Sajith Pai on India's Digital Divides

SHRUTI RAJAGOPALAN: Welcome to Ideas of India, where we examine the academic ideas that can propel India forward. My name is Shruti Rajagopalan, and I am a senior research fellow at the Mercatus Center at George Mason University. 

Today my guest is Sajith Pai, who is a Partner at Blume Ventures, an early stage Indian venture fund. He is also a well known writer, and the author of the annual Indus Valley Report. We spoke about the trends in the latest Indus Valley Report, the stratifications in Indian consumer markets, investor sentiment, policies and infrastructure to foster the AI boom in India, the AI start-up scene and much more.   

For a full transcript of this conversation, including helpful links of all the references mentioned, click the link in the show notes or visit mercatus.org/podcasts. 

Hi, Sajith. It’s such a pleasure. You’re the only guest we’ve had now for the third time, so thank you for making this part of your annual Indus Valley Report writing and postmortem. I feel like now I’m in this with you. This year, before I forget, you were joined again by your coauthors from last year, Anurag Pagaria and Nachammai Savithiri, who I believe goes by N. S., both of them at Blume and also Dhruv Trehan. I just want to make sure that this fantastic report that you guys put out every year is acknowledged with all the coauthors. Again, this year, it’s absolutely stellar, so congratulations.

SAJITH PAI: Thank you on behalf of the team. It’s a pleasure to be here—third time, is that right? Wow.

RAJAGOPALAN: Yes. One thing I love is, it’s become a major talking point now. When it first came out, it was literally like a drive link or something that you had put up online, if I remember correctly, on the Blume website. Now, from people I don’t expect, that is, not VCs, not startup founders, people randomly say, “Oh, that’s an India 1 problem,” instead of a first-world problem, or they’ll say, “That’s an India 2 trend.” It always makes me smile. You guys are officially part of the zeitgeist well beyond the startup culture. Do you get that sense as you do this year after year?

PAI: Beginning to get that now. This year, I think each year, it gets a little bigger. I think, this year, it somehow broke out of our Bangalore tech, Gurgaon tech, Indus Valley bubble, to break out a bit into the mainstream. This year was exceptionally, I would say, buzzy. The intensity of it hit us. Increasingly, in fact, I was talking to a VC from a large fund, and he was saying that, “Hey, people pitch me and say, ‘We are looking at India 1, India 2,’” and, “‘Do you invest in India 1?  Do you invest in India 2?.’” He says, “I don’t know what this means, but then I had to read up on it, and it’s you.”

I just wanted to clarify that I didn’t invent India 1, India 2, actually. All credit to Mr. Kishore Biyani, whose book It Happened in India had it. We’ve certainly taken that construct, put data around it, had this metaphor of India 1 as Mexico, et cetera. We’ve certainly done a lot of popularizing it and given it these mental frameworks. Certainly, I think this year, I felt like it had hit mainstream India for a bit, which wasn’t so much true last year.

RAJAGOPALAN: I think the reason is that a lot of the stuff you’re talking about is also like the macro reality we’re living in. It has a lot to do with what’s going on with India and the US, and the capital flows between the two countries, the way investment flows, how we think about India’s growth slump and bouncing back. You’ve just made it legible. As an economist, I’m almost envious. I’m like, “We didn’t manage to quite do this.” We’ve been writing these dense reports and very dense op-ed columns, which people don’t quite understand.

I think what you’ve done very nicely is created a very cool metaphor around it, and surrounded it with examples of things we use daily and trends we see daily. Then people, even without any knowledge about finance or economics, they’re able to grasp what it is you’re telling them. I still think it’s quite amazing, what you’ve done.

PAI: Oh, thank you so much.

2025 Trends

RAJAGOPALAN: This year’s report, it’s quite long. You’ve broken it down into seven parts. The first three focus on India’s growth and structural transformation and consumption patterns, capital formation patterns, things like that. The next four parts of the report are about general and specific trends in the startup world, and everything from seed funding, to IPOs, to AI and e-commerce and so on. 

One thing I observed when I finished reading your report was, I felt like there’s this winner-takes-all tendency going on in India. What I mean by that is, if I had to pick a few examples, the seed rounds are fewer, but larger in size.

Second is, for instance, the top of the pyramid, in terms of consumption for the India 1 and India 1A, the market has deepened, but not widened. The income tax collection trends also reflect the same reality. Is this just a function of when the growth story slows down a little bit and there’s a crunch, there’s going to be a winner-takes-all situation, and in this case, it’s going to be that very top of the cream in every market segment and startup segment that you can identify?

PAI: It’s interesting that you drew these two parallels. I certainly think over the last few years, I think through the last decade, toward the end of that, I was very hopeful about India 2 breaking out, thanks to Jio for instance, and UPI. COVID, I would say, created this K-shaped trend, and it was very harsh on India 2, and specifically people who have not connected to the digital economy as sharply.

Subsequently, given the challenges that we’ve had a little bit, and now accentuated by geopolitical trends, there is a clear flight to quality happening in capital flows. I don’t know if it’s specifically an India-only trend, because for instance, a lot of the US stock market growth is led by the —

RAJAGOPALAN: Magnificent 7.

PAI: —Magnificent 7. India, of course, has its own Magnificent 7, but it has its stock which are really driving the index upwards. This is certainly a flight to quality in the venture industry as well. There is, for instance, a clear tiering of founders as well, which is not as apparent to folks. You would not see, but to me, there’s a clear tiering of founders.

If you are a first-time founder, and we use this terminology called fluent and fresh. If you’re a first-time founder who has not raised capital before, or is not an elite operator from a large unicorn starting up, the way you’re going to be valued or the way you are going to be, I won’t say treated, you’ll be treated well, but you’ll start hearing a lot more no’s. There’ll be a lot more time that you’ll have to spend on getting $1 million and  it will be a lot of effort. Whereas, somebody else will raise like $5 million over three Zoom meetings, or something like that. Sometimes over WhatsApp messages.

For instance, there’s a hot AI startup which received 25 term sheets, apparently, 25 term sheets. It’s the math. He who has more, more will be given. That’s at play. I think specifically to consumption, and this one, I don’t need to belabor the point, historically, we have had a smaller minority, which we call India 1 and there are different terms, et cetera, for that. That, I would say, it’s got a little more accentuated of late, and there are specific reasons for that, and we’ll get into all that. I would say, certainly, I think the term I would use, it’s clear flight to quality in as far as Indus valley is concerned. Here, I would say, it’s a little bit more the K-shaped recovery becoming set in stone, so to say, to the extent that I feel like, at what point will India 2 make a comeback? We’ll have to see.

RAJAGOPALAN: I also noticed some of the other interesting trends, in that, in your earlier reports, and maybe I am misremembering them or this time, I just dug into certain nuances a little more. I think you are absolutely right that the Indian market has always tailored itself to the top of the pyramid, because that’s where most of the consumption expenditure comes from. They really set the trends. Now, there’s a further market segmentation, in that, oh, this group of people really cares about luxury.

Now, the hope isn’t that we’ll make this for India 1 and hope that it’ll peter down to India 2. It’s almost the opposite. It’s like, why don’t we find that smaller group within India 1 that really cares about luxury? Whether it’s Banarasi sarees, or travel, or something else, or even quick commerce. Convenience is enormous privilege in India, and then cater toward that without much of an expectation that this market will broaden. The expectation that is it is deep enough and they are rich enough.

