that we want to run is, and I would say that there is more than just a computational power race to run.