Ashish Kumar, Co-founder and General Partner at Fundamentum, discusses India’s evolving role in the global AI ecosystem and emphasizes the country’s potential to develop not just AI applications but also foundational infrastructure over time. Although India may not immediately match the massive funding behind organizations like OpenAI or Anthropic, Kumar believes the nation can thrive by creating specialized use cases, “small language models,” and domain-specific AI solutions. This approach, he argues, demands relatively less capital and can still generate significant impact.
He also highlights the importance of geopolitics in shaping India’s AI landscape. With potential trade barriers and export controls on advanced technology, India’s self-reliance becomes increasingly critical. Drawing a parallel to how the Indian Space Research Organisation built cost-effective space capabilities, he suggests that India can similarly build AI infrastructure to serve its own market—and potentially export it to other lower-GDP countries seeking affordable solutions.
He also highlights the importance of geopolitics in shaping India’s AI landscape. With potential trade barriers and export controls on advanced technology, India’s self-reliance becomes increasingly critical. Drawing a parallel to how the Indian Space Research Organisation built cost-effective space capabilities, he suggests that India can similarly build AI infrastructure to serve its own market—and potentially export it to other lower-GDP countries seeking affordable solutions.
Category
🤖
TechTranscript
00:00:00I think we are all slaves of momentum and ecosystem.
00:00:03Your chances of success in any area depends on how well that ecosystem is developed.
00:00:09For Flipkart, it was much harder honesty to create a business than what it required next set of companies to create it, right?
00:00:15In the West, entire ecosystem around AI is very intense.
00:00:20If AI was not part of this business, how relevant will this business continue to be?
00:00:25Innovation in digital era is never only about the final output.
00:00:30India also has a little bit of a talent problem on the AI side when it comes to actually their capability to create models and so on.
00:00:38Can you replace all humans? Probably not.
00:00:41But can you replace 60, 70, 80%? We are starting to see a lot of those things, at least in audio.
00:00:46What is good for the nation is ultimately good for the business.
00:00:49What is good for the business is actually good for people like me who invest in those businesses.
00:00:53We created services because it just made sense to do that.
00:00:57US created products because for them, creating services companies is a lot more difficult.
00:01:01I think we unnecessarily glorify saying that, you know, since US creates product and India creates services, they are better and we are not.
00:01:08I don't think it is true.
00:01:09I think we have moved away from that world today.
00:01:12It does look like that there will be some artificial walls that will be created between nations.
00:01:17Today, of course, India is very nicely poised from a geopolitics perspective.
00:01:20But will we always be like this? I don't know.
00:01:23Hello and welcome to the Escape Velocity podcast by Outlook Business, where we talk to the movers, shakers, thinkers and doers of the tech and startup ecosystem.
00:01:44Today, we have with us Ashish Kumar, the co-founder and general partner of Fundamentum, a venture capital fund which has invested in unicorns like Spinney and FarmEasy.
00:01:55Hello and welcome Ashish to the show.
00:01:57Thank you, Deep Shekhar. It's a privilege to be here.
00:02:00So today we want to talk about how India should play the AI game.
00:02:04So how we do it is, I think, will be great if you can give a bit of an intro of yours.
00:02:10Just imagine that we have beers in our hands, which we don't, but just imagine.
00:02:15You know, before I do that, Deep Shekhar, I think since you mentioned about beer, I kind of remember my first job was Microsoft in Redmond in the US, right?
00:02:24And this is when I was 21, 22 years old.
00:02:27And we had this tradition that every Friday, 4 p.m. or so, my manager used to open multiple bottles of beer, right?
00:02:36And we used to celebrate, I don't know what, honestly, because this was the heydays of Microsoft.
00:02:41This was not even when Google had made their first big move, right?
00:02:45You used to celebrate your stock price, ESOP's going up in value.
00:02:48Perhaps, because I can tell you that the stock price of Microsoft did not move for a good 10 years or so in the 2000s timeframe, which is when I was there, the earlier part of that, right?
00:02:59But the stock price, of course, moved a lot, lot, lot more when I forgot about it, right?
00:03:03So, which has been a pleasant news.
00:03:05But, you know, it's like, it's a privilege to be here.
00:03:09My background is that I was a technologist, then I became a technology entrepreneur before, you know, starting a few businesses myself.
00:03:18And then Fundamentum is a venture capital fund, which was being started by Nandan Neelakani around 7-8 years back or so.
00:03:26And I came on board like a few weeks prior to the fund officially started.
00:03:31So, if you can tell me a bit, you know, I see your posts on LinkedIn.
00:03:35You talk about AI a lot.
00:03:36But tell me, OpenAI is not in India.
00:03:40Cloud Anthropic is not in India.
00:03:43Microsoft is not in India.
00:03:45They are the people who are investing tens of billions of dollars.
00:03:48Masayoshi Sun and Sam Altman are talking about a $500 billion project.
00:03:52In India, why are you so concerned about AI?
00:03:56So, see, I think when you have to become a self-reliant country, right?
00:04:01And in the spirit of becoming a self-reliant India, and we have to do that once if you want to be the top 3, 4, 5 economies.
00:04:07We are already in the top 5th or 6th one today, right?
00:04:10And we will very soon become top 3 economies.
00:04:13We have to own the basic infrastructures at all levels, right?
00:04:16And AI is a very basic infrastructure in the digital layer.
00:04:20And what is good for the nation is ultimately good for the business.
00:04:25What is good for the business is actually good for people like me who invest in those businesses.
00:04:29The reason I care about AI is that even though at an infrastructure layer, we may not be there today, but as a country, we will have to get to that, you know, in a mid to short term time frame, which is why I care about AI, which is why I am very, you know, bullish about people in India who will create AI, not only an application, but also infrastructure.
00:04:49But today, I think the biggest debate in AI in India is whether we should play at the infrastructure layer where tens of billions of dollars are required or should we play in the application layer?
00:05:06What do you think about it?
00:05:07And as an investor also, are you looking at AI, which kind of plays are you looking at?
00:05:14Yeah, so multiple questions in that one question, right?
00:05:17So, see, I think, like I mentioned in the beginning, that we have to, in the mid to long term, we definitely have to play the infralayer as well, that we have to own our own AI infrastructure.
00:05:31And like that, we will have to own an infrastructure in multiple other digital layers as well.
00:05:36But our path need not be what, let's say, US followed.
00:05:40Our path could be very different.
00:05:41So, I firmly believe that as a country and as an entrepreneur, anybody who is actually trying to build a business there, they have to start at an application layer.
00:05:51And like what Nandan talks about that, can we become a use case capital of India?
00:05:55We absolutely should do that.
00:05:57But that has to be only the first step.
00:06:00Then we have to go, you know, like we have to start elevating ourselves up and eventually also own the infrastructure.
00:06:06And the way to do that is to start with an application, then create small language models.
00:06:12And the beauty about small language models is that you don't need that kind of resources, whether it is in the form of capital, whether it is in the form of even the amount of or the density of talent that we require.
00:06:22But it requires a lot more business understanding, right?
00:06:25So, you could start, you know, let's say, small language model by verticals, by use cases, and then go, when you have enough number of those SLMs or the small language models, there are ways to stitch together many of these to become a, to create a large language models.
00:06:41And DeepSeek is one company which has shown us one way to do that, which they call it mixture of experts.
00:06:46But there are other ways that, you know, one could potentially experiment with and then create that.
00:06:50So, I do think that we will have to, in the mid to long term, own the infrastructure, right?
00:06:55All layers of infrastructure, not only the large language model, but also, let's say, on the semiconductor side, on the energy side as well.
00:07:01But we have to start from being a use case, building multiple use cases.
00:07:07Infrastructure, like you mentioned Nandan.
00:07:09Nandan says a very interesting thing that models will be commoditized.
00:07:14So, what, how do you look at that?
00:07:16So, if models are commoditized, and we probably will be very difficult for us to go at the cutting edge, then will people use it?
00:07:25Will startups use it?
00:07:26Will companies use it if we are not at the cutting edge?
00:07:29Then what should be the play there?
00:07:32I think there are multiple assumptions that one has taken, right?