This is also true. You have slides on the hotel effect and the Coldplay concerts and so on. That’s where I was going with this. It feels like a little bit different from the old trend of India being consumption driven by its richest folks.

PAI: What has happened is the past belief that there is a trickle-down effect that starts with India 1, and then you slowly tailor yourself. FMCG, for instance, did this really well when they created this playbook by which they would take, for example, there would be a premium product and they would sachetize it in order to make it palatable. There would always be premium variants. The premiumization was a trend, but there would always be the base variant, and the base variant, the way they made it palatable for consumption was to sachetize it. 

FMCG or CPG, as it’s called in the US, they had their own playbooks. The challenge for Indus Valley, I would say, has been that they’ve not been able to take a similar playbook other than in two, three, four categories. For instance, YouTube, WhatsApp, Flipkart, Paytm, and PhonePe, these are all startups, which straddle both India 1 and India 2 reasonably successfully.

RAJAGOPALAN: All network goods. We know why.. The answer is buried there.

PAI: Beyond this, it’s very hard for someone to do both. You could adapt the product significantly to make it palatable for India 2, and that means you fundamentally change the nature of the product itself. That means that the adjustments that you do, it’s not just sachetization where you take a smaller variant, et cetera. It’s not a pack size issue. It’s a fundamentally change the product formulation issue. The playbooks that historically FMCG and big Indian consumer brands kind of use successfully to trickle down into India 2, are not available for Indus Valley to necessarily adopt.

What Indus Valley has really done is to break it up into two parts, say that, “Hey, we’ll look at India 2,” and there are a few of us who look at India 2 or typically regional variations or regional flavors of India 2, and try and do something there. Or there are folks saying, “I’ll focus on India 1,” and instead of trickling down, they’re saying, “We will use India 1 as a springboard to see if we can launch it in the US or into the Western market.” That playbook is something that’s becoming more and more clear now, and we are beginning to see brands which are successfully doing that.

For instance, there’s a company called Skillmatics in the toys business, there’s PocketFM. They’re not necessarily selling only to the Indian diaspora. They’re selling to Americans, regular American Joe’s and Jane’s.

RAJAGOPALAN: They sell to an income group, basically. That income group largely resides in the Western world, and a small fraction of it resides in India.

PAI: Correct. This is what we call the India is the 51st state of the US.

RAJAGOPALAN: Yes. I love the tweet that you had put up. I love the tweets you’ve collected in the report. Sorry, that’s a random aside, but it’s so much fun to read this report.

PAI: I’ll give a secret away. The tweets help make it fun, right? There’s also an interesting growth hack built into it. If I put a tweet, and if someone has, I’m not necessarily saying every tweet is selected by looking at the follower account, but if someone has a reasonable follower account, and they’re likely to tweet about that page, and there’s a curiosity. That is another way in which we bake distribution into the product itself. I wouldn’t say there’s a huge nefarious intent to select only popular tweets, but both help. The tweets help make it palatable, it’s also a great distribution hack.

RAJAGOPALAN: It’s fun to read because folks like you and I, we live our life online. We see this as it comes and goes, and then suddenly, when you see it in a report in a particular context, you have that little aha moment go off in your head. 

Now, the trend that you described, there are two, three different ways of interpreting that trend. One way of interpreting that trend is just, hey, we’ve gotten better at segmenting our markets, and the fact that everyone has a smartphone, and the fact that we can target so precisely, whether it’s in advertising or whether it is in use or UI/UX, we’ve just gotten better at that, and this is a great step forward for the Indus Valley. That’s one way of interpreting it.

Another way of interpreting it is, founders and investors have given up on the India growth story in some sense. Now it’s a question of, hey, what do we understand, and which market can we really cater to and do it at service levels that are good enough that we dominate that particular market? Or it could be something that you and I have written about before, which is this elitism, that the people who are making these products really don’t know India 2, and they don’t know Bharat, and they haven’t stepped outside of a Coldplay concert.

They are now catering to what they know, and that’s just the problem of the tiny echo chamber, which I don’t think is the case. I think Indus Valley is pretty deep and quite diverse. Of the three, what do you think is driving this particular trend?

PAI: I do feel that there is a little bit of truth to all the three aspects, but I would say that I think it’s not that the founders feel that the India growth story doesn’t exist. I think they’ve gotten smarter at knowing that—because once you take venture money, venture money is built around a certain speed and cadence, though, which is that the money that’s given is typically given 18, 24 months at a time. The money that you get is to help you conduct experiments to hit a certain rate of growth in 18 months, at which some other VC will come. The next stage VC will come and take it out.

RAJAGOPALAN: That’s all exit.

PAI: Correct. It’s not like this VC exits, but this VC doesn’t lead that round. The next VC has to come and underwrite that round, and they may participate a little bit. Because of the 18-, 24-month cadence, that you have to grow very fast with a certain milestone, you work backwards to say, what do I need to do to demonstrably hit? Over time, I think there’s a clear realization that there is reasonable depth in certain markets, and this is certainly what you need to do. The playbooks are there. You start in Bangalore, sometimes you start in Gurgaon, then you at least hit one more city.

If you’ve done two cities, then it’s very clear that you will be able to get the next round. You have to demonstrate product market fit, growth, retention, some unit economics, et cetera. Those matter too, but the playbooks are getting clearer. I think what’s happened over the past few years is that the playbooks for how you demonstrate to each subsequent investor, those have got clearer. There’s a clear playbook and there’s clear awareness of that playbook. Similarly, there’s a SaaS playbook for how you expand to the US, et cetera.

I would say what’s happened is, it’s not that the Indian founders feel there is no growth. I think they fundamentally tend to believe that there is a market. Sometimes, for example, some of the founders are actually surprised by, for example, how UPI, autopay, or UPI enables microtransactions or subscriptions to happen at rapid pace. They are pleasantly surprised. Just that I think the playbooks have got clearer, and now they are performing according to those playbooks. Hence, fundamentally, they’re saying, “I need to get to a certain depth in a certain market, and I’m going to give up everything else and perform according to the playbook.”

That’s what is driving fundamentally. Yes, there are perhaps a small minority of folks who probably say that, “Hey, this is what I know. I’m going to create this avocado brand,” which is like organic vegan something, like a slipper brand. There will be these niches for sure, but I think increasingly, the people who are coming in, the startup founders, are very educated, they’ve read the YC posts, they’ve read articles, et cetera. I think it’s more around the playbooks getting very clear and very structured now, and performing to that.

RAJAGOPALAN: Also, they’re usually pitching on potential, which means to communicate to potential investors, they need to signal which playbook they are part of. That could also be part of this. The Indus Valley is so diverse and thriving, you still need a shorthand to communicate. This might be one of the things driving it. Maybe in the future, they will also create UI/UX models that can broaden the brand, or at least so, I hope. I hope these things don’t get limited to a market size that will cap off at 20 million.

PAI: I’m hoping that it’ll expand beyond 20 million. What happens is, it can expand, but the energy and the effort required to push it beyond, makes the business less capital-efficient. Then pricing becomes an issue. You have to change pricing significantly to make it more palatable to India 2, because you can’t sachetize experience or offering. Then what happens is they need to work backwards to figure out, how do I reduce my cost base?

One very interesting trend, and I remember speaking with, I think it’s a journalist at the Arc, who said something very interesting. He said, “We’re talking about interesting patterns that we’re seeing.” I think it was Madhav at the Arc who said that the new wave of startups, for instance, think of Rapido—Rapido is now actually the number two mobility player.

