India's AI Opportunity
Krishna Mehra and Poorvi Vijay
India's AI Opportunity
Krishna Mehra and Poorvi Vijay
Episode Description
Unlike cloud adoption, which took as many as ten years, enterprises are rapidly deploying AI - even traditionally conservative sectors like financial services and healthcare are moving from pilots to production at unprecedented speed.
And India’s AI moment is here.
This time, we’re not playing catch-up. In this episode, Krishna Mehra (AI Partner) and Poorvi Vijay (Vice President) join Vishy V (Head of Marketing) to unpack Elevation's thesis on the massive AI opportunity for Indian founders. The conversation explores three key themes where we see immense potential:
1. Infrastructure & Middleware: Building the essential tools that connect models to enterprise data and enable AI adoption at scale
2. Vertical AI: Reimagining entire industry workflows by combining domain expertise with AI capabilities
3. Services + Product Play: A unique India advantage in delivering end-to-end AI transformation for enterprises globally
From commoditization of base models creating new opportunities and India emerging as a major AI market to the rapid enterprise adoption - Krishna and Poorvi break down the key trends shaping this space and share their advice on how founders looking to build in AI can stand out.
Transcript
Viswanathan V (Head of Marketing, Elevation Capital)
Hello and welcome to another episode of SummitUp by Elevation Capital. Today's episode is a special one, because we dive deep into the world of AI and to take us and talk us through it, are Poorvi and Krishna from the AI team here at Elevation. Welcome to the pod, guys.
Krishna Mehra (AI Partner, Elevation Capital) & Poorvi Vijay (Vice President, Elevation Capital)
Thanks, Vishy. So excited to be here.
Viswanathan V
Let's get started. And we can't have a podcast on AI without first starting with some of the hot takes on what is current and trending. And the most trending thing in AI at this point in time seems to be this Chinese upstart called DeepSeek. What's our take on that? And how do we see this impacting the broader AI space?
Poorvi Vijay
Yeah, so see, before I get into what DeepSeek is doing and why, but fundamentally, in the last two, three years, we have seen the cost of using generative AI and the commoditization of large language models getting there, right, like the cost has actually come down. If I take just the training cost, inference costs actually have come down by more than 5x and then if I take prominent examples like OpenAI's GPT 3 to 3.5 it's actually been more than 30x. Training costs have come down by 30% to 50% by similar sized models, and now coming to DeepSeek, because the world has risen to it suddenly, see what they did, fundamentally, was that, hey, a similar type of reasoning model, comparable to OpenAI's benchmark, they were able to build it, with just 2-3% of compute and inference cost of almost 30x reduction. And that's huge, right? And it was done as a side project is what we hear with a with a team of five people, and that suddenly has given rise to a question that, hey, how much compute and infrastructure do you fundamentally need to train a model. So all of this means one thing, one is that even today, we don't have a winner take all model market. Models are still evolving. People are experimenting. There are optimizations happening. The second part there is that with all these models getting more commoditized and getting cheaper, there is definitely a pressure, downward pressure on the commercial models, because people have started using open source models so much more. And the third part, which is the most exciting part, is because costs are coming down, people will start using a lot more AI. Companies that were not able to use AI because of the cost earlier now can start adopting AI and leapfrog that. So from a model perspective, I do feel that differentiation will come across two axes for newer models to jump in and who will win. I think the first one is just more specialized vertical models, providers who can actually fine tune, pack the models in specific verticals, like, let's say, healthcare, legal, medicine, etc, because then you can amp the pricing basis the domain expertise that you provide, the compliance certifications that you provide. So even if your base model gets commoditized. There is specialization that you're providing on top of it. The second differentiator that I think will happen is across latency, robustness, scalability of models. Because even again, if the base model actually gets commoditized, when you make something very, very scalable, reliable, uptime, SLAs are there, people will pay up even for that. So that's how I feel this entire model layer is going to get transformed, and we're still in early days, like while there have been so many models that have come up, we're still in early days. And I do believe there's going to be be more and more innovation.
Krishna Mehra
Exactly. I think, if you look at it, last year, everybody focused on the generative aspects, and the cost went down, but suddenly now everybody's talking about reasoning and test time and inference time compute. And you know, there, the scaling is very different. So the next year or so, you might see a lot of innovation there. But just like a child learns different skills at different age groups, these models are also learning different skills. So I'm guessing in 2026 will be a different paradigm altogether. I think, to Poorvi's point, what's very critical is, when people are building on top of it, I think they have to really pack their proprietary data and, you know, complex integrations to differentiate themselves. You don't want to be hit in the way of commoditization.
