HCL invests in Sarvam and goes down the 5-layer AI cake

Hi, I'm Vithaatri, the producer of ZeroShort. I have a quick note before we get into this episode. There are a few minor audio glitches in this conversation that we face during recording. We've tried to clean them up as much as possible during post-production, but you might still hear slight external disturbances in a few places. We sincerely apologize for this. Thank you for bearing with us now onto the episode. Hi, welcome back to ZeroShort. We're back with a new episode. It's not we is actually me. This is Rohan. I'm the sole host today. I'm not joined by either Brady, Orch, Praveen, both of who were busy with other things. So I'm donning the mantle today as the sole host, but I'll make that up with a really interesting episode. And that episode literally traces back to just two weeks ago. When HCL Technologies, one of India's largest IT services firms ended up being the lead investor in server mayi's $300 million round. That happened if I'm not mistaken on May 14th. We're recording this on May 25th. HCL Tech committed $150 million, which was 50% of server mayi's total round size of $300 million. The round was at a valuation of $1.5 billion post money. Now, this is the largest pure play Indian AI startup fundraise to date. HCL was joined by other investors like Bessimer's US venture growth desk, which put in $50 million. And then a bunch of other very interesting firms, including Nvidia, Gladebroke, and of course a bunch of existing server bankers like Khozla, Lightspeed, P-15, etc. Here's an unusual thing. The round for its size and scale was not led by a venture capitalist. Instead, it was led by a strategic, that's what they call corporates that invest as we see a strategic. So it was HCL Tech and HCL Tech invested at a 7X markup on Sarvam's last fundraise, which will be back in December 2023. If I'm not mistaken, Sarvam's revenue for FI-25 was in the region of 29 crores. So just incredible growth in valuation for Sarvam from December 2023 to now $1.5 billion now. No VC instead of strategic. So today in this episode of ZeroShort, we want to decode what exactly is HCL buying and what exactly is Sarvam selling. And for that, I have to create guests. My first guest is Abhishek Patak, who is the lead analyst, IT services and internet at Motilal Ossval Financial Services. He's a co-author of the report, Indian IT Services, assessing the narrative shock, which actually talks about the narrative shock from AI. He's also co-authored other reports, which talk about the impact of Anthropic, etc. on Indian IT services. I went through Abhishek's reports and the broad gist of what he's saying and Abhishek, please correct me if I'm wrong, is that roughly 30 to 40% of Indian IT service revenue comes from application development maintenance and testing. And that is at risk from AI. He talks about productivity gains from AI, which could eliminate 9 to 12% of total IT services revenue the next three to four years. But his point is that AI is an execution risk, not an extinction risk. And he feels that the near term narrative shock is real, but it's probably overstated. On HCL tech specifically, I think is one of his topics. Again, he should confirm that. He discrepancies all weather struck because of its balance portfolio. Welcome to ZeroShard Abhishek. You're joining us from Mumbai, right? That is right. Thank you so much for having me and looking forward to the conversation. Did I get any of the details wrong about your broad outlook on what AI does to Indian companies in IT services? Yeah, no, I think the Anthropic narrative shock report was it in probably a couple of months back and honestly the rate of change right now is is pretty magnifying. And what we've incrementally learned since writing that report is that quantifying the deflatable book in IT services is currently fought with error. And that's because with each improving model with each improving plugin, with each improving, you know, sort of, capability said that a new model brings in more and more of the services industry seems to be under the habit of the deflatable book. So we're kind of moving away from that calculation and we're kind of focusing on, you know, where the industry can kind of, you know, find itself in the next, let's say, three to four or five years. So you might not, that's the only caveat yet. I'd love to know what you think specifically about HCL tech investment, which brings me to my second guest. My second guest is Kashiap Kompela, who's the founder and CEO of RPA2 AI research. It's an industry analyst firm founded in 2019, headquartered in Bangalore with network in Bangalore, Boston, Washington, DC and Delhi. The firm serves global corporations, BC's, P firms and government agencies on AI investments and a bunch of AI strategy, most of its clients are outside India. Kashiap himself has spent over 20 years across the sector as an industry analyst and in various M&A advisory roles. He's a CFA. He's a lermaness of Bitspilani, ISB, Hyderabad. He's a master of business law plus PG diploma from NLS and he teaches as Bits. Kashiap, welcome to zero shot and that's way to 20 degrees, man. I know, and it's awkward at some point. But what you didn't mention is that I have been a subscriber of Ken since its inception. Thank you. And I don't have the time to read every story, but I read, I binge read. Many people binge watch Netflix. I binge read Ken once in a while. So much so that your bot trigger is triggered and I'm locked up. Oh, okay. Thank you for that. I'm sorry about that. This is the first time I'm in this podcast, but I teach at a few law schools. And you had a podcast with Trilegals and Killner Indran from 10 back. So that was the prescribed reading for my students, for one of the modules. I'm so happy to see all your lines being crossed between subscribers and guests and colleges. Thank you so much. Kashiap, look me forward to it. Yes, and I'm a bit while on colleges like I was looking forward to previewing because I'm a big fan of his humor. I'm insulted that you're not a fan of my humor, which actually is non-existent, but still. But let's come back to today's episode. I loved having the two of you together because I think it's a very interesting mix. On one side, we have a sales side equity research analyst, who's really bullish on HCL. And on the other side, we have an independent AI analyst and advisor whose natural tendency will be to look at the big picture and risk, etc. So therefore, it's a nice framing of, you know, I mean, the issue of HCL and server from two sides. Before we begin, quick note, we reached out to both HCL tech and