RAJAGOPALAN: Wow.

PAI: Close number two, after Uber. Historically, India mobility used to be a duopoly between Uber and Ola. Rapido has come in and become the number two. Rapido’s cost base is very different. It’s much lower than, for instance, Ola. I think it comes down to the following that Ola fundamentally said, “Hey, this works in the West, let me just try and replicate it.” They tried to build a lot of things, which are probably not required. Maybe they had more product managers; the way they thought was very different. Whereas today, for instance, there’s a lot of custom code you can build. There’s a way in which they’ve architected the company and the product to really serve what maybe more and more India 2 needs.

Rapido started out as a two-wheeler. They didn’t start with a car. They started with a two-wheeler, and now they’ve started cars too. The way in which they architected the company, is at a much lower cost base. I feel it’s much harder to do it if you start from India 1, because you can, but if you start from the notion that, hey, I need to serve the mass customer. For example, Meesho, they’ve become very large, and they’re fundamentally in India to brand cheap clothes. Rapido was another. That’s also leading to very interesting models.

Like Namma Yatri is coming and saying, “I’ll actually charge the driver and make it a fast product.” I’m hopeful that it’ll not stop at 20 million, that it’ll grow, and that we’ll start seeing founders rearchitecting models or rethinking how to deliver. Not all will do, but we’re going to see some brands which are not able to adapt and which throw money at it, and there’ll be less capital efficient to reach more and more people beyond the charmed 20 million, 30 million, 40 million set.

They’ll also be founders who are smart about it. Rapido is a great example who look at it from a very fundamental point of view. I’ve given examples.  Frnd, for instance, that they’re not trying to solve the dating problem so much as a conversation problem. I’m hopeful, but we’ll see.

RAJAGOPALAN: Actually, Rapido is a great example. When we had this conversation two years ago, our thinking was that Rapido will start like Ola, which is people building the Uber for India, DoorDash for India kind of thing. When we had this conversation last year, it was like there will be a premium and a freemium. There will be a regular and a sachet. Now it’s, I think, almost the reverse, which is, if you create this Namma Yatri for autos or for scooters, which is Rapido, there are rich people whose children need to take an auto to go to school, or who need some kind of pooling situation to go to school.

Now, eventually those parents, or those nannies will also figure out that this is something we can use. Eventually, I think, they will have product segments which offer an air-conditioned car version, as opposed to a nonair-conditioned regular cab version, and so on and so forth. That seems to be a very nice and organic way of growing, or at least connecting India 1 and India 2, which is we need to get the lessons from the ground up as opposed to from the top down. Which, of course, you can imagine how much that appeals to my Hayekean sensibility, but it just seems like this is not so much an India problem, as much as it’s just a more sophisticated and nuanced entrepreneurship story that’s playing out.

PAI: I agree with you. In fact, Rapido has actually done what you suggested. They’ve actually got cars now, AC cars. They’re not necessarily as clean, but then Uber is not as clean either.

RAJAGOPALAN: I’m a consumer. I am very much that Indian lady who will take an auto when it’s crowded and who needs a car in the evening, or when it’s too hot.

PAI: It’s happening.

RAJAGOPALAN: This part, I find super impressive. Some of this growth of things like Namma Yatri—even Namma Yatri was literally partly government-initiated—but a lot of this has happened because of the DPI infrastructure, right?

PAI: Yes.

Last Year vs. This Year

RAJAGOPALAN: In your past reports, you had set up the DPI point quite differently. You had set it up as this digital public infrastructure upon which startups are now building. This time you’ve set it up more as a, this is digital public welfare, and it is really driving the consumption end, and the borrowing end, small-ticket borrowing end, which will, of course, have some impact on how startups deal their business and target their customers and so on. This, I thought, was a marked difference from last few years’ reports. How did you arrive at that, or was this always simmering in the background?

PAI: Yes. It was simmering in the background. Last year’s focus was different. I think we wanted to talk a little bit more about the DPI tech, and how they were actually leading to innovative business models, of supporting these models. For instance, Namma Yatri just started, is built on the Beckn protocol, this is an open-source protocol, and UPI autopay was very interesting. The fact that, for instance, ONDC credit, I think, its pilots were just starting last year. Now it’s a little bit more established. The account aggregator model was gathering steam last year.

We wanted to talk about those aspects and showcase some of how they were supporting startup models. I think this year, we wanted to specifically go a little deeper, but we couldn’t invest energies into everything. One aspect that stood out is how much DPI led to this, we call it digital welfare state. I think that’s a term I like to use, that how significant direct transfers have become to keeping, I would say, a lot of it is definitely there is electoral strategy behind it. We won’t deny it, but it’s also, for instance, India doesn’t have social security. Now what we’re seeing with, Ladli, all these schemes that are there for women and pregnant women in Delhi and so on and so forth, there’s 2 trillion transfers going every year.

What we’re really seeing is the digital welfare aspect, which I think we wanted to just highlight. Because a lot of people who read the report, et cetera, India 1 people, they don’t even know that this is happening. They don’t give attention to this. We want to bring that out a little bit. I think the second was to really look at why certain technologies have yet to find their feet, why they haven’t taken off yet. There, I don’t think we are able to do the definitive research to go deeper. There were lots of space issues and et cetera, but it’s at least something we want to explore because not everything the government launches is going to work. End of the day, they try out very different things. Certainly, for instance, this ABHA, which is Ayushman Bharat Health, that is beginning to give impact. That’s something you look at next year.

Already when I talked to founders in the healthcare space, one very interesting trend is that there’s a startup we were talking to, and they were actually selling used medical equipment, which suddenly has been stopped now. They’re figuring out what next. One of the interesting things he said was, doctors, typically who would want value products, they know that they’re likely to get insurance, that it’s going to come. There’s a preference for more premium brands like Philips, over even a Samsung.

I found it very interesting how insurance, for instance, is going to drive that behavior. ABHA, very interesting and probably going to dig deeper into that a little bit over these months. ONDC credit is another interesting thing. That’s something I’m going to dig a little deeper into. I was excited, for some reason, with ONDC marketplace e-commerce thing, but ONDC logistics and ONDC credits seem intrinsically interesting. There are aspects I really want to go into and look deeper.

There’s also the fact that the government is one of the big consumers of AI now. You don’t get the credit. What is the big product, and at what point would the government launch query if that’s a scheme you can take advantage of, and you can use a voice bot or an agent to get that thing delivered to you. At what point does it happen? Very excited by all of this, but it’s something where we’ve not been able to give it as much time, which is why I think it reflects all those compromises.

RAJAGOPALAN: I didn’t think of it as a compromise. Maybe I already read this in. When I look at something like NSS data, for instance, because I see that end of the spectrum also. I see how much families get stressed because of one health emergency. Most families in India borrow from friends and family, and then they have these small ticket loans that go in. That’s just one health emergency. It could be something fairly minor.

I’m not saying this is kidney dialysis or chemotherapy, which will go on for months and years. This could be like one simple surgery. That’s it. It fits in very nicely with the consumption story you’ve pointed out, which is that actually India is just almost entirely consumption driven. The most troubling trend in India 2 is how much of that consumption is also fueled by borrowing, in a way that their investments are not fueled by borrowing, or their human capital or physical capital investment.