00:07:36One is that we will live in a world which will be where there are no trade barriers.
00:07:41You know, all countries will be able to do trade equally with each other, with ease.
00:07:46Everything will be open, available at the same time, and so on and so forth.
00:07:50I think we have moved away from that world today, right?
00:07:53And it does look like that there will be some artificial walls that will be created between nations.
00:07:58We may not even have access to the latest and the greatest infrastructure, whether it is AI or any other thing, in general, over a period of time, right?
00:08:06So, we have to be ready for such an eventuality.
00:08:09Today, of course, India is a very nicely poised from a geopolitics perspective, but will we always be like this?
00:08:14I don't know, honestly, right?
00:08:15So, that's one point that I would make.
00:08:18Second point, innovation, deep shaker in digital era is never only about the final output.
00:08:26So, what you refer to as is that the models will be commoditized.
00:08:31Yes, models will be commoditized, but the final output will be commoditized, likely.
00:08:35But when you start trying to develop a large language model, there are significant innovation in processes, different nuances that you learn along the way, which are very, very important in an area where I feel that we are at a very, very early stage.
00:08:52So, I don't think AI, even from an infrastructure perspective, we are in the fag end of the entire infrastructure creation.
00:08:57We are probably in a very early stage.
00:08:59Maybe we have traversed maybe 10 to 20% of journey and we still have to traverse another 70, 80%.
00:09:04So, very early, so that we will have to create these things, even though it may be commoditized, it may not be number one available to us.
00:09:12Even if it may be available to us, it may not be fully indenized or be applicable to the use cases.
00:09:19Even if that is available, the price point at which that may be available may not necessarily be applicable to us, given that we are a smaller GDP per capita company.
00:09:30So, which is why I think we need to do that.
00:09:32And I will give you another, maybe one example, right?
00:09:35So, if you look at space, right, what ISRO has been able to do is to create similar space infrastructure at a fraction of the cost, right?
00:09:44And India is now benefiting.
00:09:46If you look at space, India is probably one of the top three, four, five countries globally, right, which has this infrastructure and we are reaping the benefits of that.
00:09:54We need to, at a very minimum, get to the same level in AI.
00:09:58Our use cases will be different, our talent is different and which is why we need to go on that path.
00:10:04So, that was one question.
00:10:05I think the second question that you had asked was that, could you repeat the question, the other part of the question that you said?
00:10:10The second question was that, you know, you must have AI startups, kind of, you must be thinking what kind of AI startups to bet on in India.
00:10:20Right.
00:10:20So, what is the thesis that you have?
00:10:22For example, one very interesting thing you just said is about the geopolitical walls.
00:10:28Right.
00:10:28That will we have access to the latest and the best.
00:10:31As a venture capital investor, are you also keeping a keen eye on geopolitics?
00:10:37Because if there is a wall like that, you know, happened, China erected a wall from its, you know, on its own against the world.
00:10:46But today, given the geopolitical context we are in, the countries are also putting export controls, you know, on the latest technology.
00:10:56So, do you think that also presents an opportunity?
00:11:01Is that a lens that you are looking at the AI startup opportunity at?
00:11:07Absolutely.
00:11:07Right.
00:11:08So, see, I think it is one of the parameters that whenever you are investing in any company, you do think about macros, you do think about geopolitical risks and rewards as well.
00:11:18Right.
00:11:19The challenge is that sitting where I am, I will not be able to fully comprehend how will the different nations behave over a period of time.
00:11:29Right.
00:11:30So, you will factor that in, but I don't think you can factor that in completely.
00:11:33Right.
00:11:34So, there are other ways to manage that risk or take advantage of that.
00:11:37If at all there is an advantage or there is a reward framework that gets created.
00:11:41So, it will give some weightage, not too much is what I would say.
00:11:45Right.
00:11:45But let's say I do think that from India, we have been a very open country and I think that part is fairly clear that I don't anticipate that we would necessarily put up a lot of controls saying that, hey, what Indian models, will this be available to different countries or the Indian infra will be available or not.
00:12:04In fact, I think that there will be a lot of countries who may not necessarily align with, let's say, only US or who may not align with only China and they want to remain sort of more neutral the way India has managed to.
00:12:16They probably would say that, can I use the infrastructure that has been created by India?
00:12:21Right.
00:12:22A lot of these countries are also lower GDP per capita countries.
00:12:27So, for them also to take advantage of the infrastructure that, let's say, a more richer country is like the US has created may not necessarily be very appealing.
00:12:38Right.
00:12:38Think about like what the adoption of UPI across the world.
00:12:43Right.
00:12:43And it is in early stages, but it has started to happen.
00:12:46I think we have that opportunity.
00:12:48So, whenever we are looking at any investment in a company, now there are two lens.
00:12:54One lens is that, how can you take advantage of all the opportunities that are available in India?
00:13:00But the second question also is that, is this something where you could potentially export it to other countries?
00:13:06Right.
00:13:07And again, I will take the example of space.
00:13:10You know, if you look at, let's say, some of the defense companies that it has started to happen.
00:13:13And I see the same opportunity is available in the AI world, especially if you're starting to create infrastructure.
00:13:19So, the key for us is to be able to, when you ask that question, that what are the kind of opportunities that we see?
00:13:26What is the thesis that we have?
00:13:29Our thesis is fairly clear, right?
00:13:31There are bulk of the investment that in the AI world that we would do.
00:13:35We would look at application layer companies.
00:13:37And we are saying, but application layer with some differentiation or some capability to differentiate on creating either a small language model or access to a data which is a lot more proprietary, right?
00:13:53And not easily available to, let's say, the horizontal large language models, right?
00:13:57Or maybe some algorithmic invention or innovation that could help create the small language models in that particular use case.
00:14:05If we, so we would write, likely not invest only in, let's say, wrapper companies, because we think the defensibility of wrapper, when I say wrapper, wrapper over built on top of, let's say, open AI or a Gemini or, you know, any of the language models, even if they are built only on the top of the open source solutions.
00:14:24But we would want, because differentiation and ability to, you know, from a barrier to entry perspective is will be very, very low there, right?
00:14:32Especially with the rapid development and the software.
00:14:34So we will want some edge in the, either their current capability or we have to be very, very convinced that this is a capability that they'll be able to develop, let's say, in the next year or two, right?
00:14:45So that's how we would see.
00:14:48And my, the visibility that we have, we come in slightly when the companies are more in a early growth stage or a series B stage.
00:14:55But when I engage with the ecosystem, I am starting to see that not only on the software model that we are referring to people, there are interesting companies who are trying to do something with synthetic data, which is required, let's say, for AI.
00:15:07I am starting to look at companies which are trying to figure something out on the semiconductor side, right?
00:15:13And semiconductor, you don't have to be NVIDIA necessarily to just win that game, right?
00:15:17You could also, there also you could do vertical, verticalization saying that, you know, this kind of chipset is only useful for, let's say, you know, certain use cases, marketing use cases or healthcare use cases, right?
00:15:27So some of those, because, so there are some innovation that there you could do.
00:15:32There are some people who will probably say that, can I stitch multiple of these models, multiple these infrastructure and create some sort of orchestration so that the net result of this is that I can achieve the same thing that the large language model or a large model is achieving in the US at a fraction of the cost, but with some trade-off.
00:15:52So those are some of the observations that I have had.
00:15:55Another question, like you said, that there is a lot of opportunity in these vertical AIs and the differentiation, like you said, will not just be the wrapper, but the data, the access to proprietary data.
00:16:11For example, you have companies like Spinney, FarmEasy, which have this proprietary data set.
00:16:18So do you think it's an opportunity actually for these big vertical companies who have the funds, if they can deploy it and make a vertical AI and sell it?
00:16:33I mean, it could be that it's an AWS moment.
00:16:37It could be an AWS moment for a lot of those big vertical companies.
00:16:41It is absolutely happening.
00:16:42This is a great, great, great point, Deepshakar.
00:16:45I think if you look at the Zomato's announcement today or day before, they actually launched Nugget, right?
00:16:52Which is an AI tool for a particular use case, which happens to be, I think, customer service or customer care, right?
00:17:00That is what they have launched.
