AI is no longer just a technology—it's a force shaping geopolitical power. India’s current approach is fragmented and overly focused on applications. It must now think boldly and define its own AI trajectory.

The panelists:

• Ambassador Pankaj Saran, Former Deputy National Security Advisor and Convenor, NatStrat

• Dr. Sundeep Waslekar, President of Strategic Foresight Group 

• Ambassador Smita Purushottam, Founder, SITARA

• Srikanth Velamakanni, Co-founder and Group CEO, Fractal

• Anchored by Professor Rishikesha Krishnan, Director, IIM Bangalore

22 Takeaways from the session

Read Time: 8 min

How critical is AI?

1. AI is the new battleground of global dominance

“We are facing a binary configuration of these two technology powers... of the kind that we saw during the space and the nuclear era.”  — Saran

• Like the space race or nuclear era, AI is driving a new binary contest—between the US and China. The stakes include economic leadership, security, and technological sovereignty.

2. China is building a closed, sovereign AI ecosystem

"They're building their own chips. They're building their own models, creating a full ecosystem… they can leapfrog." — Velamakanni

• By controlling its data, chips, models, and compute infrastructure, China is creating an alternate AI universe—independent of Western tech. 

• India, in contrast, has left its digital sovereignty exposed.

3. India’s data is being mined—without strategic oversight

"We haven't built the compute hardware capabilities. We have allowed our complete data ecosystem to be pillaged by Big Tech...

“[We need] a full action plan on how to protect digital sovereignty and how to create a sovereign compute stack." — Purushottam

• Weak data protection laws and dependence on foreign cloud infrastructure have allowed global tech companies deep access to India’s digital ecosystem. This poses a national security threat, especially with AI weaponization on the horizon.

• China has firewalled itself from Big Tech and the global Internet. In contrast, India has allowed free access to its data. DigiLocker, for example, is on a foreign cloud service. A huge part of India’s logistics is on Google, which is now infiltrated by the CIA.

4. The AI race is already in a new phase

"We talk about the DeepSeek moment, but DeepSeek v2 is already an old story—it and ChatGPT are not even amongst the top 15 AI models….

“The big competition between the Americans and the Chinese is for new scientific discovery and controlling the entire science and technology over the next 50 to 100 years, and we are not even looking at them.” — Waslekar

• AutoML-Zero doesn't require much data; it works on the basis of calculations. 

• Models like AI co-scientist by Google and AlphaFold by Deepmind are on the path of making scientific discoveries. 

• DeepSeek R1 is capable of self-replication, scientific discovery, and even bypassing cybersecurity. 

• India is not in this game yet.

5. The world is moving towards managing “frontier risk” — the potential for cutting-edge “frontier models” that are under development, to pose a threat to public safety and security.

"The problem with Indian AI policy is, it's made by MeitY and Niti Aayog... I don't know how much the National Security Council is involved." — Waslekar

• UAE’s artificial intelligence company G42 has launched its Frontier AI Safety Framework. Therefore the US doesn't put restrictions on the UAE when exporting GPUs. 

• The AI Acts in Brazil and South Korea discuss how to manage the risk.

• G42’s Frontier AI Safety Framework is a comprehensive set of protocols designed to ensure the safe and responsible development, deployment, and management of advanced AI technologies.

• The G7’s Hiroshima Process does the same.

• India is getting left behind

  • The US puts restrictions on exporting GPUs to India because India doesn't discuss frontier risk.

  • India’s AI policy remains driven by development goals via MeitY and NITI Aayog, with little input from its security establishment.

How fast is AI technology advancing?

6. From knowledge to reasoning to action: The AI evolution

"In 2024, 2025, we had reasoning models like DeepSeek R1... Now in 2025 to 2026, what we're seeing is Agentic AI." — Velamakanni

We’ve moved from:

• 2023: Knowledge models (LLMs like GPT)

• 2024–25: Reasoning models (DeepSeek R1)

• 2025–26: Agentic AI—that can plan, act, and execute in the real world

7. Artificial General Intelligence (AGI) is not guaranteed—but even a 20% chance is too big to ignore

"Let's say there's a 20% chance that we'll have AGI in the next three years... they can really attack any installation, any security installation that we may have." — Velamakanni

• There are still many things to be solved before we get AGI. But once we have AGI, there's no stopping.