You’re absolutely right, but this welfare piece is part of it somewhere. Would it be different, as most economists suggest, if you just did a direct cash transfer, roll up everything into one instead of targeting individual vote banks, give a direct cash transfer and you’re done, or should all these central schemes exist in the first place, or something else? I’m about as libertarian as they come, but no one, including me, suggests that a billion people should starve, and live without rations. There is some element of the welfare state which needs to be involved in making sure that the gains from the marketplace are spread out a little bit more evenly. Now I’m already looking forward to next year’s report since you mentioned all this.

PAI: I certainly feel, it’s not like other people haven’t spoken about it, the more direct cash transfers you give and where customer decides where to spend it—you could do that, with farmers, for instance, instead of giving subsidies, which have other challenges. Certainly, I think that that could be a better approach. It’s complex now. The states which have certain schemes linked to certain things, et cetera. I haven’t done the exact calculation, but I think there are in kind transfers and there are cash transfers as well. Certainly, I don’t have the exact data .

RAJAGOPALAN: No one does. No one has a list of all the schemes. We’ve tried looking.

PAI: There’s an interesting startup called Haqdarshak, which does this and which tries to list all the schemes that are possible. At some point, I think that will become like an AI bot. 

RAJAGOPALAN: That would be awesome.

PAI: Try to list all the schemes that if you are, for instance, someone who belongs to this caste and this taluka, is a farmer below do bigha zameen [two bighas of land] , what do you get. You’re right.

The Role of Artificial Intelligence

RAJAGOPALAN: I want to get into the AI part of the story. That’s been one huge development this year. Last year when things were picking up and OpenAI and Anthropic and all the big American firms were releasing their models, the wisdom—and I really mean wisdom coming from Bangalore, this is everyone from Nandan downwards—was we need to layer applications on top of these, and there is no need for us to join this crazy rat race and spend $100 billion in building out foundational models. India has access to English, as well as India is open, unlike the Chinese firewall and all these other problems. India is very well poised to take advantage of this.

Now, the DeepSeek story just completely demolished this idea. India’s tech people have gone into tailspin now trying to pitch this. The government has gone into tailspin trying to invite people to build AI models, as if you can invite someone to build this out. This is not Lego blocks. One, how do you perceive what has happened? Then we can get into how you view AI both in foundational models, as well as eating up SaaS or all the other downstream consequences of AI becoming larger and larger share of this space?

PAI: I would say there is pre-DeepSeek and post-DeepSeek, certainly as you said. A lot of the shift in thinking has certainly been led by that. There’s also a term around AI sovereignty, which has come together. There’s a presentation by this gentleman called Paras Chopra, who has the oldest company, and he’s trying to build a foundation model. Very smart young man. He had the presentation, which I saw, and which talked about how each power, and India as a power, should have its own model. It’s like having nuclear weapons.

RAJAGOPALAN: Exactly.

PAI: That’s the analogy you could use with this. I would say it is interesting. I cannot get it from today’s world, where suddenly this world where geopolitics matters suddenly. For a long time, it didn’t seem to matter. Globalization is dead? I don’t know. In this world, I think the government wants its model. Some of it is to do with, I want to use my words very carefully, but there’s certainly an element of tech nationalism, tech chauvinism—I don’t know if chauvinism is the right word.

RAJAGOPALAN: Tech sovereignty.

PAI: Tech sovereignty, yes. That, we should have our models. Why should we not have our models? We should send people into space. Lant Pritchett used to say and then post, for instance, that the difference between India and Indonesia was that Indonesia focused on schools where India focused on colleges. He said it, but certainly, there was this article that came out, which looked at India versus China as well. I would say, I think, for a country which has got $2,700 per capita income, how much should you focus on some of these? Is it really as big an issue as it’s meant to be? Does AI Summit really matter?

I think we’re past that argument now. Now people want that. The government’s getting 18,000 GPUs organized which, I don’t know if Brazil has organized that, or equivalent countries. I know Indonesia has organized it, but we’ve organized it. We are calling for almost like a tender for people to create. I think there’s just too much activation energy now that’s happened. It’s like a snowball rolling downhill. You’ll get a model. Will it be an incredibly powerful model like DeepSeek? I don’t know.

RAJAGOPALAN: I can see it won’t be.

PAI: But we’ll get something on which we can put the label that, “This is our model.” I think there may be more than one model, two models, and it’ll be an English model most likely, because it’s so hard to get those number of parameters and data collected for an Indian language. If you want to do something like that in Gujarati or Odiya, or something, you need to scan so many documents. There just isn’t enough data. For instance, 2018, ’19, onwards, OpenAI has been storing up data. That’s what good scale AI growth, Turing’s growth, a lot of Indian small companies benefiting from that wave.

To really create a powerful vernacular model and store up a lot of data, which doesn’t exist, I feel, what you’re going to do is you’re going to create an English model. Yes, it’ll be your model. We probably should be proud of it, but I would say my, as well as that, I do believe Nandan was fundamentally right. It wasn’t wrong. I think there’s a lot of truth to what he said, that we really need to focus on the application. On the application side, we haven’t seen as much. There are a few examples, of course, like I mentioned NVidia, for instance, with this 60, 70 million ARR sales globally. There’s another smaller example of presentations AI, but where are they?

For instance, why isn’t an option for people to give feedback to their MLA? Why isn’t there a way in which the government can call people up? There are so many interesting ideas. I think those need to get focus as well. Fundamentally, I’m actually a little bit more in the Nandan camp, really. That said, look, end of the day, I’m also excited by shiny new things. Part of the reason why our I venture is to see cool new toys, and all of these things coming out. That excites me a little bit as well. I do feel that we will get something, but what we’ll get, you made your point, but yes. 

RAJAGOPALAN: The way I see this foundational model question is not that India can’t build it. In India, we have this, I call it the upside-down Indian state, where the Indian state refuses to build public goods, and wants to build all this private infrastructure. We won’t build to pick up your garbage and get your roads in order, and things like that, but we are going to give you LPG gas and fertilizer subsidy, and now it gets worse.

Now, to me, this AI model is the most Marie Antoinette version of that. I think the Modi government already has this fantastic digitization project, which is enormously beneficial, not just to the state, but it is beneficial to the private sector. It is beneficial to citizens. Why not, instead of spending on these GPUs, throw that money into digitization, make it go into hyperdrive, knowing full well that if the data are there, then it’s going to effectively subsidize the cost for some entrepreneur who can get everything else figured out. Private capital can pay for the GPUs. It’s doing it the world over. Nowhere else in the world does the government have to buy GPUs. For DeepSeek, also the Chinese government didn’t buy the GPUs.

PAI: Absolutely.

RAJAGOPALAN: Maybe they help them steal the chips, is about as far as the conjecture goes on the Chinese government involvement. Looking the other way and maybe they’re understating the extent of the chips they used, but I don’t see any other government doing this crazy Marie Antoinette thing. You are absolutely right. Even our PLI schemes, we don’t target our PLI schemes toward apparel and leather shoes and shoelaces, and things like that. They’re targeted toward semiconductor chips.

It’s like you are solving this very specific problem where in a bizarre way, you almost have the least comparative advantage, but it’s because your minister or whoever wants to look cool at Davos. I don’t even understand the incentives at play. Maybe it’s like the piece that you wrote about Indians living in the anglosphere, living in India, but their mind is abroad in the first one. It almost seems like this elitist weird thing. That’s where I am.