00:17:01Usually what happens when a consumer company becomes very large, they actually deploy and use and create multiple tools inside, right?
00:17:10Which are relevant to multiple other companies, which will also become, you know, large over a period of time.
00:17:18And the good team, especially because a lot of these new age technology companies have very technology DNA.
00:17:24They have an ability to basically look at that, take that entire infrastructure and basically spin it out or, you know, effectively create a different offering to that, right?
00:17:36And sort of what you were alluding to that, is this an AWS moment, right?
00:17:40Amazon that did that with multiple other offerings, right?
00:17:45AWS was one offering, but they also had a warehousing as an offering, right?
00:17:48So multiple of those services.
00:17:50So I think the same thing Zomato showed and they have probably the first ones to have talked about it a lot more publicly.
00:17:58But I know that there are other companies which are in the process of doing that and you will see a lot of innovation there as well.
00:18:04But I don't think the innovation is going to be necessarily limited to that.
00:18:09What would also happen is that many of the team members inside those large companies, whether it is Spinney or FRBZ or any other company, will also see that opportunity.
00:18:21So before maybe a large consumer companies launches it out, right?
00:18:26Because they also want it to be finished to a particular layer, right?
00:18:29But the teams inside, they would come out and they will basically say that, hey, this is what we were using in this company.
00:18:37And the same offering will be required by other companies.
00:18:41And you will notice that there, again, vertical systems will be a lot more in play rather than horizontal systems.
00:18:50Got it, got it.
00:18:51Also, another thing that comes to mind is, you know, as journalists, we also get pitches from startups.
00:18:57And it seems every other startup today is an AI-powered startup.
00:19:01As a venture investor, when you get the pitches, do also a lot of them claim to be AI-powered?
00:19:07And what is really an AI startup?
00:19:09I mean, how do you, if there's some founder watching this, how do they know that, okay, in their pitch to you, should they write AI-powered or not?
00:19:18I think we are in an era stepping back, actually.
00:19:23So, if you look at, let's say, maybe last 15 years, right?
00:19:26And this is the time that I have been investors in some capacity or the other.
00:19:30I don't think I've met any company which did not claim themselves to be a technology company, right?
00:19:36The same thing is going to happen to the AI world.
00:19:39So, AI is the new technology.
00:19:40And the reason, clearly, is that it's a wave, right?
00:19:47And everybody basically wants to take advantage of that wave, rightly so.
00:19:50The way that I look at whether it is truly an AI company or not is whether AI is strategic to this business or not.
00:19:58And I'll give you an example.
00:19:59One way to think about whether AI is strategic or not is that if AI was not part of this business, how relevant will this business continue to be, right?
00:20:09If they continue to be very relevant, I think it's very, very high chances that AI, at least at this point of time, has been a force of it, right?
00:20:18And then there have been some, you know, people will push, try to push some AI because it is cool or they probably would think that maybe that is the minimum that they need to do to raise capital, right?
00:20:30And I think most sophisticated investors will see through it.
00:20:34In my mind, there are two types of AI startups, right?
00:20:38And you kind of talked about it.
00:20:39One is an AI startup itself.
00:20:41So, they are effectively selling some sort of AI services, right?
00:20:46So, OpenAI or ChatGPT is one example and Claude and Anthropic could be another one, right?
00:20:53Similarly, there are people who have created small language models for certain use cases.
00:20:56Then there are some people who are, so, you are selling AI service, right?
00:21:01Usually, these are software applications, but some of this could also be like we were talking about, let's say, chip.
00:21:07Some of this could also be, let's say, data center conducive to, you know, creating infrastructure for many of these AI applications to run.
00:21:16So, I kind of call them as AI startup because they are effectively selling their AI as an offering.
00:21:23AI powered startups are what, you know, was technology enabled services 10, 15 years later, right?
00:21:30So, the point, and if you look at today, AI minus robotics, right?
00:21:36So, if you look at AI has a capacity or the capability over a period of time to replace a lot of digital work that was being done by people, right?
00:21:46So, they are saying that, so, we have a portfolio company which creates audio content, right?
00:21:52So, the way that it used to happen earlier was that there will be a content architect who would think about the content.
00:21:58They will have some way to figure out whether this is the content I should create or not.
00:22:02And then there will be the execution play where they would actually get artists and then the content will be created.
00:22:07Now, using AI, right, and if you do it well, it is possible for you to create a reasonably good audio content, especially on the AI, right?
00:22:19Can you replace all humans?
00:22:21Probably not.
00:22:22But can you replace 60, 70, 80 percent?
00:22:25We are starting to see a lot of those things, at least in audio.
00:22:28In video, I know that that number is probably 30, 40 percent today and we will continue to become better there in terms of the total work that a machine or the AI can do.
00:22:39So, these are the companies which I will call as an AI-powered startup, right?
00:22:43And when I say AI-powered startup, again, going back to the framework that we spoke about is that if we remove AI, will this company remain relevant?
00:22:53The answer in my mind is no, because your economics does not work if you remove AI.
00:22:59Because it has a very, if you had only 5 percent improvement, then it would, I mean, without AI, this business can work.
00:23:07But if you, the impact is 50 percent or somewhere in that range, then you know that AI is very, very critical.
00:23:15So, those are the businesses that we like to find.
00:23:18And we say that, okay, you know, like this is where AI is useful and more than often, the good companies will actually create some of the, some proprietary, some foundational models there.
00:23:31So, if you were to again create, let's say, for models for audio, one is to create a horizontal audio model.
00:23:39The other is that you could say that, you know what, for history, I will create a different audio content or history, Indian history, I could create different audio content, right?
00:23:48And audio models and so on and so forth.
00:23:49The advantage is not only that you have more relevant data, but the advantage also is that you also require less resources, both on the training of that model or inference, you know, when you are asking for information, which I think is lot more useful and relevant for a country like us.
00:24:06Got it.
00:24:06Also, understand that when today you are looking at AI startups to invest in recently, it was very different.
00:24:16It was e-commerce platforms, SaaS platforms and the like fintech.
00:24:20So, today when you are looking at AI and so much is happening in AI, there is somebody in the US, there are people who are looking to invest $500 billion into AI.
00:24:31So, has the strategy of how much to invest in a series A, in an AI startup, what is the math?
00:24:40Has that math changed?
00:24:42Is it a bit different than the usual?
00:24:44How much to invest in series A or B or C?
00:24:49How do you think about that?
00:24:51Great question.
00:24:53See, venture is a very competitive field, right?
00:24:57So, and the world is usually the playground, right?
00:25:02Especially in digital businesses, right?
00:25:04So, and especially because we are a lot more open country.
00:25:07So, which means that if we are investing in any company, we also have to look at what is happening in the same arena.
00:25:15And if somebody has access to $500 billion, right?
00:25:18And you are only talking about, let's say, $50 million, I think the math may not work out, right?
00:25:24Even with the indianization, even with the low cost, even with, let's say, a lot more algorithmic innovation, the ratio can't be $10,000.
00:25:33One is to $10,000.
00:25:34The ratio, you could say that, hey, one is to five, one is to 10, one is to 20, right?
00:25:38If you're brilliant, you could be maybe one is to 100.
00:25:40Somewhere, you know, on the algorithmic side, right?
00:25:42So, there are certain areas I don't think you will be able to necessarily play from India today from a business landscape perspective, right?
00:25:51Now, we'll, and this is where, let's say, the role of sovereign or the country will also come into play, right?
00:25:57And we don't necessarily, honestly, have to create $500 billion worth of infrastructure.
00:26:01The infrastructure, because we are also a smaller economy as compared to, let's say, what the US has, right?
00:26:06So, we also have to be in that range.
00:26:08And if we don't necessarily have to create, let's say, large language models, we have to create small language models,
00:26:16then that is another factor where we need less resources or less money and so on and so forth, right?
00:26:22But to your question that when we think about, let's say, investing in any company, do we think about, you know,
00:26:28how much money that we should be investing in any business?
00:26:31In general, AI businesses today would require more capital than, let's say, the standard SaaS businesses that were being kind of created.
00:26:41Also, because the basic building blocks are still being created, right?