• A superintelligence advantage—by even hours—can allow one nation or actor to dominate global systems. The window to prepare is now. Delay could mean digital colonization.

What is the thinking reflected in the government and policy circles?

8. Cultural hesitation is slowing India’s strategic thinking

"Our public posture is focused almost entirely on the socioeconomic aspects and the applications. The national security dimensions are being underplayed…. 

"So far the AI policies are being driven by the operational line ministries. We are not hearing very much from the security people….

“But it doesn't mean that the security part of the government is not doing anything about it… It's not visible, but it is being done.”  — Saran

• India needs more public articulation of its national security and strategic interests vis-a-vis AI.

• India has historically been late in acknowledging the dual-use nature of emerging technologies. AI is following the same trajectory. 

  • India's nuclear programme began with a very heavy accent of civil use. India started talking about the military and other dimensions only much later. 

9. A national AI strategy needs central command

"There’s zero synergy and very little vision... We need a national AI policy under the PMO."
— Purushottam

• With no unifying AI doctrine, India is working in silos—different ministries, unaligned priorities. 

  • During Operation Sindoor, Pakistan got assistance in electronic warfare (EW) capacity and in detecting our systems. It has catalyzed greater awareness of the need for AI-enhanced capabilities. 

  • India’s tech sector is mainly transactional. It has been doing work for Big Tech. 

  • A few small startups are working on drones, EW, or for ISR (intelligence, surveillance and reconnaissance). But there's no central direction.

10. India has the data, talent, and money—but lacks will

"We have the resources, talent, data sets. We just need the will and the vision, and we can do it….

“For a $4 trillion economy, if you want to spend $40 billion, it's 0.1%. We can waste 0.1%. Even if nothing comes out of that, we are not wasting that money on AI.” — Velamakanni

• India holds one-quarter of the world’s digital exhaust and has global tech talent. What’s missing is strategic direction and commitment.

11. Focusing only on use cases is a strategic mistake

“Of course, India should be the use case capital [using AI for agriculture, education and health], but that completely misses the point. It's not about use cases. It's about geostrategic advantage.” — Velamakanni

• Prioritizing agriculture, education and health misses the real game—geostrategic power. India risks becoming a digital colony.

12. India needs to swing big—not play it safe

"Let’s say we have 10 overs to go, and we have to score 150 runs. Are you going to get singles, or are you going to try to hit some sixes?” — Velamakanni

• Even if the odds of winning are low, India must go for bold plays in AI capability-building.

13. Diplomacy has a role in influencing the rules of the game 

“We are refusing to even acknowledge [AGI] risk and to play that game. It’s exactly what happened with the nuclear non-proliferation treaty…. 

“Unless India takes the bull by horns and proposes terms to prevent the development of ultra autonomous models, to prevent the misuse of AI for developing chemical and biological weapons… we are going to be left out. I'm really ringing an alarm bell here. There is a lot of scope for diplomacy.” — Waslekar

• Like in nuclear non-proliferation, if India stays quiet on risks, rules will be made without it. Diplomatic engagement is urgent.

14. The Global South is ready for India to lead on AI governance

"The Global South countries are willing to rally with India... India must be very forceful in developing a global governance regime." — Waslekar

• If India proposes a fair, development-inclusive AI regime, the Global South will rally. But time is short.

15. Don’t wait for global treaties—build first, negotiate later

“How do you enforce an international treaty?… The so-called rules based order has almost collapsed….

“At the Paris AI Summit, guess, which two countries refused to sign the Paris declaration? The US and the UK…

“There are some countries who will not accept international discipline.” — Saran

• Global AI governance may never materialize. The history of technology governance has been more of failure than success. Take the Fissile Material Cut-off Treaty or the Comprehensive Test Ban Treaty—they were negotiated for years, but have not come into force. 

“Go fast forward, develop your own capabilities. The world respects only your capabilities... if you self-censor and keep yourself away from developing those capabilities, you will not be taken seriously.” — Saran

• Like with nuclear and space, India must build capabilities first if it wants a seat at the table.