Not to say that I’m pessimistic, but the second part is the government has also done this for ISRO. What we always miss is for a Mars mission to happen today, you have to have ISRO for 50 years. You have to have incubation, and you have to have this intergenerational incubation that takes place where you have a deep bench of scientists. Where do we have that for AI? Where is that Stanford, Berkeley, Caltech, Oxford, Cambridge merged in with the private sector and the data, that you then finally get something like this going?

These are the sorts of things that make me, snark aside, extremely confused about what is it that we are approaching from a policy perspective, and why would we ever approach this problem in this specific way. There are lots of ways. Another reason AI models don’t take off in India is we punish scale. You can’t build a large captive power plant. Electricity tariff pricing is all over the place. These are the two most important things you need for a data center, or to have your GPUs up and running. Our tariffs have crazy amount of policy uncertainty. What chips you import and where you import them from, will the GPU tariffs change over a period? Nothing we know.

The government has a massive role to play, but calling for tenders as if they’re building a highway, I’m just baffled by this, Sajith. I’m sorry. I know that you don’t have to comment on this. This can be my rant alone, but the whole thing is just baffling to me.

PAI: Indeed it is. The article that you wrote, the paper that you wrote with Alex, which is premature imitation, I like to use that lens because India has different interest groups. There is certainly an interest group, let’s call it for instance, bureaucrats or the government, the government bureaucracy. There’s certainly a desire to, you mentioned about standing out in Davos, et cetera.

There is certainly an India 1 aspect, and certainly, I think it’s incredible. The fact that the $2,700 per capita income, we’ve been able to do a Chandrayaan, it’s incredible. I find it’s utterly fascinating that the fact that for instance, at some point we decided that, hey, it’s fine to get a nuclear weapon. Look, I respect decisions that led to that, et cetera, I’m not disputing, but you took all of these decisions and fundamentally I think you’ve not been able to get a lot of people into the $4,000, $5,000, $6,000 per capita income. I find that utterly fascinating.

Each interest group comes and says, “Hey, look, I want to kind of do my version of premature imitation.” That is what is really happening, and where you’re coming and saying, we are as good as Silicon Valley. We are cool. Let’s build out the thing. Yes, sure, we have very good engineers, but I think the challenge is that we don’t have the density of engineers.

RAJAGOPALAN: Or the ecosystem. You need research labs, and you need people who are just mathematicians, and just biologists. Again, I’m not dissing Indian talent at all. I think that talent is there in spades. I think it’s just the infrastructure around it. DRDO had that infrastructure built around for decades on end, and then you get a very Indianized version of a nuclear test and so on. ISRO. Again Chandrayaan, I would not compare to this AI thing, to be honest, because it was done at a fraction of the cost. That’s not the direction in which the current GPU AI foundational model is going.

No one is saying we need to build this the Indian way at very low cost. The thing is, of course, we can spend the money. We’re a superpower. It’s almost the opposite of the ISRO Chandrayaan thing that folks like you and I admire, that we can punch well above our weight. Because Indian scientists in an Indian infrastructure or ecosystem are able to solve the problem very differently from how someone else would do it in Sweden or Norway. It’s the opposite of premature imitation. It’s really ground up.

Artificial Intelligence Startup Scene

Now that the foundational thing is out of the way, and sorry for ranting, how do you think about the AI startup scene? One view is any new cool AI startup that was coming around with all these wrappers and they were getting eaten up by ChatGPT the next week. Someone would introduce something, and they would just get out-competed. On the other hand, there’s the feeling that, hey, Perplexity is the outlier. Look at how it’s managed to challenge both ChatGPT search and Google, and things like that.

One is where are you on that debate of segmentation and very targeted, specific AI startups? Or is there something about the nature of AI and compute, which just makes that very difficult and makes this very different from SaaS-style product differentiation, and very niche startups for niche consumer groups?

PAI: Fundamentally, I think all startups, whether it’s SaaS, AI, et cetera, are trying to solve consumer problems. Ideally, consumer problems that are faced by large numbers of people, because that’s how you build a very large startup. I would say that increasingly in a world where there will, of course, be foundational models fighting amongst each other but I think beyond a point unless you’re going to use it for, I don’t know, what, protein folding, calculations or something, I think maybe there’ll be a model which is—like Tyler was saying that Germany is well suited for legal stuff. I haven’t read and seen podcast yet, but it’s reserved for next week. I saw that in his interview with David Perell.

For instance, there’ll be something for doing mathematical equations or there’ll be some model that’s better suited for that, et cetera. Other than that, I think the broad layer, I think all of them are the same. I think the joke used to be that, hey, you’re a wrapper company, you’ll get eaten up. I think you can really create interesting differentiation. I would say, Perplexity is a great example. Granola, for instance, is less known.

RAJAGOPALAN: Oh, yes.

PAI: That’s an interesting example. The founder is a very sharp guy. There’s an article he wrote as well as his podcast. Very interesting. I would say ultimately, AI at one level is going to be like electricity. In the 1920s, you had General Electric, so you’ll have OpenAI. In a few years, it will be down to who takes advantage of electricity better, and who takes advantage of AI for solving a real issue.

Someone will say, “I use electricity to manufacture something like maybe Pepsi and my factory is run on electricity.” I like that metaphor, and increasingly, I find myself trying to read a little bit about how electricity evolved, the early days of electricity. You’re going to see parallels there. I would say that I’m a little disappointed in India that why do we only have one individual, for instance, which is a startup that helps you create videos on prompts?

Of course, it’s based on OpenAI and it’s one of OpenAI’s biggest customers, et cetera. Why aren’t more Indian startups using that? That’s a pity. There could be a startup that is funded by some of my peers, which I don’t know about and which is going to launch and become big. When I speak to some of my peers, they do say that they’re not necessarily finding as much in India. A lot of them are investing abroad in the US.

The other wave that’s happened is Indian VCs are setting up teams in the US to fund, and they are actually funding Indian founders, sometimes non-Indian founders, who are building in that space in the US. We ourselves came close to doing one such thing. I find that fascinating.

RAJAGOPALAN: What is the reason for that? Is the constraint talent? Is the constraint capital? Is it the ecosystem or the customer?

PAI: Talent.

RAJAGOPALAN: Yes, that’s the constraint.

PAI: It’s fundamentally talent. I think density matters. I remember this fascinating Charles Correa article a long time back. Strange to bring Charles Correa into this, but it stayed in my head, of cities. It’s about cities, and when weird people meet and suddenly a different weird person emerges, and it’s bad to bring it in, you can’t explain it. I remember seeing those images of difference between villages and cities

Cities are more spiky people and when more and more of those spiky people meet, unusual geniuses emerge. What’s happening, I think, is we just don’t have enough talent density, even in Bangalore. Maybe it’s there. Sorry, but I don’t think there’s enough talent density. There was this fascinating chart. We actually said that 7% of top AI researchers are Indian. Not my comment, but none work in India. I feel like that is one key factor. Because once you build, if it’s cool enough, customers will get created, customers will figure out. 

All my dollars are going to non-Indian companies. All my AI dollars, whatever I spend, like $50, $60 a month on all your different tools, are going to non-Indian companies. What is stopping us I think it’s fundamentally talent density. You need a certain amount of talent density in one place for a lot of ideas to be ripped and discussed, and then suddenly something emerges. We lack that.