00:26:46What happens when an ecosystem matures, there are a lot more elements, basic elements, foundational elements which are available.
00:26:54So, when they are available, you could simply reuse it rather than creating your own.
00:26:58And I'll give you an example, right?
00:26:59Let's say, in the consumer internet era in India, when Flipkart started, they had to create that entire infrastructure.
00:27:07They had to create, let's say, the payment infrastructure.
00:27:09They had to create, let's say, the last mile infrastructure.
00:27:11They had to create, you know, inventory holding infrastructure and so on and so forth, right?
00:27:16They also had to create, let's say, that entire, you know, customer care and all that.
00:27:21Cash on delivery.
00:27:22Cash on delivery, right?
00:27:23So, later, you know, when Nisho started, as an example, they could, and they are also very large and they are a very interesting company as well.
00:27:31But they could reuse or they could use multiple infrastructure that were created with third parties, right?
00:27:37And the net impact to all of this is that the first company which will create will always require more capital.
00:27:43And you could argue that the capital efficiency for them will be lower.
00:27:47But as ecosystem matures, you know, those are the advantages that you are able to get.
00:27:52So, I think the same, and this is, by the way, this is true across any time.
00:27:56This is the usual curse of the innovation cycle, right?
00:27:59The first one will always require more capital to do that.
00:28:02So, we are in that cycle today, right?
00:28:04If one were to, but I think the innovation cycles are the, is also shortening very, very quickly, right?
00:28:11In the AI world, six months is an eternity, right?
00:28:15Whereas in the traditional internet, you know, maybe that number was three years as an example, right?
00:28:21So, what is going to happen now is that we, in India, when we have to invest, we have to look at marrying product expertise with some domain expertise, right?
00:28:34And go for those models or those businesses where, you know, you have a combination of, you know, like product brilliance and business brilliance, right?
00:28:44In the US, you don't have to necessarily think about a business to begin with.
00:28:49So, when OpenAI, as an example, came, did they know about enough use cases?
00:28:53They probably did not.
00:28:54It was a research project which was in the making for a good seven, eight years.
00:28:57And then they got to a particular stage where they thought it was good enough for them to kind of just put it out and customers will start paying it.
00:29:05We know that and this is what Sam Altman also said that even at $20 per month that they're charging or $200 that they're charging for, let's say, deep research, that business does not make sense.
00:29:15Which means that on the business model, there is still, you know, a lot more work that will happen.
00:29:21And US can afford it, right?
00:29:23Because that's how that economy kind of works.
00:29:26In India, I don't think that is the right way to do it, right?
00:29:29Which is why we will have to, our entrepreneurs will also have to be smart to say that whenever they're starting a business, they need to have a CTO or a chief AI officer who understand models, who understand how to make use of the existing models and or if need be, how to even create their models.
00:29:47They also need to understand that they have to have a domain person.
00:29:50So, and they have to go deeper into a vertical, choose the vertical, figure out specific use cases and then create that business.
00:29:59And those are the businesses in my mind will require a lot less capital than the $500 billion that you are referring to.
00:30:06So, how much should a founder of an AI startup and a series A or series B, you know, what is the range that they should be looking to raise as compared to what was happening in the platform economy in, let's say, consumer tech,
00:30:19e-commerce, edtech, so compared to that, how has that shift happened?
00:30:24So, I think the answer will also depend on the use cases that they are solving, right?
00:30:29But let's say if it is a consumer use case, right?
00:30:33In my mind, the consumer use case, the amount of money that if you are doing with an AI should probably be 20% more than what a business could have been started without AI.
00:30:44It is not a multiple of that.
00:30:47It is still 20%.
00:30:48The reason I'm saying 20% is that the X amount, let's say, is used for the basic business and reusing some of the AI infrastructure that are available to you from third parties.
00:31:01But 20% money or 20% resources is what you need to use to start creating a competitive edge, whether it is in the form of, you know, data that and you putting together data in a fashion that it is more useful to the machine or the AI.
00:31:18Or it could be some algorithmic innovation that we are referring to, all stitching together and creating a unique small language model, which is a lot more, a lot less resource hungry and creates the same output in a, gives the same output or better output in a smaller amount of time and resources.
00:31:41So that's how I would think about it.
00:31:43If you do the same thing on the, let's say, B2B side, right, which is the B2B software, it's a slightly different story, right?
00:31:50And we can maybe talk a little bit more about, let's say, the SaaS versus AI SaaS, how that is happening.
00:31:57Before we get into AI and SaaS, I just need to understand this, that let's say $5 million Series A was the norm.
00:32:04Right.
00:32:04So now you are saying 20% more, $6 million.
00:32:07$5 to $10 million I think is still good.
00:32:10In my mind, the Series A stopped being $5 million two years back actually, right?
00:32:16But I think $8 to $10 million is the typical Series A that I think, if you look at like relevant company, that's what the numbers are today.
00:32:28Series B has probably become maybe $20 to $40, $45 million.
00:32:32So I would probably say instead of $10 million, you have to think about $12 to $15 million.
00:32:38And it is also a function of, honestly, how much confidence does the investor have in your capability to be truly able to create competitive edge in some model?
00:32:50India also has a little bit of a talent problem on the AI side when it comes to actually their capability to create models and so on and so forth, right?
00:33:00If you look at like maybe last 10-15 years, having PhD in the co-founding team was necessarily not a good thing, right?
00:33:10Whereas if you look at the state of where AI is today, I think it is almost a necessity that you need to have a PhD in the co-founding team or at least a very plus one layer team.
00:33:21Because we need people with independent thinking, right?
00:33:27You know, PhDs by their training are trained to think a little more independently in terms of technology.
00:33:33And I'm talking about PhD in technology or data, right?
00:33:37So those are some of the points that I would make.
00:33:42I want to pick at this talent problem.
00:33:46I think there are a lot of people saying this, that high-tech talent is going out even now from India.
00:33:51And this is not like 20 years back when there was no capital in India.
00:33:56There is capital in India.
00:33:58Is it not that insufficient amount for deep tech?
00:34:01Is that the problem?
00:34:03Or are there a lot more better opportunities outside?
00:34:07And you will not even get a sliver of that in India.
00:34:09Is that the case?
00:34:10Why is the best talent going out?
00:34:13And connected to that, like you said about the talent problem,
00:34:17can people like you go to Silicon Valley and find the best people working in Microsoft, Google, Anthropik, OpenAI?
00:34:27Because there are a lot of Indians there.
00:34:30And tell them that I'll give you $10 million, $15 million.
00:34:34Come and do it for two years and let's see.
00:34:37Are people like you willing to do that?
00:34:41So, see, I think two questions that you referred to, right?
00:34:46One is that from a talent problem, can we even create the companies that are required to be there?
00:34:55And the second is that can we bring talent from outside to India, right?
00:34:58Because China has been able to do that.
00:35:00Yeah, so maybe we can talk about the second one first, right?
00:35:04Since you talked about China, right?
00:35:06I think we are all slaves of momentum and ecosystem, right?
00:35:12And people will, your chances of success in any area depends on how well that ecosystem is developed.
00:35:20Like we spoke about the example of Flipkart.
00:35:23So, for Flipkart, it was much harder, honestly, to create a business than what it required the next set of companies to create it, right?
00:35:30I mean, they had a different challenge that, you know, Flipkart was there, big and so on and so forth.
00:35:34But in the West, especially, let's say, US and likely China, right?
00:35:40The entire ecosystem around AI is very intense, right?
00:35:45So, you know, you go to any meetings, you go to any conversations, you know, there are different people trying to do something truly meaningful, a lot more deeper than superficial conversation that you'll usually have in India.
00:35:56And that is the reason why people are also kind of saying that, hey, as founders, founders are very smart, have to be smart, right?
00:36:03It's a very intense, competitive world.
00:36:05They are saying that if I were to start a company using AI as a technology, should I be doing it in India or should I be doing it in the US or let's say any other area where the ecosystem is not mature?
00:36:18Many of them are choosing a path to say that I would do that in the US because there are 30 people that I can speak to on a daily basis and I'll probably get more and more ideas.