server for their participation in this episode. Server declined. HCL tech did not respond. Therefore, our conversation today isn't some census and outside in reading off the deal by two professionals who watch this terrain professionally. But we have no absolute insider visibility into the deal mechanics. So therefore, if you're a listener, please treat what follows accordingly. Okay, let's dive in. Abhi Shake. Let me start with you. Before we get into the specifics, can you give me a simple one paragraph reading of what exactly it's HCL doing here with this massive foreign Indian IT services company, $150 million investment into server? And why is this the right move for them? Is this the right move for them? Yeah, no, I think the old IT services model as we know it, right, where you had this labor arbitrage game that you paid for the last three decades where you've got an engineer sitting out of Bangalore, Mysore, Vishakapattana, Mad 1.6, the cost of an engineer in the US, that's certainly breaking down, right? And at a given point in time, the cost of tokens will definitely be converging with the cost of the cheapest engineer out there. So AI has found its perfect product market fit and that's unfortunately software engineering, right? So a lot of the building that's happened, not the building that happens in IT services firms such as HCL has been time in material, led coding revenues, right? So I lower X amount of engineering, for X-Summer of Oz and I kind of, you know, build you for that. And so, so there needs to be kind of almost a feverish urgency to disrupt your own book of business, to kind of, you know, to kind of keep looking for answers as to how does, let's say, you know, the IT services form of the next five years look like. And to that effect, I think it's here in my opinion has kind of gone ahead or has possibly, I don't know if the news is confirmed, but they've kind of gone ahead and let this down in service. And I think it's difficult to really bend down right now the exact number or the exact dollar amount that servom may contribute, let's say two, three years out. But I think it's a more non-linear bet around what a sovereign LLM company like servom can do to HCL. I mean, if you look at, you know, five, six years out, there's a couple of sort of, you know, a key, a key area to kind of look at the first is, you know, I'm going to come back to you for a deeper dive into the deal mechanics. But before that, let me just ask Kashyap also that same question. Kashyap, quite simply, how do you see this deal from HCL, HCL, X point of view? No, I think I've covered IT services as an analyst before I started covering the AI space. And prior to that, I covered the IT product space, evaluated IT product space. So this question, so we always had this question 10 years ago, 15 years ago, why are in services companies increasing or building their own products, etc. So that's a larger discussion. And the question that now we have the backdrop to this deal is why are in IT services companies building foundation models themselves? So the answer then when we're talking about services was products, plus the services is different, the products are different. It creates a lot of organizational clashes as well, like you need to pay product, guys, more than you pay the services for. And that retention is so much more in AI right now because AI talent is so much more expensive. So I'm broadly positive about cyber me. And I noticed like since my tracking of the space likes 20 years ago to now among the large Indian IT services providers, HCL and Cog is end. I've proven themselves to be different to do different Apollo different playbook compared to the day it appears. So I think this is a continuation of I think HCL trying to do differently and it makes sense. We're not getting into the valuations, etc, etc. So if you can't build the model yourself, it makes sense to invest to be the anchor investor in a deal like this. But one question I have is from whatever little I see about a server for example, server strength and capabilities. They're positioning themselves as a regional like if you see their benchmark of their latest models, they benchmark against us. They are comparing us to other regional front-aid models. They're not openly comparing themselves with say the open AI's or the anthropics. So for an IT services company, who's business is so much outside of India while service focuses more on sober and India and India based applications. They have fantastic technology when it comes to Indian Indian languages, a speech model, etc. So how does that squareness something that that we need to unpack and understand a little more. If an Indian IT services company was investing in a front-aid model company that is focused on the western clients, it makes a lot of sense. So the only sort of a work I see in this largely positive deal is what does HCL take get out of this. Kashyap, I think it's interesting that you called it if I'm not mistaken a foundational model, I think between you and Abhishek, one of you called it foundational model Abhishek called it sovereign LLM. I think only one of them is right. And I think and the reason I sort of say that is that yes, we know that Sarvam is a foundational model, but I think one of the parameters or checks for me to actually determine, I mean, how a foundational model fairs is to look at some kind of global benchmarks on where a foundational model fairs. And Sarvam frankly is invisible there. It's just not there on most global benchmarks. Now that's fine because Sarvam strategy seems to be to essentially target India and all the various Indian languages and Indian data sets, etc and stuff like that, which is actually there for closer to that. And if you look at Sarvam's makeover and I think the makeover was pretty drastic in the last 12 months. If you remember the Sarvam website before that to at some point it just suddenly became every third word was sovereign. And literally it was that they use an LLM to insert the word sovereign every third word. And now it's like you go to the Sarvam website and it's we are sovereign, we are sovereign sovereign is best. So I think I'm sorry, I do feel your question is valid because it's here. So the real tech seems to have invested. Let's forget the valuation for now in the absolute amount into a fundamentally sovereign. And that's a great point for me to turn it around to Abhishek and ask why Abhishek because I mean you obviously agree with it because you also said sovereign LLM. What's the economic argument is this essentially I mean what's the economic argument they let me not put words in your