RAJAGOPALAN: It’s funny you say this. I’ve been an economist for a long time, but the older I get—now, what you described is exactly what we call the endogenous growth theory in economics. Paul Romer came up with it, and he won the Nobel Prize for it and so on. That’s exactly this cluster effect that you’re talking about, where you have this talent density and this density of ideas. One thing I realized over and over again is just how difficult it is to create these clusters.

Some of these cities have been there for millennia, and that’s where people end up going. The big hope in India was following China, that as India gets richer and urbanizes more, we create these huge pockets of talent density. China somehow managed to do it. Maybe, as you rightly point out in your report, it’s because they focused on primary education. They spent so much on R&D. They launched 100 universities for every one university India launched through the public sector and so on.

That has been one enormous disappointment for me, that somehow we don’t understand how growth really takes off. To the extent that both human capital matched with the right physical capital, all of it being in a dense place and all of it being relatively in a free and liberal environment without the threat of uncertainty and expropriation, that well-established truth, one, it keeps coming back when I read the paper over and over again.

Again, semiconductors also happened in Silicon Valley. That’s how it gets its name. The internet boom also happened there. Now, the AI boom is also happening there. Isn’t that really weird and bizarre?

PAI: Yes, that said. I want to say that in the case of China, the path dependence, or I don’t know what you call it, the second-order effect, really, not path dependence, but the second-order effect of Xi Jinping’s crackdown on finance and tech was what led to DeepSeek, because suddenly that company was told, “You’re making too much money from the wrong things,” and they said, “Okay, we’ll pivot.” Because it was a tech company trying to eke out nanoseconds of time to get a trading advantage.

They were very good at managing GPU and eking out some advantage of GPU, which has actually been DeepSeek’s strength because they focused on how to connect. Anyway, I think we need to say that in a way, China got a bit lucky. I don’t think Xi Jinping, or maybe this is what Xi Jinping anticipated. If so, it’s brilliant, right?

RAJAGOPALAN: No. This was just someone else did it, and now the government will take credit for it, I think.

PAI: Absolutely.

RAJAGOPALAN: I appreciate that. I love that kind of political opportunism. 

PAI: We’ve done a little bit of reading. Some of that didn’t come through, but my colleague Nachammai who worked on the report, she had actually read a couple of Substacks. There was this, I forget the name of the body, et cetera, but it was an Australian body which actually looked at 50 or 60 different subtechnologies in AI, and they found how papers from China were consistently—and people would just say, “Oh, they’ve just created such a paper mill.” They just publish papers.

But they were also getting better and better in quality. I think what happened is I think China systematically improved over the last few years. Now, some of it has to do with the fact that the Indian state, we encourage immigration, and it’s very easy for a very elite Indian math, computer science guy to migrate to the US, et cetera. We are okay with that. China, for example, it’s a little harder, maybe. There’s a lot more numbers, so there’s a lot more density there.

Some of them go, but some of them stay back. I think these big numbers and these things just worked out much better in China’s favor, and they haven’t worked out in India. Because what happens is if you’re going to send the top 100 folks from, let’s say, top 100 folks who crack JEE and let’s say, probably many of them will do very well in the Maths Olympiad or the Physics Olympiad, et cetera, top 100, top 200, typically—and this is what research says—60%, 70%, 80% of them today just land up in the US.

RAJAGOPALAN: I think Deedy did a whole thread on this, right? Yes, I remember reading that.

PAI: He did do that. That’s right. Deedy is the guy who did that, Deedy. Das. I think that is really one aspect that at some point if India wants to retain, like for example, try and create an institute, like IAS which is the Institute of Advanced Study equivalent like TIFR. TIFR should just say, if you’re top 100, don’t go, come do what you want here and try and create what you call a cluster effect where they all gather together. Something special could come, but it’ll take time. It won’t happen like tomorrow, but it’ll happen in a few years.

RAJAGOPALAN: No, I’m also optimistic about that sort of thing. I’ll tell you what I’m pessimistic about. For the China density to happen across the country and also in individual sectors, we have to fundamentally change the Indian mindset and get out of misallocation of capital. Most of our engineers are engineers because that’s a ticket to a better life, not because they want to do engineering. Everyone wants a government job because that’s a ticket to a stable life, not because they actually want to work in government or be a PWD engineer or anything like that.

We end up in civil engineering because you got a lower rank and we end up in computer engineering because you’ve got a higher rank, and not because you are actually interested in that. Instead of the price system or some other system, we are using rankings as the allocating mechanism, which ends up fundamentally misallocating talent. Once you are in chemical engineering, then four years, please hang out in chemical engineering. That’s it.

We get people who are very smart, we’re able to signal they’re very smart because we took in the right lot. They haven’t learned that much over the four years. Of course, that depends on who they are. At the end of the day, they’re utterly useless for being an actual chemical engineer. That I think is a social problem which we need to solve at some fundamental level where the parents stop telling kids in fifth standard that they need to become an engineer or a doctor or crack JEE. If I meet one more doctor who has become an actor—this is really just such a ghastly waste of social resources to do this.

Every single one of those TVF—they make these amazing sketches, and now Netflix shows and all right, they’re all engineers. Why in God’s name do you need to be an engineer to write a sketch? This is the sort of misallocation that just happens way beyond TIFR and IISc and IIT. I don’t want to just blame them. I think that human capital problem, which you’ve pointed out also in the report, I’m very pessimistic about this, Sajith.

In fact, I’m more optimistic about the startup situation because I think it gives a lot of people exit. People drop out of college or they have startups alongside college and actually start allocating their human capital in the place that really matters. I feel like the Indus Valley has given one potential exit route, which is very respectable, can make money, and allocate human capital well. But we need 7,000 more for the size of the country, which is not very comforting.

PAI: Very true. Interesting way to put it on, or the fact that there’s no price mechanism that regulates talent allocation, but it’s a strange ranking mechanism.

RAJAGOPALAN: You have people my age who put their JEE rank in their CV. It’s an embarrassment. If I still put my 12th standard board marks on my CV, I’d want to shoot myself. It just means the last, I don’t know, 20 years have been a total waste. It’s a funky thing, and I don’t want to blame them individually or ridicule them. There’s obviously a reason for this system, but I see this on LinkedIn all the time, like people put their JEE rank way beyond the years of JEE.

Gold and Land

Now, a couple of other trends I quickly wanted to pick your mind on. One very cool thing that you pointed out is the increase in gold. Gold has historically been a hedge against inflation in high inflation economies. In India there’s that gold obsession. We started getting inflation under control, but it was still going on. Gold is also very good hedge against crazy policy uncertainty. There was also that. Now you’ve introduced the third element to this, which is gold is a better collateral, a safer collateral, because other assets that can be collateralized like land are, one, too expensive, two, the transactions costs are too high, and it’s just too crazy.

Now I have a fourth element to add to this and tell me if I’m anywhere close to what you’ve observed. Does this also mean that there is a certain amount of women’s empowerment and better sharing within the household? What I mean by that is a lot of the gold is still held in jewelry. Is it women trying to stake a greater voice and say in how the wealth is both accumulated and saved and distributed within the household? Of course, it is available for collateral, so it’s kind of wise.

Part two of that, is the downside of that because we haven’t included women in the equities market and the systematic financial economy because either we just like to not have any financial literacy, and we generally make it very intimidating for women, and we kind of poo poo them when they show up? A lot of the women I know, even though they’re educated, like my mother’s generation, they don’t even like going to the bank. The whole thing is a bit much for them, even though they’re very competent, very well-educated women. Is that another element of what is playing out with this gold business?