Viswanathan V
Very interesting. I think the big takeaway is cost of compute goes down. It gets more democratized. I think for me, the reflection is I need to figure out how to get my cost of AI down because the number of tools I am using is quite a bit. I think another question in general, when it comes to AI that seems to be on most people's minds is the ability for startups and new entrepreneurs to disrupt, right? Because there are large incumbents, whether it is at the model layer that we spoke or even at the application layer. And I think there is still a lot of skepticism around, how are new startups gonna come and disrupt? What is our take on that?
Krishna Mehra
I mean, if you look at it like, unlike the last wave where incumbents were found sleeping on the wheel or they had an architectural lock in and they couldn't move to cloud quickly enough. I think this time, incumbents have read the same memos and they are playing around with the same APIs. So if you are building a thin AI layer on top of an existing workflow or application, I think it is incumbents who will clearly win. Where I think startups have a real opportunity is being more crazy and more wonky. And I think you have to really play out that Innovators Dilemma do stuff that innovators can't copy easily, like it can be around pricing models, outcome based pricing is something everybody's talking about. Add a lot more multi modality. So I think the more you kind of do those things, layer in more services, human in the loop, I think incumbents will find it hard to change their business model from those. I think that's what people need to do, and that will, that is what will help differentiate and give a 10x higher ROI to businesses, to their customers. So, just as an example, if you look at Bret Taylor's Sierra, they started building customer service agents, and obviously, with their pedigree, I think they can land any customer they want in the world. But it's very interesting that so early in their life, they've actually switched to outcome based pricing, and they're really trying to differentiate themselves from existing customer service options that are there in the market. So I think it's interesting to see that even such strong pedigree founders are actually going down that path, and that's what's showing like, how we have to be different as we think of starting new companies and new products that we can kind of sell to businesses.
Poorvi Vijay
Yeah. So see, as a fund also, like, while there's so much innovation happening, we've been focused on three primary areas. We look at everything that's interesting, but from a thesis perspective, these are the three areas that we really are interested about. One is the infrastructure and the middleware. The second is this entire swarm of vertical AI agents and applications. And the third is, very interestingly, service plus product plays coming out of India. So, if I take the first example, and we were talking about this commoditization of large language models, right, while we do feel that the large language models are getting commoditized, what's critical and not getting commoditized is, how do you actually connect models to your internal data. How do you use data in a very compliant and AI first way with a lot of governance? And then use all of this to build agents, applications that can actually make the experience for your customers very well. And that is not easy, and that is where a lot of innovation we feel can happen. And we've been very active, right? Like in the last 18 to 24 months, we've made a lot of investment in this area. So one of the companies is UnifyApps, which fundamentally does this. How do you break those data silos in large enterprises, take that data, connect them to the models, then build applications and agents on top of it for the different use cases that an internal team might have. The second one is a company called as Composio, which essentially, again, is doing something very interesting to say that, hey, for anybody who's trying to build agents, any developer out there trying to agentify their ecosystem, they need a bunch of tools to connect with. You can't build a powerful agent if you're not connected to at least 200 applications. And that's where Composio comes in as a middleware to say that, hey, no matter what use case you have, I have these pre built 200 plus integrations. Auth of these agents is a very important aspect of it. I do the auth, I do the integrations, and then you can build agent on my platform, right? The other one is Maxim AI, which fundamentally says that, hey, while you're building things in production, whether it's an application, agent or something else for internal team, you need to test and evaluate it. Because nothing can go in production and unless you run it through benchmarks, you do human evaluations, you do other evaluation, and that's where Maxim AI comes in. So that's why we've been very excited about this space. But having said that, we've always said that, hey, obviously, application market is very, very big. The way internally we think about it is that infrastructure and applications always go in tandem. It's a sinusoidal wave. Because there was no infrastructure, applications were not getting made. So there was a lot of investment in infrastructure companies to build that tooling to build applications. Then people start identifying use cases on what applications can be built. Once application ecosystem grows, you start realizing that, hey, for scalability, for robustness, you need more infrastructure on different things, because you start finding faults in what's existing. You build more infrastructure again, and then once that gets solved, you start building applications again. It's a sinusoidal wave that keeps on going. So that, as a thesis, I think, is very, very exciting for us.