00:36:30Because you're also developing ideas, you know, over a period of time.
00:36:34It's not that, you know, you think of an idea today and you start business and that it will remain as is.
00:36:38It does not, it is never like that, right?
00:36:40So that's the challenge that India has had today and which is why I think I am, I agree with you that a lot of people start wanting to start, they are going to India.
00:36:52But like we mentioned, I think there is no one formula that how it works that, you know, in the US, again, if you look at the kind of AI company that people are usually creating,
00:37:03they are creating infrastructure, they are creating horizontals, that's what is works there.
00:37:08In India, you know, you would likely not be able to do that and I don't think people should look at that as a negative.
00:37:15In fact, you know, when one door closes, the other door opens.
00:37:18And the other way to look at that is that can I marry business use case with, you know, like some potential model improvement that are only relevant in this and then I combine this.
00:37:29And then I have access to capital from people like me, which was not as easy, let's say, you know, maybe five years back, so on and so forth.
00:37:38Right. So so and whereas if they go to the US, I think they will also find that there are other people in the US who have been working on AI, horizontal AI for last eight years.
00:37:50How do you compete with them? Right.
00:37:52So it may look that the ecosystem is very developed and I will be able to do something, but there are other disadvantages to it.
00:37:58Right. So there are different people who will choose.
00:38:00In my mind, people who are a lot more of a technologist, do not have as much familiarity with the domain, will likely go to the US and they probably should go to the US.
00:38:10Whereas people who are have experience or expertise in both areas, they will be able to partner, partner with and they will be able to start in India, partner with somebody like a technologist who could go deeper down there because they will be able to demonstrate that they have access to capital sooner than what.
00:38:29Let's say somebody in the US could potentially have had now.
00:38:34So that's the that's the first question.
00:38:36Second question, can somebody like me go to the to US and basically speak with people from Microsoft and bring them back?
00:38:44I'm not a fan of that.
00:38:46You know, I think like I mentioned, we are all slaves of momentum.
00:38:50I don't think to the any founder, you can convince too much as investors.
00:38:56I believe that our job is to either agree or disagree with the founder, not to convince them.
00:39:02So if I were to convince them and the founder is truly convinced in one conversation or two conversations from somebody like me who will always or who should always know lesser than what the founder knows about that subject.
00:39:15In general, it is a bad news. So I am not a fan of that.
00:39:19I certainly the way that I would do is that, you know, there are a lot of people who are in two minds on whether to start, let's say, Indian guys want to start in India, start in the US and and all of that.
00:39:31But I will always be happy to talk to them and I do talk to them, right, you know, like today and saying that with the intention that somebody, let's say, in India wanting to start up, can they and this person in the US, right, come together and basically utilize both of their skills and create something nice, which should only come from India.
00:39:53So we have to always look for what is the India advantage in this business.
00:39:58If there is no good answer to that question, then it's a bad idea, right?
00:40:03I think so. I won't even honestly, China is also a lot more mature, right?
00:40:07So China created DeepSeek.
00:40:08I won't even encourage entrepreneurs in India creating for profit businesses to even attempt doing a DeepSeek.
00:40:16I would, because again, DeepSeek has the same problem, you create a model, you have a way to monetize, which OpenAI has shown, OpenAI itself is struggling with that, you are, your value proposition is lower cost, how will you make money, right?
00:40:31So you will have to start from a business first mindset, you will have to start with a use case mindset, right?
00:40:38And then see that how can I either make a process more cost efficient or I can I add a higher profit pool to a particular industry or a domain and then build this entire solution out.
00:40:51But DeepSeek, they claim that it was made within $6 million and there are questions about if it was that amount.
00:41:00But assuming that it is $6 million, it is better than some of the OpenAI and Meta.
00:41:06So then there is a use case, right?
00:41:10If you can make something like that from India and then it is very cost effective.
00:41:16Isn't that then a strategy that?
00:41:19So, see, I think we are a country, we need to see certain business cases being solved sooner than the amount of time somebody will have a patience,
00:41:35let's say in the more developed economies or more developed economies.
00:41:40They have basic profit pool being captured by existing companies and which is why their objective usually is always to create new profit pools, right?
00:41:51So, if you think about, let's say, OpenAI or the Anthropic and all of these guys put together, right?
00:41:56I think they have generated a total revenue of what I think OpenAI is some four odd billion dollars.
00:42:03Anthropic is also probably close to a billion or so, right?
00:42:06So, if you're going to put all of this together, it's maybe eight to ten billion dollars, which has happened in a span of what two to three years, right?
00:42:13Of revenue.
00:42:14Of revenue, right?
00:42:15And I'm not even counting, let's say, what Accenture is doing or what Microsoft is doing.
00:42:19You know, that could be significant.
00:42:21That would be tens of billion dollars, probably.
00:42:22So, the point is that this was a revenue which was not available, not present in the world two years back, right?
00:42:29So, those companies are great at creating and they have a need to do that.
00:42:34It is not, I disagree with the fact that, you know, they are smarter than us.
00:42:39They're not, right?
00:42:41I think people everywhere are smart.
00:42:43It is just that people choose different problems to work on depending on what is available to them.
00:42:48Whereas in India, we don't have to do that.
00:42:50We have some basic problems or some basic business problems not sorted out.
00:42:55Why should we not look at that and then use technology to serve that?
00:42:59And in this journey, create differentiation by creating technologies that are a lot more meaningful to us.
00:43:06And it serves two purposes.
00:43:08One is that it is more relevant to us, right?
00:43:10And the second is that, God forbid, if truly we stop getting access to some of those horizontal technologies that were created in a different country, we are still okay.
00:43:20We can make work with, make do with the vertical goals.
00:43:24So, those things, so are deep, and China also, if you think about China, right?
00:43:29I don't think, and DeepSeek is just one example, you know, like there is, there are memes that China, you know, the competition, people in China are not competing with the US.
00:43:39In fact, Beijing is competing with Shanghai and Shanghai is competing with the third province and so on.
00:43:44So, and of course, we know so little about China that all of these are good memes to read.
00:43:49But the point I'm making is that it's the natural evolution of economy, right?
00:43:54We created services because it just made sense to do that.
00:43:58US created products because for them, creating services companies are a lot more difficult.
00:44:02I think we unnecessarily glorify saying that, you know, since US creates product and India creates services, they are better and we are not.
00:44:10I don't think it is true.
00:44:12People will always create whatever is the easiest thing to do at this point of time where they can make money, right?
00:44:20Or they could create value.
00:44:21So, so from that perspective, I think absolutely starting with an application, going, creating small language models, creating an AI orchestration layer to stitch all of the SLMs together and create large language models is a journey that we have to traverse.
00:44:39This will take time, but I don't think with the talent that we have and whatever the activity that I'm seeing, I'm not talking about a 10 year journey.
00:44:48We are probably in a three to five year journey.
00:44:50Three to five year journey for getting to the infrastructure layers.
00:44:54Absolutely.
00:44:54We will have infrastructure, which is at least, you know, like as good as many of these countries are today.
00:45:02Of course, they will also move, right?
00:45:04And we'll have to continue to play catch up if we continue to try to play catch up with them or continue to on the same path that they are creating.
00:45:12So, so I think we have to do some algorithmic innovation as well.
00:45:15So we deep seek if you think about it, you know, open, what is open AI?
00:45:19Open AI, the way a good analogy could be that it is one expert who you can ask any question and the expert will basically understand and will tell you the answer.
00:45:30But it is one expert.
00:45:31And naturally, every time the expert has to learn something, the expert has to effectively consume information, place this in their right graph or right areas of their head.
00:45:43So that, you know, next time somebody asks, the person can tell you that this is the answer.
00:45:47What is deep seek?
00:45:48Deep seek has created multiple experts, right?
00:45:52You know, you could be one expert who knows a lot more about journalism.
00:45:55I know a lot more about venture capital.
00:45:57Somebody else will not know a lot more about health care.
00:46:00We put them all together.
00:46:02We create an AI orchestration layer.
00:46:03And we say that depending on what that question is, you go either to deep seek or you go to Ashish or you go to somebody else.