mouth. Honestly, the one framework that that's always worked when we're talking about IT services is cost. And by cost I the principle I always think about is Christensen's principle about what happens when the cutting edge becomes unnecessary. So you're going to have these frontier models, you know, absolutely obliterating compute across the globe, you know, and they're going to push all benchmarks out of the way. But the typical enterprise beat in India or beat in the US they're not going to need such intense, you know, capabilities. And my broad sense is more than sovereign, I think it could be a bet on how cost effective a servom would be vis-a-vis, you know, an opus 4.7 for let's say 80% of the tasks. So the framework that I am using is 10 HCL build a some IP on top of a servomodil which costs 20% of what you know a Claude opus 4.7 or whatever, you know, like this 15 models are going to be called can they execute 80% of what the Claude's frontier model can do at 20% of the cost. Why do they need to make a venture investment, you know, for that why can't they just like I presume license servom's model just like any other business out there, right? Or I mean, I could argue that why can't they just use one of the open weight models out there in the world, many of which actually perform much better than servom currently. But I can assume that that argument will very quickly come down to or but they are Chinese models and the Indian government and businesses will not want to be seen as using Chinese models, etc. That argument I can buy, but why do you need to invest be a lead investor in a company to use on build upon its models. Yeah, no, I think it's funny, right? I mean, as Kashyap was mentioning Indian IT services, they've been notoriously bad at committing capital to new technology, right? And that's kind of led to the situation where, you know, the sector has been sometimes a sitting duck in terms of new technological frontiers. And I think this is probably a very, very good corrective step from that perspective. I mean, sure I can license a server model and I can license a client, etc. But what might happen is if I'm really owning the IP or if I'm really kind of participate or I'm really owning even the model layer that I kind of gives me the freedom to monetize different aspects of of a client rather than just being an app services platform, right. So AI is basically a five layer AI, five layer cake, right. The bottom most layer is energy. The layer above that is ships. The layer above that is infrastructure, right. The layer above that is models and the last layer of causes applications, right. And we've always played in in a layer actually just above that which is application services where we've not even owned the application IP, right. And the application's layer might not really make us any money. So forget about app services. So I think this is a this is and to be honest, I don't know if we if we asked CVK, I don't even know if CVK has an answer to what to how you know as as to how he plans to monetize this. But I think it's a it's a really good sort of you know, so the dice so to speak in terms of how to monetize that five layer cake. What about you? I think of being a strategic investor like this would probably help them inflame the roadmap. That's that's one reason. The second thing is I think it feels India is I think this probably it's also a bet that the India market is going to grow and they'll be in good position to leverage that. Probably I don't know this I wish I could know this better but probably 20% of their revenues could be India focused so that that's something. So but more broadly, I think what is fascinating out here is how everybody talks about this new role that is coming up, which is like the forward deployed engineer. So the whole item services industry is based on the offshore programmer. So now the official programmer versus the forward deployed engineer the forward deployed engineer I mean fancy term, but you could think that it's roughly I mean not not Apple's tap is but roughly analogous to the onsite engineer. So since the time I have been in the industry for the last 25 27 years or so. The quest always has been to bring that down from say 50% on site to 10% right now, I think the metrics is 10% on set and 90% remaining off-short. So now the model is at a crossroads. That is, that is the model is at a crossroads. It's quite interesting to see. Like I think Abhishek made a fairly interesting point. We said that the cost of this off-short program, entry level off-short program is like sort of equivalent to the cost of tokens that you put in. Sorry, I mean, Abhishek was saying is that inevitably it will get there as the cost of tokens falls. But I mean, the news from the last few days has been exactly in the opposite direction, which is really that at Microsoft, et cetera, they've been cutting cloud code subscriptions because it's too expensive for even Microsoft to be able to handle. So the longer term bet is that tokens get cheaper over time and you're not at some point token costs will converge with salary costs, et cetera, but I don't think we're there yet. I do want to read out this very interesting quote, which was said by Adele Founder and Chairman Sunil Mithal, who recently said earlier this week, which is, we do divot last week, actually. We do dividends and buybacks, but we'll never become like IT companies who have done nothing but just take money out as dividends and buybacks and become a shadow of themselves. Many of those companies should have been buying leading-edge businesses in their own industry in the last 10 to 15 years. I think this quote is instructive because it makes a point that much of this should have been done 10, 15 years ago and not just now, but let's move past that and come to India. And I want to talk about this point that we were talking about, which is the service and the India business. One starts here, roughly out of the 112 crore, which was dispersed as part of the India mission, GPU subsidies, close to 99 crore went to just one company, that was Sarvam, of the 4,420 through GPUs allocated, 4,096 went to just Sarvam. And this is out of the 12 labs that were selected. So pretty much Sarvam has emerged as the face of India's sovereign AI/foundational model. So that for me is really interesting because there seems to be a contrast now because if you really look at India has always said, and I think Nandan has also said that India follows this very unique third way, which is between the US and China. And I think a third way is sort of emerging here. If you look at the US, which is really private