Krishna Mehra
I think one more interesting thing that I'll add is for both infrastructure and middleware, earlier, the market for all these things was only, you know, North America, for instance. But I think we are actually starting to see more and more of India being a market for many of those things. What's incredible is the number of companies, even in the valley, where we've seen pitches which have customers, which are focused based in India. So I think India is a real market which is also emerging in this that helps founders and companies to actually learn faster, iterate faster, work at very massive scale. Indian you know, consumer applications have, like, hundreds and hundreds of millions of users. So I think you can kind of test at scale, learn very quickly, and all of that, which is a huge advantage for founders in this domain.
Krishna Mehra
I think that's a big departure from the previous waves. You had to struggle to go to the US to find your large customers. Excellent. Okay, so you spoke about three opportunity areas, I think we spoke a little bit and go deep on the infrastructure and middleware layer. Vertical AI agents, that seems to be the latest buzzword now.
Krishna Mehra
Yes, so I mean vertical AI obviously, is a huge opportunity. I think the one, one way to look at it is not just the software, but also the human labor pool that is going in. And the ROI from that is not just cost. Obviously, cost is a huge advantage. You can automate a lot more, but you can also get a much faster turnaround time on, you know, complex tasks where you know it would take a lot longer before So, for instance, you know, we've invested in a company in the fraud and risk space which does investigations on, you know, cases that come up, and large banks and financial institutions have 1000s and 1000s of people doing that manually today. So there's a real opportunity to automate a lot of that work. Then similarly, if you look at, there is so much revenue cycle management work for healthcare that happens in India today, and the domain expertise in that space is actually largely sitting in India. So there's a lot of companies which have come in that space which automate a lot of that work. And, you know, kind of mix this whole human plus labor aspects of it. Again, globally, there are a bunch of very interesting examples, like, we talked about Bret Taylor's Sierra, but also Hippocratic, which is building a AI nurse to doing a lot of the non clinical work, because you can automate a lot more of those things. So definitely, on the vertical side, I think the opportunity to go beyond just software is incredible. The one thing which is important to remember, though, is, you know, while the technical capability is new, I think the business equation, the business case that the customer will look through, I don't think that has changed. So they will still look at, you know, hey, what is the ROI that I'm getting, getting out of this solution. So in 2024 I think ROI was out of the window, yeah, because everybody was experimenting with AI. But I think it is going to be very much in the conversation, especially the moment you go into vertical solutions. I think people have to really demonstrate very high ROI as they build these new capabilities.
Viswanathan V
Very interesting. So it seems like that $600 billion question will get answered.
Poorvi Vijay
Finally, I think the only other part that I'll add there is that historically, in vertical spaces, everyone has been very selective. And hey, is the TAM enough? Can it grow? Can it not grow? And that's always a question when you invest in a company at an early stage, right? And that, I think, has had a real unlock because of what Krishna was saying, that you're not attacking the software spend only, but you're actually attacking the labor spend, which just doubles, triples the TAM in any vertical sectors that we can think about.
Viswanathan V
The third space you spoke about, which is very interesting for us as a fund, is the opportunity in services, right? And that seems at the surface very intriguing to me, because while India has built its success on the service model, AI was supposed to disrupt this, right? So how do we see this becoming an opportunity for India?
Poorvi Vijay
So see, the way I think about it is not just services, but a product plus service play, and I'll explain what it truly means, at least for us. See again, India has had a huge legacy, and that's where our leverage has been, that we've implemented, delivered services at scale. We know how to do it today, if you think about and speak to any enterprise customer, right? Everyone wants to do something in AI. They have some idea what ROI can be. How do they want to do it? But they're very confused, because there are hundreds of point solutions, and as a big enterprise CIO I don't want to keep using different solutions, because A, I don't have the know how of it. And B, it's very difficult to actually integrate a bunch of things. So when a company comes in that, hey, you know, you don't need to worry about which database you use, you don't need to worry about which pipelines you use. You don't have to worry about how do you clean that data, etc, right? I come in, I do everything and deliver an outcome to you. That is a very interesting proposition for an enterprise. So I think from a margins perspective, yes, it will not be a software margin but it will also not be a service margin, because internally, you are creating a lot of efficiencies with AI. So it will be somewhere in the middle. But any company serving to enterprise, does the implementation end to end and delivers outcomes, I think could be in a very interesting place, and from India, I do feel we could do it much better than anyone else.
