India’s AI bet: how to win the data centre race

The nature of global power is changing. All over the world, big blocks of ordinary buildings spread over acres of land are running every hour of every day. These data centres are built by big companies better known as hyperscalers, and these companies are pouring billions and billions of dollars every year into running this infrastructure because they have entered a race: the artificial intelligence (AI) race.

The ones at the top will be the ones who can run it longer. In other words, the ones with the most data centres. But alongside these companies, the nations that are hosting those data centres are also competing in the same race. These nations provide the resources, land, and labour needed to build data centres and support their infrastructure. However, in recent years, many of these companies are struggling to secure those resources from countries.

This is already generating growing resistance to data centres. New York Governor Kathy Hochul announced that New York will become the first state to impose a data centre moratorium, while communities elsewhere are moving towards permanent bans. In India, water crises in Haryana and Rajasthan show how seriously resource scarcity can affect communities. As data centres come to India, these could become India’s problems too.

But one country has caught the attention of all these hyperscalers: India. Over $200bn in investment has been promised to build India’s AI infrastructure. About $40bn has been invested in India since 2010, while Google is establishing a full-stack AI hub and investing in infrastructure in the country.

This is a pivotal moment of economic development for India. India skipped the manufacturing stage last time, and it massively slowed the country’s growth, especially compared with countries like China. Now, India has one more opportunity — probably its last for the next 20 to 30 years — to set itself up for the future. This is why the government is doing everything it can to host these companies in India.

But India will also pay a cost. One that turns this whole strategy into a big bet. A gamble India will have to play no matter what, and it must win. Because if it doesn’t, far more is on the line than what India will get in return.

The Visakhapatnam data centre project shows why this matters. One of the AI data centres being built in Visakhapatnam, Andhra Pradesh, is planned as a one-gigawatt facility spread across 480 acres, with an investment worth $15bn. But in Taluvada, one of the villages where the data centre will be built, local people are concerned. According to the Human Rights Foundation, in places like Visakhapatnam, where groundwater depletion has already created huge water stress, such a project will almost certainly intensify the city’s crisis.

It would be hypocritical to say that AI data centres in India are not crucial. AI is at a point where it just cannot be stopped. Every prompt, every question, every image people want AI to edit and every piece of code they want it to write needs AI data centres. That is why this infrastructure is so important to build, especially if India does not want to be left behind in the race.

But how should a data centre be planned in a city? How should resources be allocated? And how should India negotiate with these big companies to ensure a fair deal?

The fight is essentially over resources that locals must share with data centres. Traditional data centres typically had a total facility power capacity of about 10 to 30 megawatts. AI data centres now operate on a much larger scale. Powerful graphics processing units (GPUs) perform huge numbers of calculations, and because these chips work continuously, they generate massive amounts of heat. A single high-end GPU puts out 10 times more heat than a human body does. Tens of thousands of these chips can exist inside a centre, creating the need for enormous amounts of cooling and water.

It is not just water. It is electricity. In 2024, data centres in the US used about 183 terawatt-hours of electricity, roughly 4% of total US electricity use. If that electricity were instead used by households, it could power about 17m average US homes for one whole year. By nature, these facilities draw heavily on the grid, and the cost is shared among the locals who use it.

Companies continue to build in such places largely because of incentives. States offer tax benefits, land is abundant and cheap, and construction costs are lower. India is particularly attractive because it is much cheaper to set up data centres here than in the US or Europe. State governments also offer electricity-duty exemptions, stamp-duty exemptions and other subsidies. But the question remains: what is India getting in return?

India wants technological gains, domestic AI models, reduced dependency on foreign clouds, quality employment and local economic development. The plan is to become the data capital of the world, leverage the country’s advantages and build sovereignty. But what does sovereignty even mean? Microsoft, Google and Amazon have made sovereignty promises, including local data storage and customer-controlled encryption keys. However, their infrastructure is connected to global systems, operated by US-headquartered companies and subject to US law. Even if the data is physically located in India, technical dependence remains.

Neysa, an India-based AI infrastructure provider, promises sovereign GPU access and has raised $1.2 billion to deploy 20,000 GPUs in India. But these chips are bought from NVIDIA. If NVIDIA stopped selling to India, the stack would stop working. Deployment does not automatically create technological independence. Even with the 20,000 GPUs deployed, only 6% of them are currently used.

How can India make the right bet?

India is not negotiating from a position of weakness. These companies are not coming to India as a favour. They want access to one of the world’s largest digital markets, cheaper construction costs and a deep pool of technology workers. The conversation cannot end with India asking them to build in the country. It has to continue with India asking what they are willing to build around the data centre.

In exchange for tax holidays and electricity-duty exemptions, India can ask hyperscalers for more technology-intensive activities, research ecosystems, AI hubs and partnerships with academia. A data centre in Telangana, for example, could come with compute units for a university so researchers can train AI for applied sciences.

Building infrastructure in India is not necessarily a bad thing in itself. But if India is providing the resources, subsidies, and infrastructure, it should get more out of hyperscalers than tax breaks and data-centre plans.

And there is one final concern: how will this growth benefit local people?

If people are not included through jobs, training or related opportunities, AI growth could increase the gap in the Indian economy and leave a large section of the population behind.

India cannot stop the AI race. But it can decide how it participates. The real question is not whether India should build AI infrastructure. The question is: how does India win this bet?