India’s AI Options Beyond Optics: Real But Untapped
India’s AI push risks focusing on data centres and optics while neglecting chips, power and funding. Yet, redirecting capital and talent could still build sovereign capability.
As the AI impact summit makes its way past a major ‘pretend-intelligence’ faux pas (a robotic dog of Chinese make passed off as the creation of a local private engineering college), missing WiFi and digital payment infrastructure at the convention centre, India occupies both the spotlight of AI achievement and the deep shadows of policy confusion.
Stock market indices have tanked beyond information technology stocks, which seem to be recovering from a panic-driven reaction to AI’s apparent ability to replace swathes of software services. There is opportunity in this crisis, which we will explore further down.
Tensions Spur Arms Shift
The US has amassed a massive strike force in the Middle East, and is threatening Iran that it would act, if the Islamic Republic did not abandon its plans to enrich uranium, to whatever level. Iran has, so far, stood its ground, saying that it is willing to negotiate but not be bullied. If hostilities do break out, it is unlikely to be restricted to bombing of Iranian military targets. Iran would probably attack targets in Israel as well as American bases in Qatar, Kuwait, Bahrain, Iraq, the UAE, and Saudi Arabia, and even the military deployment in Syria.
The Post-Traumatic Stress Disorder this would initiate is unlikely to remain at the level of the affected military combatants’ psyche. Oil prices could go up, there could be an increase in terror strikes in many different parts of the world, whatever happens to the Ayatollah regime in Iran.
Norway joins Poland and Estonia in choosing a South Korean supplier for deep penetration rockets, as European countries diversify their arms purchases away from the US. South Korean Hanwha Aerospace’s Chunmoo multiple rocket launcher system has been adjudged superior to America’s HIMARS and Israeli-German PULS systems, reports Nikkei. South Korea’s successful foray into the growing arms market shows the potential that awaits an economy that has mastered the art of precision engineering and manufacture of high-end electronics. The implication for India is obvious.
Data Dreams Need Coal
India’s AI Impact Summit was preceded by an open invitation by Prime Minister Modi for the world to store its data in India. There would appear to be some delusion in policymaking circles that an abundance of data centres means abundance of AI capability. While passive storage of data in solid state memory is not particularly power-intensive, data centres that process the data stored on memory to generate AI results are extremely power hungry.
Data centre capacity is measured in megawatt (MW) and gigawatt (GW), and not in terms of the number of server racks a centre can house or the floor area of a data centre, because the essential entity being measured is data processing capacity, and that is measured by the power spent on cooling the processors, which generate heat as they crunch numbers at high speed.
Housing data centres means generating dedicated generation capacity for tens or hundreds of GW of power. Yes, there are some developers who promise renewable power for their data centres. Renewable power sounds clean, but their intermittency has only dirty solutions.
The easiest solution is to turn to thermal — meaning, in the Indian context, coal-burning — power when the sun ceases to shine or the wind stops blowing. That would mean that, for the benefit of a handful of data centre companies to getting the landlord’s rent for storing someone else’s data, ordinary Indians across the land would breathe the polluted air that Delhiites come to expect as standard, come winter. Air pollution stunts young brains, decreases productivity and raises healthcare costs.
Data Storage Illusion
Another solution to intermittency is battery electric storage systems. BESSs are an invitation to walk into strategic dependency on China, which has the best battery technology and the bulk of the refining capacity for battery minerals.
Sure, India figures among a group of 50 countries that join the US in a critical minerals coalition that would create non-Chinese supply chains and a mineral reserve. If we wish to move on AI when this fairy tale reaches the happily ever after stage, then, of course, there is no problem. But if we are serious about building AI now, and supplying all Indians with sufficient, round-the-clock, quality electricity to run small industries in rural areas and cook at home, we must give up the mad idea of building data centres for the world’s data.
As the Economic Survey points out, the AI that matters for business is in the form of applications that run on existing models, big and small. India’s most competent home-grown AI, from Sarvam, runs on 2 billion parameters, while Gemini 3 and GPT 4 are models that have a trillion-plus parameters. Indian business can make do with models from western companies or open-source, open-weight models from China, for developing their business applications. However, the AI that India builds for national security purposes must be indigenously developed.
Having lots of other companies’ proprietary data and foreigners’ data protected by inviolable privacy guarantees stored in India does not help Indian development of AI. India does not need to house foreign data to develop AI. Indians are themselves the biggest generators of data. The point is to capture that and store it here, for which alone we need enormous data storage capacity.
Chip Capability First
Strategic AI can also run on small models. However, India needs the capacity to develop on its own a foundation model comparable to the best from the rest of the world. The raw talent required to build it, young people who can be trained in advanced branches of mathematics, abound in India. But the constraint that would bind AI development efforts would be, as pointed out, once again, by the Economic Survey, is the availability of advanced processors.
Strategic AI capability must be insulated against denial of technology embodied in chips, as well as against foreign models that play foul at critical moments. That means developing indigenous chips of the most advanced kind. Can India do it?
An India that continues to spend just 0.65% of GDP on R&D cannot. But an India that identifies every individual assembly, sub-assembly and component of the complex array of machines used to manufacture advanced chips, and funds five start-ups to develop their own versions of each piece of kit, would be able to. All the money need not come from the exchequer.
Fund Future Champions
Indian stocks are overpriced by global standards, an analyst told the Core Report recently. He advocated Indians being encouraged to deploy abroad a portion of the money they pour into domestic mutual fund SIPs. Investible savings are chasing too few decent companies, and inflating their prices. What if Indian investors were given the opportunity to invest in large pools of risk capital that fund start-ups trying to build strategic capacity in AI, chips and weapons systems?
In other words, what if venture funding were no longer the preserve of ultra-high-net-worth individuals, but could be drawn from ordinary savers eager to own a slice of future national champions and contribute to building national strategic capability? What if India’s retirement saving schemes, the National Pension Systema and the Employees’ Provident Fund, were to offer their subscribers the option to allocate, say, 5%, of their savings to such venture funds?
India has the potential to, in fact, lead the world in AI but that potential cannot be realized by merely holding summits. India must mobilise the local talent that has takers in global capability centres into developing the relevant tech in India for India, and must fund that talent. That calls for political will, going beyond rhetoric and posturing.