equity led, private, not much of government interference. Don't know what Trump does to that over the next six months or so. But labs operate on their own, right? And then you look at China, which is where you show there's a lot of private labs and open schools, but the state strongly influences that. And then you have Europe. When you have examples from, which is a Mril Plus Daso plus AMID framework in France, then in Germany, you have Alfa and Schwarz and the Kohir combination. So this entire Sarvam plus HCL plus India AI mission sort of converging to be India's AI champions stack sort of makes sense to me because that was a whole that was sort of existing in India, right? You can't be such a large economy and not have, you know, a sovereign stack of your own. So for me, at least from the outside, a certain premium to Sarvam seems to be because of the fact that it is de facto India's sovereign AI and that will determine a lot of the business that it might get in the future similar to what's happening in Europe, in France, Germany, et cetera. How do you two see that? I'll go first. It's, I think given the scale of our, the scale of the budget for India AI mission, I think it makes sense to concentrate your bets and back on our two companies. So my only wish out here would be or ask increase the budget and have maybe two, three bets. But if we didn't the scale, it makes sense to do that. So one point and we also need to be prepared, prepared that this, we're in it for the long haul. It's not like that you're going to invest this year and expect results in next year. This is probably going to take like three to five years, 10 years and we should be prepared that these are fast depreciating assets that we're investing into and that's okay because we need to have for sovereign theories and systems. I think the next class of sovereign AI companies that we back should probably something related to national security. I don't think it makes sense to house both linguistic capabilities or it's a trade year related linguistic capabilities into one. There's also spread out, say what's India's sovereign, palantir, will be a palantir. So that that's I think. Yeah, to your first point about we need more funds. The reality is if I'm not mistaken, the India AI missions budget was scaled back from 2000 crores to 1000 crores because of underutilization. So I don't think the problem at any point was that we have too many players to give it to and therefore let's pick a handful. I think we had in some sense as the inverse of that problem, which probably ends up in concentration as well. So I think it has to be a combination of what we say. I'm generally against billionaires because I think they have too much power in the world, but there could be some ways in which those billions can be put to use in like say funding, private funding. So at a small scale, folks like Paris Chopra doing after he's made his money, he's like championing this as to loss funk, probably, probably, you can include any show notes. What is loss funk? So I mean, it's sort of, it has to be a combination. There is no strict playbook and I would also say that like I briefly alluded to fast depreciating models. So other countries like in the list of countries that you mentioned, UAE was not there, but the United Arab Emirates had a fairly solid sovereign model called Falcon. It was like reasonably state of that at some point in time, but with the newer advances, this space keeps changing so fast they had fallen behind. So now what do you do? Do you sort of reinvest and try to be in the game or do you just license Western models? So this is not a problem that is unique to India, but given the suddenness with which the whole LMT abilities have burst upon us, versus the capital intensity. So if you look at, I'm getting a broad stroke number is broad stroke numbers. Probably the entire VC investments that have come into India in 2025, probably is of the order of like say 20 billion dollars or so. Roughly I would say. So OPEN EI itself has raised, if my number is correct, about 190 billion dollars, anthropic 70 billion dollars. So these are like asymmetric fights. And add to all of this thing, like the restrictions that are there when it's something is publicly funded versus always the questions of is it the best use of our capital, limited capital it's it. So I mean, I think there are valid opinions in different directions. And I think we should be okay to have that kind of healthy debates. All right. Abhishek, let me ask you this question, right, which is in the context of Sarvam being a sovereign AI opportunity as a business. I want to connect that to something that you said about that five layer cake. And you said that Indian ID services companies were now going down that cake. And we see that because infrastructure data centers, sovereign cloud, et cetera seems to be where the opportunity lies. And let's be honest, right, a certain amount of that exists because there is some kind of regulatory guarantee and wrapping around that. I mean, cloud is a cloud. The moment you say some clouds are sovereign and some clouds are not, it is some some, you know, some senses. It's a market which is created due to regulatory action. Now, how do you have you done at Motilal Oswal, any kind of analysis that looks at India and say, Hey, if you look at the sovereign AI into cloud into data center model, how big is that opportunity? Because my sense is going to be huge because if you look at the number of companies like Adani is Google, Amazon, Adel next try. Everyone is going into, you know, that cake, right, that you mentioned down the cake. So have you tried to contextualize servum into that cake and the opportunity that it represents? Yeah, I think more than servum, I think I want to also kind of continue on on the point Kushyap was making right. And that kind of ties down to the earlier point I was making where these battles have always been asymmetric because India's capital markets have never been that deep to kind of really hold 70, 80 billion other investments in any technology cycle. And the way we found our way around that is we've always played in managed services. So if I were to use the cake analogy, going down the five layer cake needs more and more capital, right, chips need more capital, fabs need more capital energy needs even more capital. And through the last 25, 30 years, we've actually played in between the two layers of a cake and that's the plumbing, right. What we've always said to the world is, hey, you know what, I really can't build you and as your or a Google Cloud platform. But if you need to help implementing it, I've