00:46:12Sometimes you may have to ask both deep seek and Ashish, but you don't have to ask all 10,000 people.
00:46:17All the time.
00:46:18Right.
00:46:18All the time.
00:46:19So there are so and there will be certain cases where you will probably fail.
00:46:24And that's fine.
00:46:24Those are the areas.
00:46:25So the point I'm trying to make.
00:46:26Where you will not be able to recognize probably which is the right expert.
00:46:30And this we have just talked about, let's say, software layer today.
00:46:36Think about what it does to the hardware layer.
00:46:40Now, you could also create custom hardware, right?
00:46:44Which are a lot more conducive for an algorithm like this.
00:46:47Right.
00:46:48You mean to say we can do GPUs in India?
00:46:50We, whether we call it GPUs or not, we will figure it out.
00:46:54AI chips.
00:46:55Absolutely.
00:46:56Absolutely.
00:46:57Absolutely.
00:46:57I think, and I don't think we are, we will, we have an option not to create.
00:47:03We absolutely will have to create compute infrastructure.
00:47:07Right.
00:47:07And GPUs is one component of that compute infrastructure.
00:47:11The point I'm, the bigger point that I'm trying to make, Deepshaker, is that a resource
00:47:15heavy country always will have a belief that resources is not a problem.
00:47:20Think about the scale that, you know, when Sam Altman was basically asked that, how much
00:47:26money would you require to do AGI?
00:47:28I think he's, he's, he's, he talked about some hundreds of billions of dollars.
00:47:31Right.
00:47:32You know, I don't think entrepreneurs in India will be able to think that much.
00:47:37Right.
00:47:37Because that's the, so, and I believe that Sam, Sam Altman was talking about it.
00:47:42You know, he had some maths in his mind, right?
00:47:44It is not that people are, people, especially when they, you know, create enough value for
00:47:50themselves and others, they have some thinking, which will look crazy to a lot of people, but
00:47:56there is a substance about it.
00:47:59Right.
00:47:59So in India, what we have to do is that we, people will always think about a lower number,
00:48:04much, much lower number, right?
00:48:05Because that's how we are trained to think.
00:48:07And we don't have to copy that, right?
00:48:10At some point of time, when we will, at some point of time, we will also be 20 trillion
00:48:13economy.
00:48:14We will be able to afford many of those things, right?
00:48:16But the path to 20 trillion also goes through 5 trillion and 8 trillion and 12 trillion and
00:48:22the likes, right?
00:48:23So we have to be aware of all those limitations and be mindful of that.
00:48:28Yeah.
00:48:29And to ask you a question about this, the path, you said that the path through 220 trillion,
00:48:34the path to 20 trillion goes through 5 trillion, 10.
00:48:37In India, if you look back at the past 30 years, the first wave of tech was IT services.
00:48:43Then came the platform marketplace economy.
00:48:46And now we are looking to get into deep tech, but still a large chunk of techies in India
00:48:52and from India work in these IT services companies.
00:48:56You said IT services is not necessarily a bad thing.
00:48:59A few days back, Anderson Horowitz posted about a company which is probably kind of they are
00:49:08aiming to disrupt the whole BPM, BPO sector.
00:49:13And a lot of other kind of tools like this will come out, which they say will disrupt IT
00:49:20services.
00:49:21And IT services has been kind of responsible for creating the middle class in India to
00:49:26a large extent over the past few decades.
00:49:29How do you see it playing out?
00:49:32Do you think that our IT services companies are at risk?
00:49:37Well, services is not bad.
00:49:39Do you think it's time they have to pull their socks up?
00:49:41I think if you were to create sensation, you basically make some bold claims and then you
00:49:49back it up with data, right?
00:49:50Which is what I think a lot of the conversation, the one that you referred to is also about
00:49:57it.
00:49:58The Anderson Horowitz.
00:49:59Right.
00:50:00Yeah.
00:50:00Right.
00:50:01I think that there is a lot of substance in what the statement actually talks about,
00:50:07right?
00:50:07Anderson Horowitz statement.
00:50:08I do think that the way services or IT services are delivered today will go through
00:50:14massive, massive, massive change, right?
00:50:16And it is in the two areas.
00:50:18One is that, of course, people will say that I want greater efficiency because there are
00:50:23tools and there are technologies to help create that.
00:50:26The second thing that also happens is that people will also say that right now, the way
00:50:31IT services were also created because services were supposed to be used by certain people,
00:50:37right?
00:50:37I think both these are now questions that are software, is software supposed to be used
00:50:46by services or is software to be used by agents, right?
00:50:50Which is the agentic AI kind of claim that people are talking about.
00:50:53I think what is absolutely the IT services, the way it is delivered will go through a massive
00:50:59change.
00:50:59But I don't think necessarily IT services companies, the current company will be relevant.
00:51:06Some of these companies will actually shape up, become better, become stronger, right?
00:51:10And I'll give you an analogy, right?
00:51:12So if you look at, let's say, automobile industry.
00:51:16Automobile industry, there are multiple in the era of ice engine, right?
00:51:21The combustion engines.
00:51:23They had like some 1,700, 1,800, 2,000 odd parts, right?
00:51:27And there were multiple companies which were created only to create those parts.
00:51:32There are OEMs, but there were OEMs.
00:51:34Now you could argue that if parts are there, anybody could become an OEM.
00:51:38But there is a lot of value when it comes to putting all of these parts together, creating
00:51:44a delivery layer on top of that, creating a customer service layer and so on and so forth,
00:51:48right?
00:51:49I think we will see the same thing in the IT services as well.
00:51:53So while there will absolutely be like Anderson Horowitz talks about is that there will be a
00:51:57lot of agents and agentic AI that will come up.
00:52:00Absolutely.
00:52:01But I think they will, and they will also make systems efficient.
00:52:05Some of these companies, the IT services company that we talked about will probably not be able
00:52:09to cope up.
00:52:10They will go down under, but some of these companies will actually become stronger, right?
00:52:14As a country, I have no fear that the IT services will not continue to remain a big employer of
00:52:23the companies, right?
00:52:25I think the way if you look at, but the job functions may change.
00:52:28So right now you have, let's say, a developer and a QA and a business analyst and a technology
00:52:34analyst.
00:52:35A lot of these functions may change and that will happen.
00:52:39But do you think this sector will still be such a large employer and such a large exporter
00:52:45if this disruption happens?
00:52:46If I look at, let's say, one way to also look at this sector is don't look at IT services
00:52:53as different from other digital services.
00:52:56So if you look at, let's say, IT services plus, let's say, BPOs and the KPOs and the
00:53:01legal process outsourcing.
00:53:03The way, the reason these companies existed separately because software was initially created
00:53:08which were used by people, right?
00:53:11So if you look at which is why there was a UI layer and then there's a backend layer and
00:53:15so on and so forth.
00:53:15So that is how the services were created.
00:53:18Now, when the AI world, people are saying that maybe we don't need for meat.
00:53:24If I were to do, let's say, marketing or if I were to do a data entry job for my company,
00:53:30I would, in the pre-AI era, I would hire two people, two companies.
00:53:35One is an IT services to create that workflow system software, right?
00:53:41And the other is to hire people who will use this workflow software and effectively create
00:53:48or enter data and so on and so forth.
00:53:50In the agentic AI era, this could be all combined together, right?
00:53:54So if you now, and there could be just one contract or agreement that could be given to
00:54:01one company saying that you do this end-to-end rather than somebody doing software and somebody
00:54:05doing, let's say, BPO, right?
00:54:07So, and if you know, most changes, Deepshaker, do not happen abruptly.
00:54:13So if you look at, let's say, the India's best IT services company, whether it is Infosys
00:54:18or a TCS or a Wipro and the likes, right?
00:54:22All of them, it is not a coincidence that they have, they provide IT services and BPOs and
00:54:27KPOs as well, right?
00:54:29So, and they have the, they own the customer servicing layer themselves, right?
00:54:33So the buyer may change in the, on the customer side.
00:54:37So they will have to become smarter at it, right?
00:54:39But I think these companies in my mind will continue to thrive, right?
