India’s artificial intelligence (AI) journey must be rooted in its existing strengths, rather than attempting to compete across every layer of the stack at once, said Arjun Rao, general partner at Speciale Invest. In a conversation with Udisha Srivastav, on the sidelines of the India AI Impact Summit, Rao also spoke on sovereign AI, gaps in the country’s AI ecosystem, Nvidia’s India push, the RDI fund, family office capital, and exit strategies in deep-tech firms. Edited excerpts...
India hosted the first AI summit of this scale. What is its significance and what are the key takeaways?
Till now, we hadn’t necessarily showcased our ecosystem on the AI side in a consolidated way. The AI wave is now 3–4 years old. There has been a lot of activity, and people inside the ecosystem know that, but there hasn’t been a collective positioning of how good our ecosystem is, what all we are doing, and even where the areas of improvement are.
About takeaways, we need to focus on our areas of strength. But there are some sovereign areas of AI where we have to build, using our own data, our own language-related models, and making our own attempts at global modules.
Apart from sovereign AI, what are the pressing gaps India should work on immediately?
There will obviously be gaps because the AI ecosystem is vast. My first view is that we should play to our strengths. We have a very large developer ecosystem. If that ecosystem transitions well into AI development, and we ensure that it happens at good pace and quality, we can benefit a lot by building a variety of applications.
Where there is a gap is in large foundational models. Recent announcements, for example from Sarvam AI, show that the first step has been taken. That’s important because a year ago people thought maybe that ship had sailed.
The second area is semiconductors and infrastructure. If we think ahead, not just 12–24 months but 5–10 years, this cycle has just started. We need some level of self-sufficiency in data centres and AI chips. We don’t need to build every chip in India — that’s not realistic — but having some AI chips and infrastructure capability built domestically is important.
Do you think India AI founders are moving to the US for capital, talent, or markets?
It depends on what they are building and their ambition. If they need large amounts of capital or are targeting enterprise customers in North America from day one, it may make sense. The Western pool of capital is much deeper — maybe 100 or 1,000 times deeper. The VC industry there is 60 years old; ours is under 25 years old. But if a company is building India-specific applications, like in defence AI, robotics for Indian manufacturing, drone tech, sovereign chips, then there is no reason to move.
Relative to 5–10 years ago, fewer founders may be leaving. Earlier, it could have been eight out of 10. Today, maybe closer to five out of 10.
You have applied to the government’s RDI fund. What are your expectations?
We applied for a fund which is a Rs 1,400-crore deep-tech growth fund. The RDI fund has not committed to a fixed percentage. They will evaluate and may support some portion — maybe 20 per cent, 30 per cent or 50 per cent — we don’t know yet. They will act as limited partners (LPs), similar to any other investor.
The advantage of applying to that fund is that it’s a vote of confidence. If around 190 funds apply and a smaller number are selected, that helps when raising the rest (of capital) from private investors. Raising capital is never easy, for startups or funds, but government participation accelerates that process.
Tell us more about your Rs 1,400-crore growth fund.
With this fund, we will invest in Series A and beyond, writing $5–8 million cheques. These are companies that have built products and have early traction. We expect to back about 12–15 companies and deploy over 3–4 years.
Our earlier Rs 600-crore fund, which closed in August, has made four investments so far. That fund will back around 20 companies with around $1 million ticket sizes at pre-seed and seed.
There is a conversation that family offices are showing great interest in deep tech. Do you think this is the case?
We have had family office support from the beginning, for nine years. Family offices provide patient capital, and they can wait longer for returns. It’s because many have built wealth through manufacturing or product businesses, so they understand hardware, engineering and innovation. Also, sometimes deep-tech startups can even become customers or partners of their core businesses. In our case, over 90 per cent of our capital is domestic, largely from family offices and HNIs.
India ranks low on global IP indices. Is IP creation a concern?
Innovation is very important. IP is beneficial. But a patent by itself does not make a great business. If we do more research — which initiatives like the RDI fund can catalyse — IP development will naturally improve. Research and IP are connected. Can we do better as an ecosystem? Yes.
Given the long gestation period in deep tech, what do exits look like?
Deep tech in India is less than 10 years old. We have seen M&A (mergers and acquisitions) as the primary exit route. In our portfolio of about 35 companies across three funds, we have had six to eight M&A exits so far. IPOs are also beginning, and Ather, ideaForge and MapmyIndia are there. These are early examples of hardware and deep-tech IPOs. Going forward, exits will come through IPOs, domestic and global M&A, and secondary transactions.
 "Fewer AI founders leaving India now, says Speciale Invest's ArjunRao")