got English speaking engineers who can do that. very, very well. And I'm a platinum partner with Azure, and I have a platinum partner with SAP and Salesforce, and I can do it for you. So the managed services provider model has always worked for India, right? And the same model, I think, will also eventually take its place in AI as well. And SRVM, I dare say, would be an exception rather than the rule. Because by its very definition, frontier models are capital intensive. They need a tremendous amount of investment, a sort of pockets. And what we will see, Indian IT services companies do in the next three to four years, is look at the next generation managed services providers who are platformizing the implementation opportunity. >> What is that? >> So earlier, yeah. So earlier implementing, so if I were Walmart or a JP Morgan, implementing Azure or AWS involved 500 engineers who are manually recording and kind of really changing the architecture of my own platform. And this was our labor intensive exercise, where a thousand engineers would work for 18 months. And that's how a typically smaller business promoted the cloud, right? AI changes all of that. So we've seen examples in the industry, where we've talked to a few founders where they're agentifying this entire exercise. So a migration exercise, which would earlier take 500 people and let's say 18 months, can today be implemented with 50 forward deployed engineers in a macro of three weeks. And what Indie Lite would have to do is along with SRVM, they're gonna have to look for these new age managed services providers. And they have to be new age plumbers, because the nature of the competition itself means that we'll always make more money on the plumbing side and not on the IP side, because creating IP is a whole different ballgame. So I'm not sure if I can kind of answer the question, but that's the framework that I've always gone with and that always works. >> Hmm. I still know we've largely been speaking about, I think even this answer that you're talking about, is still talking about IT services in the context of its traditional international business, right? I mean, bring back the focus to India, because that's more aligned to the sovereign opportunity, but most specifically, there is a lot of money, which is lots of zeros, thousands of zeros of crores, which is floating around, I think SRVM signed with some kind of a agreement with the Tamil Nadu government, which if I'm not mistaken, now again, please take these numbers with a pinch of salt, because this is the equivalent of those states doing investment, summits where they announced, we announced like, you know, five black crores of investment, so we've got to apply deflation factor, but still, 10,000 crores with Tamil Nadu, then there was, UP government did some many thousand crores around AI, Maharashtra AI policy, if I'm not mistaken, 2026 with 10,000 crores, Andhra has the quantum valley, Gojarrath sovereign AI park, Karnataka has been linked some 9,000 acre township, so I'm just saying that whether it's data center infrastructure or investments or companies, etc., or government spends by various state central government bodies, etc., there's enough money, which is going to get focused into some kind of a bucketed AI spend in India. Is so I just have to go for some that wouldn't serve them, allow it's CL to essentially have a very strong narrative there to align itself into many of these opportunities. So, you know, the data center opportunity is also predominantly split into three buckets, right? The first bucket is the most obvious, you know, the sovereign AI bucket, where the government is basically telling, you know, everybody that, hey, if you're storing data about Indians, it better be in India, right? That's the first sandbox AI sovereign AI opportunity or sovereign data center opportunity, right? This is not really AI, but data centers, right? The other opportunity is sort of, you know, where I'm hosting data from private Indian enterprises on the cloud. I mean, in the sense that HGFC Bank and, and so those, you know, airtelling all these guys, they don't really need my data to go to the, to go to the US and back. I can just run everything, all the analytics right in India, right? That's all right. But a huge chunk of the opportunity is still in data centers, which are built in India, but for open AI and cloud, right? And that'll still be a huge part of the stack. So even if these data centers are being built in India, we really want open AI and cloud and Google to kind of bring their data from the US to India for inference, not necessarily for compute, because in compute, I'm actually training those models, but there's just so much data floating around in the world looking for a home. And India is a huge chunk of that building is wonderful. And David, I'll just thank for that. There's so much data floating around in the world looking for a home. You know, I mean, just a couple of decades ago, that would have sounded, you know, when the internet was like, you know, there is no, the internet is global and data, every information deserves to be free and data has no home and so on. Here we are, where data is floating around looking for home. Yeah, absolutely right. I'm not denying that, which is essentially the problem of the opportunity. But please go ahead. Yeah, but I mean, yeah, so I mean, see the theater society, the underlying point is we're going to have to build for, and we will build for those people because nobody pays as much as the US clients, right? I mean, you can have all the governments in India and all the Indian plans, but, but, you know, the dollar value or what a US enterprise is, if you know, can pay, it will still spot super seed, what it can build locally. So from that context, you know, TCS has also kind of come up with their own plan to go down that file area, IK, they're building their own data centers. And the story is very similar, right? So if I have a sovereign AI stack with serve as the foundation model, can I offer a US enterprise client just thinking out loud here, but can I offer a US enterprise client an opportunity to train their own payments, data set on my server AI stack in India? And is that let's say NX cheaper than maybe doing it in the US or in Singapore with the frontier models, right? So there's this, there's various non-linear use cases. And when we think about sovereign AI, I think, I think to think about it in just Indian context will kind of be doing a disservice to it because eventually when we build a data center, we need to build so that the global enterprises, they kind of come looking for these