16. What’s holding India back?

“We barely invest in R&D.…

“We got the $10 billion fund [India’s $10 billion incentive plan to attract semiconductor manufacturers], we went and gave it to a foreign company.” — Purushottam

• India invests far too little in R&D—only 0.64% of GDP—stifling innovation and local AI leadership.

• The Indian diaspora creates trillion-dollar value abroad. 

• India needs to create a supportive ecosystem for agile domestic companies to prosper.

17. It requires a mission at the India level  

“There are so many easy, low hanging fruits, why would you go and pick the higher hanging fruit, which requires risk taking R&D?...

“Why don't we bring 10 big companies in India, maybe $20-25 million from each of them and let’s say the government matches 4:1. We’ll have a couple of billion dollars… Advertise Indian AI and get the Indian diaspora to come and work for us.” — Velamakanni

• Industry prefers low-risk, low-investment paths to profits—that’s how capitalism works.

• Without government-led mission-mode thinking, bold R&D won’t happen.

• A public-private innovation fund could galvanize action.

18. Two parallel tracks: domestic industry comes together and Global South collaborates

“Certain kinds of AI models can be developed through the national cohesive effort… Certain other models which are going in the direction of advanced science, mathematics, physics — for that we will have to have cooperation with seven-eight emerging economies… The two are complementary.” — Waslekar

• Domestic AI initiatives can be complemented by strategic South-South cooperation.

• National efforts can focus on domain-specific AI (e.g., agriculture, weather, healthcare).

• With Federated Learning AI — a distributed machine learning technique that trains AI models across multiple servers holding local data without directly sharing the data — you automatically have data sovereignty.

19. Focus on building foundational models

“We have to solve for one problem: how do you build the best AI models that take us towards AGI.” — Velamakanni

20. Human capital bottleneck

“Our educational system is catering to humanities, to the elites… we are not looking at the Swiss or German model of vocational education. They do rigorous training for four years, and all the multimillionaires in Switzerland are from the vocational education background…

“You invest a few 100 million now, and you get trillion dollar companies at the end.” — Purushottam

• India’s education system doesn’t support deep-tech talent at scale.

• We need a Swiss-style vocational model and better incentives to attract top minds.

• Invite talent back by creating a special mission.

• It will require several strategic actions

  • The semiconductor story in Taiwan wasn't completely indigenously developed. A semiconductor plant was acquired abroad; efforts were made to bring in critical talent. 

  • Samsung got into the memory chips business by first acquiring an earlier generation plant in the US and bringing it back to Korea, building and learning.

21. Who will lead?

“It’s got to come from within. You can't import this kind of vision… You need the foot soldiers who will actually go out, roll up their sleeves and do it…

“You can have a creator, a catalyst—a government body like a Department of Space, a Department of Atomic Energy. If you think that was a useless model, your only option is to wait for market forces to throw up a leader…

“Today, for AI technology we are much better structurally and institutionally literate than we were in India’s nuclear or space programmes.” — Saran

“The moment we have an inspiring mission and a vision, a leader will emerge automatically. India has the raw talent…. The amazing talent we have around the world, they will show up. It's a big enough cause.” — Velamakanni

• India needs both visionary political leadership—who will give power, authority and resources—and committed executioners.

• Past successes in space and nuclear were pioneered by visionaries—Vikram Sarabhai and Homi Bhabha.  

• An inspiring national mission can mobilize India’s global talent pool.

• India has learned from past tech revolutions (space, nuclear). It is much better placed with better structures and is institutionally literate.

• Open experimentation is better today. 

22. Is AI overhyped?

“Somebody like Xi Jinping thinks that this is just the beginning of AI…

“In 10 years from now it won't be overhyped. We will, in fact, find that the game has totally changed.” — Waslekar

“If there was a 10% chance that an asteroid will hit Earth in the next 10 years, we would be planning today… There’s more than a 10% chance, probably a 50 to 60% chance, that we’ll have AGI in five years.” — Velamakanni

• India has to be prepared. The risks of missing out are too high. If it doesn’t do anything, it risks becoming a digital colony in 10 years from now.

• Even China sees this as just the beginning of the AI revolution.