India's challenge: Building sovereign GenAI

It needs more than just ambition; attracting Indian researchers back to the country is just one of the hurdles

India needs to build sovereign AI, including sovereign generative AI (GenAI) capabilities. As this columnist has argued earlier, artificial intelligence (AI) is a foundational technology, and depending on others for it could turn us into a techno-colony. For the fifth-largest economy in the world, aspiring to become the third-largest economy, not building sovereign capabilities in this critical technology would be a catastrophic mistake.

The Union government is fully cognizant of the need to build domestic AI capabilities. Prime Minister Narendra Modi has recently spoken about an AI mission. In last year’s Budget speech, Finance Minister Nirmala Sitharaman mentioned three centres of excellence for AI. The NITI Aayog too has brought out papers on what can be done in AI in India.

But plans, while a good start, need to be backed by a clear understanding of both the inherent advantages we have as a nation that can be used to build AI capabilities and, more importantly, the hurdles that need to be surmounted.

There are only a few advantages. The most important one is the amount of digital data we generate. AI, and GenAI, in particular, needs copious amounts of data to work properly. India is the second-largest generator of digital data in the world following China. It is the data we generate daily in every subject that can be harnessed to train GenAI models properly.

The other partial advantage is our IT industry’s capabilities in building applications, especially as fast followers (or copycats) instead of originators. While we have been largely absent in AI research and development, our growing capabilities in applications will certainly be of some help, though our IT industry is only now exploring creating AI applications.

On the other hand, the hurdles in our path to building sovereign AI capabilities are formidable. The first challenge is that we have been absent in AI research and are only starting now. The US and China are clear leaders in this field, though others such as Canada, the UK and some European Union members have also built some degree of expertise in specific areas of AI research. They have decades of AI research behind them.

To be able to build any significant AI research centre, we first need to think about how to attract experts in the field to come and lead the effort. We could possibly take a leaf from China’s book on AI research. The Chinese government, as well as some of its biggest technology firms, aggressively wooed Chinese-origin AI researchers in the US. They attracted Chinese researchers from Universities as well as from Silicon Valley’s Big Tech firms and startups. This, along with a focus on improving the quality of research in Chinese universities, provided them with the brain power to almost catch up with the US in AI.

Some of the best papers in AI in the US have been co-authored by Indian researchers who have moved to that country. The government as well as the private sector need to make efforts to attract some Indian-origin researchers to return. This would mean a change in policies on the government’s part and in attitudes within Indian private companies, which have not been known to pursue cutting-edge research so far.

The second hurdle is the lack of graphics processing unit (GPU) cloud infrastructure in the country. This may be less of a problem than initially thought. There are plenty of cloud service providers globally who are willing to service clients ready to pay. Also, Nvidia has recently announced tie-ups with both Reliance and the Tatas for AI infrastructure. Finally, the government itself plans to spend Rs 10,000 crore to set up its own GPU cloud infrastructure. Though GPU demand has shot through the roof, the infrastructure required for AI research should not be a huge problem.

The third hurdle is something that rarely comes up in discussions about AI, though it may be the most significant going forward. It is the enormous amounts of electricity (and water) that Gen AI models require, and this demand is going up every day.

It is projected that global AI research energy consumption will pretty soon overtake the annual consumption of countries such as the Netherlands and Sweden, and the demand for energy will keep going up as GenAI progresses. Sam Altman, CEO of OpenAI, recently admitted that energy demand could be a huge stumbling block for GenAI’s further march — and he hoped that a breakthrough in nuclear fusion would provide the huge amounts of clean energy that GenAI will need to progress further.

Mr Altman and others in the US have backed and funded startups working to make nuclear fusion practical. Their thinking is that despite the falling prices of solar energy and the rapid increase in installed capacities, it will not be enough to meet the insatiable energy demand of GenAI (and later, artificial general intelligence or AGI).

India’s energy demand projections may need reworking to include the GenAI factor. It could end up being the most challenging problem that the government faces in its energy transition journey.  

The writer is former editor of Business Today and Businessworld,  and founder of Prosaic View, an editorial consultancy