data centers in India, where they can host their data, where they can run analytics on it, and then they can kind of, you know, that that's where the real money is in my opinion. So quite the same. I want to come in here. So this is this formulation of maybe we'll do inference in India for the rest of the world. So that doesn't necessarily have to happen through servant stack. So before that, I think your formulation that there is a lot of money in Indian state comments, municipalities, central comments, that opportunity is going to be pretty big. And I think one of the motifs for the deal that we're talking about, server plus HCL. So that is the one of the driving forces. So in software plus services, the general, there used to be a thumb roll. If you pay one dollar for software licenses, anywhere between five to 10 times of that is the implementation revenue, the services revenue, the systems integration, consulting, etc. And that's not going to change because we're using foundation models now. So if we want to pay this much more budget in tokens, then probably it's not going to be 10x, but let's say 5x. Because these agents are not going to write themselves for the next five years, at least I can guarantee you, there is like tremendous amount of integration work, tremendous amount of plumb being, etc. to be done in services. So this is like, I mean, you'll see like these, what does that come for? James Google on the one side, partnering with the their Indian service providers versus the IBMs and the Microsoft. Everybody is going to be circling with the same opportunities. So the services pie is probably going to be slightly shift because of the splitting cost of tokens, the splitting or small slides that are going to be operating, scaling loss. It may still come to think. So the services opportunity is pretty big. So that's why some of, I think in the beginning of the conversation, there's not an existential crisis. But I would think the services companies broadly are at crossroads. The models have to change. They may not necessarily be generating as much employment as before, but their business models with enough tweaks, if they're able to do it, will be fine in the longer run. Will be fine for the longer, medium to longer. So that is one thing. So whether an extensor captures it, whether hits CL captures it, it will be pro captures. It's that remains to be seen. So we want to see a lot of consortium plays. So it is not necessarily in the case. So if you wanted to set up, for example, I'm slightly even going to a shake's point of inference in India for global clients. If you wanted to set up say a one gig of what data center in the US, I think it's roughly about $80 billion or so. They would take a few billion. In India, it's like $20 billion or so. So but we haven't approved that yet. We haven't proved that we can execute those data centers. I think Ken itself has I think our series recently on Love Me don't have the electricity ready for that. So before that opportunity is realized, we need to fix like other layers of the cake like electricity. So that is fixed probably that can happen. And there are also certain data residency rules, just like India is saying all data should be resident, all Indian data should be resident in India. Other countries have their own data residency rules. So there is going to some regulatory concerns there as well regulatory factors that play to see if that that opportunity will play out. Correct, but broadly there seems to be the movement towards kind of the internet's becoming sort of like a splinter net at country levels, etc. Right? Like, you know, and to come back to India as well, where you talked about the opportunity, the fact that reliance has committed, if I'm not mistaken, $110 billion over the next seven years to build energy data center infrastructure AI stack, etc proves that there is going to be a significant sovereign AI opportunity. The way I kind of read it is that, you know, I mean, probably even a year or two years ago when we talked about foundation labs, we always tend to look at whose performance is best. But it doesn't look like that matters because every country is going to choose a couple of home-grown AI companies who made kind of days as local champions and it doesn't matter what their performance is with the rest of the world because they are their companies. Right? So I think Sarvam is in that sense that company for India. I want to sort of flip this question now from Sarvam's point of view, right? Because we've been discussing this largely from an HCL tech point of view. Now, Sarvam is a venture funded company. It is a private foundation lab. It does have a business model. I would normally assume that, I mean, the way you typically look at venture investments is that investment led by a strategic is never an obvious option, right? It is, you, I mean, labs, especially in early stages, whether it's a lab or a startup, if you see significant growth ahead of you over the years, then and you feel envisage taking multiple rounds to fund that growth. You typically want to raise from investors, not from a strategic because the moment a strategic comes in, one of the questions becomes what control will this strategic execute over you? Is this strategic going to buy you? Are you going to be in some kind of a proprietary arrangement with this strategic? So it deffers or like future investors look at that as oh, this strategic has got dips on this company. So what does it say from Sarvam's point of view? It seems to be a, I mean, like I said at the beginning of the conversation, I have no information from Sarvam, but this isn't an obvious choice, especially because AI is still at a very early stage and companies are coming out literally every three months, six months and raising. So the appetite for capital from global investors for AI is significant. In this kind of an environment, if Sarvam goes and allows a strategic to take the lead, how do you guys read that? Yeah, maybe I'll go first on this one. I think this is, I think from Sarvam's perspective, I think it's probably cut and dry. I mean, it's, it's, what Sarvam is getting is sort of the roller dex access to, you know, fortune 500 clients across the globe. You know, it's, CL has been this marquee technology service is provided globally and they operate with top five US banks, you know, top ten US retailers. I mean, they operate with auto EMs across the globe chip, chip manufacturers. You're getting access to this, this gold standard of enterprise clients that you might probably not have a fruit in the door and possibly, you know, considering, let's say a small