00:54:44Similarly, but at the same time, what would also happen is that there will be, you know,
00:54:49just like let's say when cloud came, some of the large companies continue to remain
00:54:53relevant, but some of the companies actually did not, at the same time, some other companies,
00:54:58mid-tier IT services companies actually became a lot more relevant.
00:55:01We will see the same thing in the AI world as well.
00:55:03So I do think that let's say the top, maybe 10 companies of IT, top largest IT services
00:55:09companies of India, at least eight, if not more will continue to remain very, very relevant
00:55:14is what I think in the AI era.
00:55:16But there will be maybe 30 more AI services companies that will also come up.
00:55:22Some of them will create very specific agentic AIs in horizontal workflows or horizontal processes.
00:55:28Somebody will actually go a lot more deeper into the, let's say, verticals and so on and
00:55:33so forth.
00:55:34And maybe span across the entire profit pool on what the work has to be done rather than
00:55:42basically breaking that by saying that, hey, this is software and this is BPO and this
00:55:49is KPOs and so on and so forth.
00:55:51All of this AI work will be done on AI infrastructure.
00:55:56And the fundamental thing of that physical infrastructure is the chip.
00:56:01And you said just some time ago that we can expect to make an AI chip, build a GPU in the
00:56:08form of a graphics processing unit or any other form.
00:56:11In India, it is said that we have 20% of the world's chip designers, but we don't have
00:56:18any major chip design company from India, even though all the big chip design companies
00:56:22like NVIDIA, Intel, they get their stuff designed here.
00:56:27Are you talking to people who are thinking about designing AI chips?
00:56:32And it is also in the context of in the outgoing Biden administration, just a few days before
00:56:38Biden left the presidency and Trump came in, they kind of notified this.
00:56:47I think it's not, they have not yet notified the change, the directive, but what they, the
00:56:55plan that the US has that the Biden administration kind of revealed is that India
00:57:01India will get only 50,000 GPUs annually.
00:57:05So there are these three tiers of companies that closest allies will have no limits.
00:57:10Second tier of companies like India will, sorry, second tier of countries like India
00:57:15will get 50,000 chips.
00:57:19And then there are these adversarial countries like China and Iran and Russia will get no
00:57:24chips at all.
00:57:25So we are not the closest ally.
00:57:27Probably it seems we are one of the countries.
00:57:30They are probably maybe a bit neutral about, and we are getting 50,000 chips.
00:57:35So is that a concern?
00:57:37Is that an opportunity also?
00:57:38Because that scarcity gets, if it gets created, then we need AI chips in India.
00:57:45Otherwise, how do we scale up AI and do all this work?
00:57:48So I think, first of all, we have to be truly thankful to our government, right?
00:57:54That we are not in, we are actually just a layer below where, you know, the US and we
00:58:01have access to 50,000 GPUs and whatnot, right?
00:58:04I have not looked at the numbers, but it's still not applied.
00:58:09I mean, it's still not has kind of come into practice, but it might down the line.
00:58:15Yeah.
00:58:15Yeah.
00:58:15Yeah.
00:58:16So the point I am making is that, so what it does, Deepshaker, is that it gives us time
00:58:24that, you know, 50,000 GPUs is a particular number.
00:58:29It will probably be able for, we will be able to do certain things and so on and so forth,
00:58:32right?
00:58:32Now, and maybe if we are smart about our algorithms and the way that we spoke about, the way that
00:58:40we architect our entire, let's say, AI infrastructure today, this 50,000 GPUs is probably a lot more
00:58:46in terms of capacity is higher than what somebody uses it for horizontal models, as an example.
00:58:53So that's basically one, so we buy more time, we get more time.
00:58:57What also happens, if you look at founders are also, our founders are very smart, right?
00:59:04So they will start understanding the gaps that today, India is at 50,000 and it is, we are
00:59:10able to do certain things, right?
00:59:11The use cases will start developing at one point in time, we will say that, you know what,
00:59:15now we need X more and some of this will be available through the global companies, but
00:59:21at some point in time, we'll have to leapfrog.
00:59:24And when we start leapfrogging, what would happen is that in this entire compute and
00:59:29GPU is just one part of it, right?
00:59:32There are different companies, which will start playing on different components of that
00:59:37ecosystem and say that, you know what, let me do, let me build a part of it, even if I'm
00:59:41not able to fully build out GPUs and so on and so forth.
00:59:43So I do think that with the technology capability, which is required to build GPUs, it will take
00:59:51longer time for us to do it.
00:59:52But we don't, again, we don't, we have to understand that the founders by the use case
00:59:57will figure certain things out that we are, you and I have not even thought about, right?
01:00:02And I am starting to notice that there are conversations in the US from Indians who are starting to think
01:00:10like this, people who have actually gained some financial independence, either working
01:00:14at NVIDIA and the likes, and they're saying that, you know, now, now I also want to kind
01:00:18of come back and then try to do something.
01:00:20Will everyone do it in a for-profit format?
01:00:23We don't know.
01:00:24So one example I'll talk about, right?
01:00:26So as an example, if you look at the initiative that Nandan runs across multiple, you know,
01:00:34technology stack that he creates using his philanthropic capital.
01:00:37His capability to attract great people from the US who will come to India and effectively
01:00:45work pretty much pro bono is very high because these are some of the people who are very high
01:00:49quality people.
01:00:50They have earned a significant livelihood and they've created a reasonable corpus and they
01:00:56also want to serve India, right?
01:00:59I will, I'm saying that you will start noticing some of that happening as well.
01:01:03At the same time, when some of the, let's say manufacturing, whether it is electronics
01:01:09or the foundry business, and we will have to attempt to get, start getting some of this
01:01:15in India as well, right?
01:01:16And it will happen slowly, perhaps it will happen.
01:01:19We will also start developing a lot more talent there, right?
01:01:22So it's a, so in, when you look at the entire AI infra, right?
01:01:27There is model, there is a, let's say hardware, and then there is energy.
01:01:30We are, I feel a lot more confident about the software layer today, which probably is a
01:01:35three to five year, I think we will be in a reasonably good shape is what I, what I feel.
01:01:40The energy and hardware, we are slightly farther behind, but I think it is a matter of time
01:01:46before we will start seeing activity there.
01:01:48Do you think it is time for India to have VC funds, you know, which are focused on this
01:01:54kind of deep hard tech that we will need maybe 10, 15, 20 years down the line, which, which
01:02:00have that kind of a very long time horizon, maybe a fund like yours, could it have a fund
01:02:08like that?
01:02:10Is that possible?
01:02:10I think in general it is, so there are VC capital may not necessarily always be the best
01:02:16capital to service these high gestation period projects, right?
01:02:22So if you think about, you know, there's a lot of conversation around our VCs investing
01:02:28enough in deep tech companies or not, or do we have the risk appetite or not?
01:02:32My take on this is that I, as a venture investor, have a fiduciary towards my, you know, limited
01:02:41partners or my investors as well.
01:02:43Anything that I don't see exitability, at least in my mind, in a five, six year time period
01:02:50is what I will be very variable, right?
01:02:52We may not like it, but that's the truth that we have to all accept, right?
01:02:56And the way Indian markets are designed or are present today, you will selling only an
01:03:06IP is not very lucrative business today, right?
01:03:10So when it comes, and typically if you look at deep tech companies, it will take them
01:03:13five, seven years to create that basic IP before they start monetizing.
01:03:18So we have to see some semblance of, you know, monetization in that five, seven year time
01:03:23period, which is why it is a lot more difficult, honestly, to do deep tech.
01:03:27In spite of that, there are actually funds which are starting to look at deep tech a lot
01:03:34more closely and they are investing in it.
01:03:38But I'll also tell you that, you know, many of them, when they invest in these deep tech
01:03:42companies also use the same model that you and I spoke about is that, can you start just
01:03:47like in the AI, you have to start with an application layer, look at monetization and
01:03:51then create infrastructure.
01:03:53So people are saying even in the deep tech, while your entire product will probably take
01:03:58seven years.
01:03:58Is there a different way to look at the same problem that instead of doing all, spending
01:04:04all seven years and then thinking of monetization, could you marry some of the business use cases,
01:04:09elements, find a customer and service them, start servicing them maybe in two years, three
01:04:15years time period.