PAI: To be honest, I haven’t thought of it that way because I also do probably don’t have enough data. It’s interesting that you say it. I would only say that, for example, when a girl is married into a different household, she does carry gold with her. Certainly, it doesn’t always stay that way. It does sometimes enter into as collateral. I haven’t thought of it that way, Shruti. I don’t have enough knowledge of this one to say that—

RAJAGOPALAN: Nor do I. This is just a hunch.

PAI: One, at some point, if they do have agency, they do have disproportionate agency over gold, you’ll want to see data in terms of gold loans to women, for instance. If what you’re saying is specifically true, I’ll probably want to then, and I could do that, I could try and look at gold loans to women. Are they increasing specifically? If so, then you know that she has agency on this matter, disproportionate agency, and that is reflected in her using gold to possibly try and use it for education or try and use it for a productive asset like a sewing machine. So I think will try and see that to really see if there is something there, but I’m not 100% sure yet. I don’t have the data point.

RAJAGOPALAN: There is a counter to that. My counter to the point that I made is maybe they just buy it more in jewelry rather than in gold bars or gold options or something like that, it’s harder to expropriate jewelry. It would be a public scandal if you make women take off their bangles and things like that. It would just be unheard of. Maybe that’s the reason, like it’s another layer of protection against all sorts of crazy. Plus of course, if you can wear something beautiful and also have it as collateral, it’s a multiple-level win. That’s what would be the counter to that argument. This gold loan thing that you’re talking about is super interesting.

Another is this is of course the older generation, the way women used to use chit funds to save up and buy gold, which is a sachetization scheme if you really think about it, same with sarees for Diwali, like all these big—Nalli and all of them used to have a chit fund version for the lower end of the market, even high end sometimes. We saw that disappear post 2001, and maybe this is the version of it coming back. That would be another interesting thing to see if you put in small sums of money.

PAI: My wife used to do this with Tanishq. They had Titan. I think they had one of these gold—Tanishq, it was. They had this thing where you go and put in X amount of money and that becomes a gold, saving on x amount of gold at the end of the day. There’s a company called Jar, which is fascinating. They have this roundup thing where you spend 90 rupees and 10 rupees, can round it up, and you can set an algorithm with each time you pay like 163 or 173 rupees, 27 rupees gets rounded off and buys a gold, like a digital gold.

RAJAGOPALAN: Oh, how lovely.

PAI: Yes. Some interesting things, and full credit to Joseph Sebastian who’s a colleague and like a joke that all Mallus have an obsession with gold, but in this case, land and gold is what he identified. I know that land is an area of interest for you. Like on land for instance, I also bring it later into the section of the— 

RAJAGOPALAN: On land, labor, capital, yes.

PAI: Yes. There’s another one on addressing, which is to do with why India has poor addresses and how much they cost. I would say that if we can just figure out, the government just figures out addressing and mapping, like digitization of land. some scheme by which it goes down to like nine square meters or something, which is the size of a room or whatever, and says if we can map India into that and then that becomes how UPI or Aadhaar became a very common thing. The amount of efficiencies and the amount of benefits and everyone can build on top of it.

RAJAGOPALAN: We’ll just unleash it.

PAI: Yes, reduces litigation because a lot of litigation is around land, as we found out. It’s incredible. It will improve addressing, will improve other benefits. For example, collateralization will improve, even logistics efficiency will improve, et cetera. 

RAJAGOPALAN: My solution to this is, we’re already a very highly taxed society, but governments need incentives like anyone else. Why did we have better land records under Queen Victoria? It’s because she’s a residual claimant and capturing that revenue from the land. One of my favorite movies on this is The Englishman Who Went up a Hill but Came down a Mountain.

It’s this hilarious movie about this small village somewhere in England where they think they have a mountain, but the geography surveyors, geographers actually show up, and it’s a few feet short. The entire village, to maintain its pride in its mountain, will actually collect soil and dirt and raise the height of it. It’s a charming, cute film. Once upon a time, we had geographers and surveyors as a class of government workers, and their sole job was to survey, map—and of course, a very large part of it was a little pernicious because it had to do with revenue. 

If we start having property tax and the core basic simple things—now everyone wants to protect their property. One of the best ways to show that the property belongs to you versus someone else is if you have a long record of paying taxes. The municipality has an incentive to guard it if they, in fact, can gain revenue out of it and actually funds them. There is a Georgian way of threading this needle very well. We just refuse to do it.

Now we have drones we can use to map. There are so many things we can do, but we just refuse to map, survey, and title. It’s a bit depressing. You have Karthik’s book out,.

PAI: I wanted to quickly get a stat, but I don’t think I’ll find it because there is a very interesting passage where he talks about how India is undertaxed on the property front. That is something that we’ve just not used. I think he also has the usual caveat saying we’re anyway taxed and this one. I’m just trying to find that particular—

RAJAGOPALAN: No, but we are very regressive because we are overtaxed on consumption. By definition, those taxes are regressive because poorer people spend a larger proportion of their income on consumption than rich people. If you want to have a truly equitable tax system, you want to truly be pro-poor, you got to actually tax property and reduce taxes on consumption. We’ve got it flipped exactly the other way. We can’t easily tax capital because in this globalized world, there’s capital flight.

We just have to find saner ways of doing it and somehow we only want to tax consumption and capital, which are the worst things to tax. Sometimes, I just wish there were a little bit more economic literacy.

PAI: Yes, you’re right. Actually, for instance, US—it’s not like I’ve studied the US super deeply or anything like that—in the US, for instance, property is significant. For example, I’ve seen this movie called Houseof Sand and Fog, where Jennifer Connelly loses her house because she goes through some phase which she forgets and the house is possessed. US is where when you pay property tax, and of course, the nature of the US, this thing where, for instance, schools are free, but you—anyway, sorry, school parallel, we should ignore.

Fundamentally, I feel US is a country where there’s a very strong local connection. Whereas India, the Indian elite, barring a few who have lived in a particular place all their life have no local connection or empathy. For instance, I live in Noida, and I live in this society, and I have no big interest or attachment to the places outside of it. I live one level above all of our -- , and I think of myself as far more India. I never think of myself as, “Oh, I’m sector .” Presently, US, for instance, they’re very proud. They think of themselves as belonging to a particular locality, for instance.

RAJAGOPALAN: Even the Noida example you gave, you actually bought into the fiscal federalism aspect because your society presumably provides all these basic public goods and water and electricity backup and basic security and the garbage is picked up and all of that stuff. That’s really why you chose into that society even more than Noida. What the US has managed to do is it’s managed to do that.

In India, we managed to do that through gated communities because the private sector has filled in for this terrible vacuum that local governance has left. We just thought local government is not important or enough in India. In fact, we didn’t want it. Because when they were writing the constitution, they said, oh, these villages are all like dens of parochialism and all of that. We need a state level and a union level. We’ll be like the Soviet Union.

The consequence of that is garbage doesn’t get picked up, and you don’t have traffic police, and traffic lights are a suggestion. In Noida, you have pizza delivery boys whose pizzas are stolen before they can deliver them. I actually read this in a paper. It’s all a little bit upside down. It just always keeps coming back to the same theme. I guess one important theme that I keep going back to every year I read your report is there are limits to what the Indus Valley can do without the scaffolding of a functional state.