size and that's not to say the capabilities are bad, but it's just that the go-to market and sales is a whole different animal rate. So from that perspective, I mean, I mean, my, my broad sense is, you know, all the complications of, you know, a strategic investor aside, if I can get access to auto five US bank and if I can, I mean, imagine ACL walking into the, you know, the, just out of context, the city bank or JP Morgan boardroom with a servom LLM and it's CLS built an agent on top of servom which kind of smokes a cloud or an open AI at one 10 the cost, you know, that's what I think servom hopes to achieve. I'm assuming from this. So, okay, now, we shake like sick and I'll tell you why, right? So, so firstly, there is absolutely no evidence we've seen to allow that servom has been trying to target international clients. Because if they were, then we would have seen a concerted effort from servom to be part of global benchmarks, etc. And so because let's speak one of the scenarios that you created, right? Say servom, we're to partner with HCL and go together to a global organization that global organization, this CTO, its staff is going to ask, okay, you proposed to use this particular model. Why? And they're going to demand to see benchmarks. Now, you know, I mean, so let's say, let's agree that you're not going to be beating foundational models from the likes of opening eye or anthropic or Google, etc. That's ruled out. So then it comes down to so servom says, no, we're not aiming for that, but we'll be lowest cost. Now again, a lot of the, then they get compared against the likes of Gwen and deep seek, etc., which is coming out of China, which by the way, are open weight and can be, right? So there too, I haven't seen absolutely any benchmarks. And even if they were, I'm pretty sure that they get smog. I think we are forgetting that models are not going to be modes in the next 10 years, you know. What I mean by that is, I mean, anthropic and open eye can be absolutely, they can be winners at the frontier, but models will be utilities, right? And the money will be made in agent orchestration, the money will be made in LLM ops, the money will be made always on the cost side, right? And why will the server be able to operate at a lower cost for the same level of efficiency than any of the open weight models? Because it just has a lower number of parameters. And to be honest, again, I have not really gotten into the integrities of a server, etc. But I think, I'm just thinking from a first principle perspective here, and I may have agon my face five years out, and server may not have a single person's right side. Exactly. So, I mean, like I said, I mean, this may be, you know, it may turn out to be ridiculous as to it, but from a broad perspective, when a product company ties up with an MSP, it generally is for go-to-market. That's the only, that's the only first principle logic I have. Except what if that go-to-market is India? That I can kind of fully get on board with that. Sarvam plus HCLT for Indian use cases, governments, clients, etc. would be a killer combo. The only issue I have, and maybe, and forgive me, you know, sales head analysts are just, we just notoriously linear thinkers, and you know, we are, we're slaves to our Excel sheets, to be honest. But what is the quarter? Yeah, I mean, I wish it were different, but it comes, I mean, yeah, it's a hazard of the game. But, but, but, the point I'm making is, it's been 25, 30 years, and HCL, Kushyap, to your point earlier, only 3% of the revenues come from India. The best IT services companies in the world, everywhere, be it Accenture Cognizant, will have 50% of revenues delivered from India, but India accounts for less than 3% of their business, 3 to 5% max. And there's a reason for that, because Indian enterprises cannot pay you as much as the US enterprises. And it's not just India, I mean, Europe is barely 20% of business. So, if you really have to make money in tech, you're going to have to service the Fortune 500 lands in the world who can pay you an outsize amount. And you know, you can take, you know, a sovereignty, a stack, we end enterprise. But, but, but the, but the paying capacity will, will, will just be very materially different. So, so my framework for, for SRVAM plus HCL, they will always be, how can I take this, this, this new stack to my existing client base, which is in the US or, or possibly in the Euro, or maybe Australia, rather than kind of building for India, because, because the model is just not trying to. Okay, Kashyap. Yeah, I think a few quick follow up to strands to that. So, one is that, so that one needs the money right now. So, they'll take this 300 million and then become competitive, then they can raise follow on raises. And I think if I'm not mistaken, they already have the Indian woman on their capital. So, so you're essentially saying he who survives, lives to fight another kind of, correct, and probably, I mean, to give them credit from whatever I have seen, they believe in the mission as well. So, they believe in the mission that said they have the Indian government has the strategic investor on their capital. Now, they have like an India focused systems integrator also, like an Indian system integrator also on their capital. So, I mean, so that that's pretty interesting, but, but the India, one interesting sort of a byproduct of this discussion is that I think a lot of Indian enterprises who are reluctant to pay for software licenses will probably try to wipe code so the India IT market itself may not expand significantly. You know when it comes to wipe coding my theory is that it's going to create such a tsunami of security issues and downtime that it's probably going to create even more opportunities for Indian IT services companies to come and fix it as long as they're not using AI for that but that's one of my thesis actually. Please go ahead. That's one of my thesis because if you look at the this sort of in application development and maintenance services the composition of her quality software quality testing is about 15 to 20 percent and what kind of validation happens is very different. So there's a huge opportunity for the services companies to actually convert that into testing of wipe coding security evaluations. You need to re-skill like a mid-men so if you have to look at if you look at the amount of effort that goes into like say building an application right now the bulk of it is in coding which is 50 percent, let's say 50 percent of the project time is towards coding there is specifications there is like testing sign-offs etc that is like the rest of 40 