Krishna Mehra
I mean, I have managed... I've been lucky to see Innovaccer kind of play this story out very closely. So if you look at it, they took Palantir's forward deployed model, and really took it to healthcare. And when you look at in the beginning, their business looked a lot more services heavy, because they took on all this extra work to generate ROI for their customers, but over time, they managed to chip away and actually have now gotten to software margins. So I think when you look at that, I think there's a lot more opportunity to replicate that playbook more and more given this tailwind of AI. Also, it's changing so rapidly, it's hard to just build a pure software solution. And I think it's a real differentiator, plus a huge advantage that Indian founders have. And I think when the world, when you start a new company, the world is almost stacked up against you, so you should take every advantage you can get. So I think it's a logical thing for ambitious founders to do.
Poorvi Vijay
You know, we've all seen how Palantir stock skyrocketed, so it does make sense.
Viswanathan V
One more thing that seems to be skyrocketing at this point in time is the interest in just general compute infrastructure, right? Whether it is the debate around where, who has those $500 billion and Satya said that he has $80 billion. Stargate or Jio's plans to build the largest data center in Jamnagar, it seems like there's disproportionate focus going on just the compute infrastructure part of things. How do we see that playing out, and what is the relevance of that for, say, the Indian ecosystem?
Krishna Mehra
I mean, traditionally, if you look at this space and you open up AWS' products page, there are probably, like 250 products on that page. So in a typical data center, you need to manage so many different kind of workloads. But when you look at AI workloads, they're fairly simple, there's training and there's inference, and you're running a model and that, but that workload is so much larger, and in terms of price and cost, you can actually run a full data center on that. So I think that is the one thing which is also a big tailwind for this, because given this, you know, new kind of workload, there's real opportunity to unbundle, you know, the hyperscalers, and that window didn't exist before, which is where I think this has, this piece has become so hot and interesting. Of course, the capital requirement and how to do it, build versus replace. There are lots of questions to answer, and I think there's nuances to based on location, based on power ability, based on water infrastructure, cooling, like all those things have to be taken into into account as you build it, including, you know, sovereign considerations. But I think this is definitely a very active debate.
Viswanathan V
Interesting. One more thing, I think you touched upon it a little bit, Krishna, when you spoke about some of the large Indian tech companies being the customers for these AI companies. You know, a year ago, we had gone and we had surveyed a bunch of founders, and at that point in time, AI adoption was fairly low. Around 10% had something in production. Very marginal number had something even in terms of experimentation, my sense is that numbers, those numbers, would have changed materially. And I think you spent the last few months having a lot of conversations with founders. How do you see AI adoption having evolved in the last year?
Krishna Mehra
If we just look at AI by itself, you know, we'll get impatient and say, hey, why is it not moving faster? But I think the best way is to compare it to the cloud wave. In the cloud wave, you know, a lot of people took 7, 8, 10 years to start adopting cloud, especially larger enterprises. But that's not happening in AI. There's a lot of top down pressure directly from the board. And, you know, CIOs and CEOs are being asked, 'What are you doing with AI?', that's an active board conversation. So, like you said, last year, there was a lot of pilots and POCs because I think these decision makers also didn't know how to use AI. So a lot of pilots and POCs and lot of horizontal solutions that got bought OpenAI Enterprise, Glean, etc, because those are very simple unlocks when you look at it. But I think now, as people are starting to understand, there's a lot more of this vertical AI adoption that is starting to happen. What's fascinating is we've seen adoption of AI in even large banks in India, which, you know, if I think back from the cloud wave, there was no way you could have pitched to a bank in 2012 or 2013 I think you could only do it by 2018 so it's definitely moving a lot faster. Of course, we all want it to go even, even faster, but I think this wave is definitely moving very, very quickly.