In fact, they punch above their weight. They go above and beyond to solve a lot of public goods’ problems that technically are not the business of the private sector. Even past that point, there is not a glass ceiling; there’s a concrete ceiling, and the state eventually has to do its job.

PAI: This cocooning, as I call it, increasingly, what’s happened is you’re relying on the private sector to take over every single thing that would be called through public goods. When finally it’s down to you can’t solve for air. If Delhi could solve that, and I’m just waiting for a new fancy building which has a glass caves and the entire thing inside will be purified.

RAJAGOPALAN: I think it’s called Oberoi Hotel because they keep advertising minute to minute what the AQI is outside and what it is inside.

PAI: Probably, we’ll have a glass-enclosed dome, like a dome in which people will grow up and stuff. You don’t even know how much property tax you pay, it’s very little. There’s no strong connection with the land, and when more and more services are being co-opted by the private sector, I would say it’s almost like when you’re taxed, it also forces accountability from the government.

This is something I think I mentioned last time as well, that more and more, I think the Indian state is fundamentally saying, this time or so, they just reduced the number of people who are going to pay tax. It’s going to half. It’s 2% pay tax and then income tax, and it’s going to come down to 1%. A lot of the tax that you’re paying is going into GST or corporate tax, et cetera.

RAJAGOPALAN: Which means you’re taxing the poor instead of the rich; it’s mortifying.

PAI: Yes. India is a country I think a good way to understand it is it’s unique. It’s different interest groups. Some of these interest groups align. There’s an India 1 interest group. The India 1 interest group, because it consumes a lot of US products, wants to keep the rupee-dollar exchange rate toward itself. That means there’s a second order impact on manufacturing. The ideal exchange rate actually should be something like 110, I don’t know, 100, maybe. I’m not a microeconomist but actually, the real effective exchange rate says that we have, for example, overvalued.

The one luxury product India 1 loves is to send their kids to the US. For that, they have to keep the dollar at a certain rate. There’re interest groups like bureaucrats, which want to collect rent in terms of pensions, and they want to go back to a fixed—different interest groups are extracting their rent and tax in India. India is becoming a country where, when you get the ability to do that, you will extract that rent. It’s sad that we’re not sitting and solving that. The farmers, for instance, they want free whatever. The second-order effects of that are in pollution.

RAJAGOPALAN: Yes, groundwater.

PAI: Groundwater, yes.

RAJAGOPALAN: I don’t mind a certain minimum amount of rent-seeking and that’s just the way of doing business. You’re absolutely right. Now, the kinds of things you’re mentioning, it really threatens macro stability. Now this is a public bad for everyone. It’s not just a matter of a minor transfer that I took the subsidy away from leather apparel manufacturers and I gave it to textile manufacturers or something. Now this is much more systemic and there’s a much bigger thing that’s actually going on. You make these amazing connections, both in the report and otherwise, because I’ve been reading you for a long time.

Sajith’s Reading Habits

Before we end, I want to ask you, how do you read? Because I think that might be an interesting thing to know for people. I don’t just mean books because you read blogs and Substacks and you listen to so many podcasts. I have no clue how much time you have in a day, because you’re on top of all of it. I know some of it, of course, is just a direct input into your daily work and writing the report and things like that. How do you choose what you read and how do you curate that for yourself? Because I think people would be very curious now that they are consuming what you are putting out.

PAI: Yes. All right. Thank you for that. Yes, one change I did in my information consumption a few years back was to go into a highly dense—and a kind of joke to use this food metaphor for going into proteins, like what are the equivalent of proteins and reduce carbs. One, especially in the startup world, and I’ve discovered that podcasts have a disproportionate influence because it’s very hard to sit and create now because AI makes it simpler. There is a certain friction to creating content through writing. Now how AI will change it will be very interesting.

Podcasts meant that a founder could come in and share his three, four, five interesting things. So I overindex on podcasts a lot and a lot of the insights, a lot of the interesting perspectives actually come from podcasts. That’s one. The other one is I do consume a lot of reports, which are these equity research analysts and these macro reports. The World Bank put out an interesting one on India, et cetera. I think over time, what’s happened is really that as you read a particular thing, you have a certain mental scaffolding and then you add stuff to that. That makes it a little easier and easier to consume. I think these two are content formats that typically others don’t overindex on.

Podcasts and research reports, especially equity research reports, and try and take out interesting data points. For instance, there was a very interesting one on air-conditioner market. From that, two interesting data points came out by Jeffries or I forgot, I think it was Jeffries. Two interesting data points came out on how room air-conditioner market consumption and at what income level does it suddenly spike. The other one was interesting one on how import bans led to suddenly manufacturing growth in room air conditioners.

It’s keeping that curious thing and trying to see if there are very interesting patterns, et cetera. I think it’s a lot to do with how I read, I would say some of it is that I read the same thing that others do, 

RAJAGOPALAN: I’m not so sure. I don’t know too many noneconomists who are reading Karthik’s book with the kind of detail that you’re reading it.

PAI: Yes, that’s true. I’m halfway through. It’s a fascinating book. I know you’ve interviewed him and I’ve read the transcript of that as well.

RAJAGOPALAN: That was like climbing a mountain, speed-reading that book. This book is so big and the research is just so good. You can’t skim Karthik’s work. You have to really pay attention. I was cursing him the whole time. I was cursing him through the recording of the podcast. I was like, “Karthik, make this easier for the regular mortals.” You also read a lot of history. Is this coming from your Times of India days? Because I don’t see too many people who are VC investors or startup founders reading very much. I see all of them listening to podcasts, though. I think that fits.

PAI: I’ve always been an eclectic reader. People, typically they’re reading science fiction or nonfiction and nonfiction with tech fiction. I like to read diverse stuff. For example, three, four months around the time I’m working on Indus Valley, the first two months, a little more slow burn, the last month, heavy burn, I tend to pick up books like Lilliput Land, and Accelerating India’s Development and books like that. I read the one by Raghuram Rajan and Rohit Lamba was interesting as well. I do find history fascinating. Last year, I read India s Broken by Ashoka Mody, which gave a very interesting—

RAJAGOPALAN: Very pessimistic.

PAI: Very pessimistic. I think history is a particularly interesting thing. But I think, Shruti, it finally is down to, I think, over time, I’ve read so much and different things. Certainly unusual for a VC that there is an intellectual scaffolding, and it’s easier to absorb certain kinds of data and easier to connect some kinds of data. I think that probably is the reason.

RAJAGOPALAN: Like everything else, that also compounds, which is amazing.

PAI: That is true.

RAJAGOPALAN: There’s some advantage of being older.

PAI: Yes, that’s true.

RAJAGOPALAN: Thank you for doing this, Sajith. One thing I hope in the future is in addition to the decks, you also write a part of the report, because I love your writing. I’ve been reading it for a long time. That will also be very cool. That short, four-pager, five-pager of your insights, which always sound different when you write them versus when they’re in a deck versus when you speak. That would be lovely. A huge congratulations to your team. It’s a real public good, everyone reads it. It’s become this point where everyone converges. Thank you for doing this. Thank you for coming back.

PAI: Oh, thank you. Thank you for that. It was a pleasure. I’ll keep that in mind. What you said about writing maybe a four-pager, five-pager, I’ll keep that in mind.

RAJAGOPALAN: Thank you for doing this.