01:04:16And then along with that.
01:04:17It's the Palantir model, right?
01:04:19They first started off as a service company and then now they've become a product company.
01:04:24Correct.
01:04:25It is cutting edge kind of defense tech in the US.
01:04:27Correct.
01:04:28And Palantir has done a lot more for government and defense and so on and so forth.
01:04:31But you could technically do for many other deep tech areas, right?
01:04:34You know, telecom also probably requests a lot more innovation that should be done in
01:04:38India as an example.
01:04:39So I do think we will have to do that.
01:04:44And there will be the way to think about this capital, about this capital is some of this
01:04:49will be done through, let's say the government fund of funds and the government actually has
01:04:53talked about, you know, similar things in this budget as well, right?
01:04:56So I think they have sanctioned certain things for deep tech and space and so on and so forth.
01:05:00So some of this is kind of happening as venture capital pools also become competitive, right?
01:05:08And it has become competitive.
01:05:09Like 10 years back, we probably had maybe 20, 30 funds.
01:05:13Now we have probably 500 funds, maybe more, right?
01:05:16And if you look at more developed economies, there are thousands of funds and we will also
01:05:20get there.
01:05:21People will start to ask for differentiation and then there will be pools of capital who
01:05:26will say that, hey, you know what?
01:05:28I have to, I will only do it in a particular area, which could be deep tech somebody.
01:05:33So I don't think we are too far away from it.
01:05:35For somebody like us to do that may not necessarily fit into our scheme of things today.
01:05:40But yeah, who knows over a period of time, why not?
01:05:42So my last question, so how have you been using AI?
01:05:48And it's not just about which app and that app.
01:05:51Has it been really beneficial for you?
01:05:54Has it been saving you time in your, you are an investor, you must be doing research,
01:05:59you must be going through pitches, evaluating pitches.
01:06:02Has that been helpful?
01:06:03Because I have this one feeling that from my use of it, that it's not actually, I mean,
01:06:12it's like an intern probably, who you always have to kind of keep guiding, right?
01:06:19But is it still kind of, is it still, can it still achieve objectives with any degree
01:06:25of autonomy?
01:06:27That is one.
01:06:28And second is, is it really additive?
01:06:31Yeah.
01:06:32Right.
01:06:32So if you have five interns working on a problem, hard problem, and then maybe they can't solve
01:06:40it, add five more, add five more.
01:06:42You can have hundred interns, but will they be still able to solve a hard problem?
01:06:48You have some great questions, Deepshakhar.
01:06:50You ask two, three questions in one course.
01:06:52I have to remember those questions.
01:06:55So let me try to answer that, but I think two questions that are fairly important, the
01:07:00way that you've asked that, right?
01:07:01One is that, is this an intern, right?
01:07:03And the second is that, is this additive or not, right?
01:07:06And I think second question particularly is an outstanding question.
01:07:12So I'll come to that, but let me first maybe tackle the easier one, right?
01:07:16I think one of the challenge of being user of a very fast moving technology is that if
01:07:26you keep your, if you, you know, don't look at it very closely, even for a month, you will
01:07:33probably become outdated.
01:07:34So the point I'm making is that AI stopped being intern only maybe six months back.
01:07:43The way my framework is that, so see in any work, what is that we do, right?
01:07:47Not only venture capital, there is data that we collect, right?
01:07:52And then there is analysis on insights that we extract from this data or from our experiences,
01:08:00right?
01:08:01So insights are based on my experience and the, any fresh data that has come up, right?
01:08:05That's the second one.
01:08:06And then based on these insights, you take a decision.
01:08:08Very interns usually are useful for the first layer or the bottom most layer, right?
01:08:15They will collect data because you'll say that, Hey, you go to this field and collect this
01:08:18information and this and that, right?
01:08:20Or a lot of times interns are, don't even get primary data.
01:08:23They actually get secondary data.
01:08:25If you look at the current models that all of these, the large language model guys have
01:08:31come up, I think their ability to come up with insights based on both all the past data
01:08:38that is available, because of course they have all the data, right?
01:08:41Do a deep research on it.
01:08:43Come up with multiple ways that you could take a decision table it in the form of saying that,
01:08:48Hey, these are the three potential decisions.
01:08:50And these are the things to watch out for is there today.
01:08:53If you have, it requires a little smarter prompting because technology is also improving,
01:09:00right?
01:09:00So you also need to spend a little bit of effort.
01:09:03And I think all of this will become easy over a period of time.
01:09:06So I think AI today is at a place where not only does it collect data, right?
01:09:12Or it gets data from the current layer of data.
01:09:17And if you also provide new data, it is also able to do analysis based on that.
01:09:21And then also tables three different decisions that you could take.
01:09:25So I think, you know, not only does it solve a job of an intern in many, many, many, many
01:09:32cases, it actually has started to solve problems of that typically a mid-tier manager also actually
01:09:40does it, right?
01:09:41You will, since the technology is so new that we don't fully have confidence that this data
01:09:48correct, this analysis could be right.
01:09:50But I think this will all keep on improving, right?
01:09:52Very, very rapidly, right?
01:09:54So I'll tell you, I think my thinking is that it is a matter of honestly, very short period
01:10:01of time where I think all the things that we do based on experiential data, AI will be
01:10:07able to do bulk of this, you know, in as little as maybe a year or so, right?
01:10:13Which is what I think when you asked a lot of these Sam Altmans and the other people running
01:10:18these models, they are actually talking, referring to some of them, some of them are referring
01:10:22to as the AGI moment and whatnot, right?
01:10:24I think we are not, genuinely, I don't think we're too far away from it.
01:10:27What will be important is that this will become a commodity.
01:10:32If, you know, if OpenAI or the Claude or the, you know, the anthropics of the world actually
01:10:38provide you this data, you will have the same information as I.
01:10:41So your experience and my experience is not as valuable based on what you have done.
01:10:47So how do you differentiate?
01:10:49And the way to differentiate is that you will have to go out in the field again.
01:10:53I will have to go out in the field again.
01:10:54And, you know, typically mid managers and top managers have stopped doing it.
01:10:58Many of them have stopped doing it.
01:11:00So they will have to start doing it, collect fresh data.
01:11:03That is where the competitive edge, I think, will start coming from.
01:11:06So I think, yes, I think it is a lot more than an intern.
01:11:13So that's one net additive.
01:11:17I think that's the reason it is also a great question because the way current algorithms
01:11:23work today is that they will use existing data and based on that, make a decision based
01:11:31on that.
01:11:31We don't know enough whether it is possible for them to extrapolate and create and anticipate
01:11:40new data and then make a decision based on that.
01:11:43So once that starts happening, it will be net additive to the entire information that
01:11:48are available to you or me.
01:11:50It could, since we don't have access to all the information in our head, it will, I think
01:11:53it is already net additive.
01:11:54But to the entire world, it will start becoming net additive at that point of time.
01:12:00And I think it is going to be a very fascinating world when that happens.
01:12:04And they probably require a lot more algorithmic innovation than we know of, certainly I know
01:12:11of today.
01:12:11Hope that algorithmic innovation comes from India and hope you are an investor in the company
01:12:20or the project that does it.
01:12:21And on that note, thank you so much, Ashish, for coming here and enlightening us on how
01:12:29India should play the AI game.
01:12:31Oh, thank you so much.
01:12:32I think, you know, if there are entrepreneurs listening to this, you know, I'm very happy
01:12:38to kind of brainstorm with people.
01:12:41I have a technology background.
01:12:43I still consider myself as a technologist before I consider myself as an investor, right?
01:12:48And it is a topic very close to my heart, not only because it is good for business, but
01:12:53also because I think it's good for the nation as well.
01:12:56And we have started taking, I would say, not only baby steps, but, you know, some, we have
01:13:05started to make progress.
01:13:07We have to, of course, accelerate this.
01:13:09And the way it would happen is that when we think of this as an ecosystem, so a lot
01:13:13more people will have to do this and not only in isolation.
01:13:17With that, thank you so much.
01:13:18I truly enjoyed this conversation.
01:13:20Thank you so much, Ashish.