Poorvi Vijay
I think the other anecdote that I can add here is that obviously we keep seeing a lot of AI native companies right in our own portfolio. There's where we invest. But if we look at our overall consumer portfolio also, which is very huge, people are actually making insane progress. Like we met NoBroker recently, and they have launched a complete customer support and customer service platform, which is completely AI-first. Actually other companies have started using their platform, you know, because the ROI is very high, Meesho has actually open sourced their AI deployment platform because they have seen insane amount of, you know, excitement there, usage there from different open source folks. So I think it's not just the AI native companies, but even in the broader ecosystem, people have started experimenting and adopting AI, and we have obvious proof points of that.
Krishna Mehra
Yeah, everybody is talking about, hey, what are best practices around AI usage? And you know, we've been pinged up from so many founders like, can you give us ideas on where we can adopt AI more? So I think it is a very active, intentional discussion, versus, like previous waves, where, hey, it's all working. So let's figure out and watch, wait and watch. That's not happening now. People are leaning in.
Viswanathan V
Super interesting. I think this has been a very interesting conversation. I have learned a lot, and I'm sure anybody who's been watching will find some things that would have, you know, been thought provoking. Just as we close this out, what would be your advice to young builders and founders out there trying to disrupt and build something in AI from India?
Poorvi Vijay
Yeah, see, I'll take a very India first lens to it. I think it is very, very exciting, like as an investor, I'm so excited. And I think we are in a very exciting time, and see, because in SaaS, we were actually lagging behind, right? Like, if we think about just the first IPO in the US for SaaS, was Salesforce around 2004 and for us, it was Freshworks in 2021 so it's literally two decades, like almost two decades, right? But in AI, that's not happening. We are seeing companies getting started almost at the same time, as compared to our global counterparts. So we are starting at the same time, so we're not lagging behind, and that's very interesting. We have all the talent. We're obviously moving fast. We have been super agile. We've been super cost efficient also in certain ways. I think the only thing that we need to think about is that we shouldn't think about us as an alternative to x in the AI game. We should aspire to be actually category leaders, build solutions, attack problem statements, which are very native. Nobody has thought about it. And I think there is a clear right to win. Also historically, we didn't have those playbooks, right? The corridor. And I think Krishna can allude to that so much more because he started a company in that time. I think now we have that corridor, the India-US corridor, the India global corridor, for that matter. It's really opening up, so many companies that we meet, are also partnering and piloting in India. US companies who are trying to pilot and partner with Indian companies. So that corridor has opened up. So the Indian companies also should aspire to be that global leaders, and not just think about as either a cheaper alternative or a second or a third alternative to an X company. So I think that is definitely an advice to most founders that just go big, be ambitious, and I'm pretty sure that we'll see a lot of wins in this case.
Krishna Mehra
I am so amazed, because when I think back at when, you know, I started Capillary almost like 15, 17 years ago now, there were no playbooks. We had no idea how to sell in international markets. There was no SDR, BDR, sales enablement, like all these terms, didn't exist, especially for founders from who were building from India. There were no resources. We could not find any GTM talent. We actually pretty much trained all our GTM talent from scratch, or we hired them from services companies. So much has changed over the last like decade and a half. I think founders now have talent in almost every function that you need. People are selling large enterprise deals sitting out of India. There is so many more resources. There's so much more capital available. There are playbooks, and there are communities like, you know, serving founders and helping educate them, etc, as well. So I think the constraint is only the level of ambition. And I would say, I think that's what founders need to index on. I think we have the opportunity to build category leading companies. We should not be focused on building, hey, you know, I want to build a small company and make it successful. I think we can build really large companies, and that is what is very exciting. I think there's real opportunity for massive value creation in this wave.
Viswanathan V
I think this has been a superb conversation. Thank you so much, guys. If you are an early stage builder thinking of building in AI or already building in AI, do reach out to our team over here. You can reach out to Poorvi, Krishna or the broader AI team here at Elevation. You would have got a sense of where do we think there is a large opportunity, and also a sense of how excited we think we are and the overall ecosystem is in terms of the opportunity to build out of India. So if you're a builder, now's the time to build and reach out to us if you need any help. Thank you.
Written by Krishna Mehra, Poorvi Vijay
Related

Harnessing Better Experience And Innovative GTM: Marketplaces Unleashed Part 3
Exploring the last two pillars of our marketplaces framework through case studies
03.08.2023

Vridhi: Reimagining Home Lending For Bharat's Self-Employed
Ram Naresh Sunku, Co-founder, Vridhi Home Finance
11.12.2024

Investing in Atlys
Building the world’s largest digital visa provider
21.09.2023