Everywhere all at once makes India a safe AI bet

Everyone is looking for the next big AI bet. They’re searching for energy-rich places that can run data centers cheaply, for bottlenecks in the semiconductor supply chain that will earn massive profits, or for companies that might own the next breakout algorithm.

Usually, India doesn’t feature in these conversations. It isn’t going to be a chipmaking superpower any time soon. And, although a couple of big data-center projects have been announced, high energy costs and land scarcity limit its ambitions.

And yet India may be the biggest, safest bet in the age of artificial intelligence. Not because it will build the models, but because it will use them.

The large language models players already suspect this. In recent months, three companies have rolled out free access to their paid tiers exclusively in India. OpenAI Inc.’s lightweight ChatGPT Go plan will be available at no cost to Indians for a year; Alphabet Inc.’s Gemini Pro will be provided to every single one of Reliance Jio Infocomm Ltd.’s 505 million subscribers for 18 months; and Perplexity AI Inc. will offer its Pro version to Bharti Airtel Ltd.’s 350 million users.

That two of the three are going with telecom providers is partly to build scale. Nobody gives you numbers like India does. And young Indians are particularly ferocious adopters of technology.

The telcos, for their part, are always looking for products to bundle with their subscription plans. But some analysts have pointed out that it’s different this time: Instead of an entertainment package, they’re selling their AI add-ons as a utility.

We are on the cusp of a planetary-scale social experiment: What happens when you push free, unlimited, cutting-edge AI onto a billion-plus peoples’ phones?

Indian officials know what answer they’re hoping for. This might be how the country finally breaks out of the low-skill, low-productivity equilibrium it has been trapped in. Growth numbers look impressive. But they’re driven by a few high-output sectors; the vast majority of people work for themselves or for informal enterprises, and according to the International Labor Organization are only half as productive as the average.

Last month, the government think tank NITI Aayog argued in an AI-focused report that it could triple the productivity of India’s informal workers in the next decade, taking it from $5 an hour to $15 an hour. The officials said that their calculations showed widespread adoption would add between $500 billion and $600 billion to India’s output by 2035.

New Delhi is an upbeat town, and this figure is almost certainly exaggerated by the most flattering possible assumptions. The brutal fact is that every other attempt to give India’s hundreds of millions of young people the skills they need to compete has failed. And, as my colleague Andy Mukherjee has argued, entry-level white-collar jobs are as much at risk in India as they are anywhere else.

But official optimism about this technological transformation isn’t entirely unfounded. Youths here are not just enthusiastic about tech, they are unusually verbal users. The reason why almost every explainer on YouTube is made by and for Indians is because many of us search for answers on video sites first.

This sort of curiosity is almost designed for an era defined by language models that let you talk your way into competence. The one thing that we know about LLMs is that they seem to flatten the skill curve. Someone who has never coded a line of Python can suddenly create halfway-decent websites, and her friend without any experience of this country’s Byzantine regulatory environment can suddenly navigate opaque government forms.

Some of this is already visible online. Look at X, for example. ChatGPT’s characteristic syntax of LLMs is identifiable in tens of thousands of blue-tick accounts from India. And those posts are getting the engagement they crave, nevertheless. It’s irritating. It’s depressing, even. But it’s working.

A similar dynamic will play out in the real world. Indians will break down instructions they earlier couldn’t understand. They will teach themselves new systems step by step, transcending their flawed educational and skilling system. Speakers of Hindi or Marathi will see their horizons expand now that they’re able to navigate different languages in a multilingual society and world. They will be able to provide services that cross deep cultural divides.

What OpenAI and the others seem to have realized is that there’s more than one way for a country to provide AI infrastructure. Data centers, power plants, semiconductor fabs — sure. But the last and most essential ingredient is people. And people India has.

That’s why the biggest AI play in the world might be India itself — not as a chipmaker or algorithm owner, but as everything else. Not in any one sector, but in all of them. If language models really lower the barrier to achieving competence, if they really work as enablers for the unskilled and disconnected, then this country’s hundreds of millions of underperforming workers become the world’s most consequential growth story.

India’s government has failed to empower its people. It has failed to grant them skills. Perhaps it’s time to let the LLMs try.

(Disclaimer: The views expressed above are the author's own. They do not necessarily reflect the views of DH.)