percent or so. So now that because you have AI systems which are fairly decent, coding coding from application development problem is probably going to be 15 percent also but the pie that is towards the quality what we call software qualitative that is going to increase because of this technical debt, AI debt etc. So we have which which the Indian IT services industry is missing they have a million testers on their roles so they need to be seriously upskilled to test AI applications to become proficient at security testing etc. So that is the massive opportunity that is right in front of our eyes. Very interesting QA and testing were considered the least sexy most boring nobody wanted to work in those functions in IT services over the years. You're essentially saying as wipe slop invades software that becomes an opportunity because the bottleneck is no longer creation but the bottleneck is validation. I think I'm really sorry guys but my channel checks suggest something completely opposite. Testing and QA are sold like they're going to be eliminated. What is being served offered is the testing and qualifications in their current form that you don't need like a manual tester but testing of AI applications itself that is going to be completely different ballgame. It's a good show. Yeah I think I don't know what format takes but before that I have a fantastic chart that I keep pasting in all my notes right to your point on wipe coding guys I think we agree on all of this which is a really fantastic chart. So in 1985 50% of all US enterprise software spend was done on self-built software. They basically not wipe coded but literally slop coded their entire enterprise stack. That number in 2024 was less than 10%. Over the last 40 years the enterprise spend moved away from self-built and custom software to prepackage software. You went to SaaS and why did you? You went to SaaS because owning code was just horribly painful exercise right? I mean owning code maintaining code securing code was just so awfully boring and tough with so little ROI that I just went to kind of SaaS and now that everybody is thinking about wipe code people have forgotten that owning software is still horribly awfully boring right? It's like saying that if my fuel prices are down I'm going to go and say, "I don't know if you can make ethanol out of it." Exactly. So I don't think wipe coding is definitely something that's validates one of the most fundamental truths about business which is we're just destined to live the same cycles go up and down every few decades right? It's something in the end bundled and then something is being re-bundled. Okay let me bring this discussion back to a close. I want to ask you folks sort of two closing questions. First question is while all of this is happening the foundation labs, the most ambitious foundation labs specifically Anthropic and Open AI have been doing very interesting things and now even Google which is they've been forming these joint ventures with these massive private equity companies in order to go after these opportunities that in banking financial service in other sectors that IT services companies target. Now this argument I've been hearing for the last maybe a year and a half or two which is really what does AI do? What AI does is up till now what used to happen in the context of software was AI IT services companies or software service companies only used to capture a certain percentage of the value in a particular sector whether it be healthcare or insurance or real estate or finance because they would go to this company and say here use our software or use our services and we'll make a 10% margin from you or 15% margin from you in return for providing this software. Now these the the narrative behind these massive JVs and the interest of private equity in this space is they're saying look can we just go to these sectors combined with AI roll it up and capture all of the value ourselves. We don't want to be in this business of merely going and selling software and services to some sectors. So there is that very real threat that IT services companies and software companies are not going to be competing with you know peers but AI lapsing you know let's just cut to the chase we'll just partner with P's and capture all of the value let's just do an end game. How do you see these large P back JVs that anthropic and open AI are forming? I think they're better off or eventually they'll be forced to partner with the technology services companies in India that it happens and at the badges of the IT services companies or that it happens their GCCs like I said I mean agents are not going to write themselves there is a lot of domain knowledge there is a lot of sort of software building experience there is a relationship that is not captured or that doesn't exist with the and this is not just so the models the model the foundation models are in some sense products products on steroids so none of the product companies are been successful in having implementation of businesses as large as the services companies so I don't see that why it should be any different because there is a certain DNA like it we're not talking about Google's but there is a Google graveyard of products that they have sunset and so for example I'm not picking one Google but if you look at for the cloud providers themselves Google's technology is sort of very good in some areas Google fraud and talking about Google fraud but because their reputation in the market for services is not very good for support is not very good Microsoft steals a lot of cloud share from them so with the and with the sort of frontier labs also it'll be some sort of a similar story so it may be good like I think one of these joint ventures that you're talking about is for specifically focused on the chip flair so that's a different ballgame chips I mean it makes absolute sense for these frontier labs to partner with the private equity labs whoever is giving capital to do that but for the application services I think that play and involve India in some form or the other is that through the GCCs or through the technologies that this was companies hmm how is it you want to see something? Yeah I think kind of a very similar vein you know I guess I'll talk about the deploy co sort of announcement that open air had right where they've basically formed a GB with a few companies and and I think the entire IT index sold off that day because they thought yeah this is the IT services killer you know we've got deploy co-nau who can take business away from TCS and FOSTS etc. I think there's two things right a platform or a product's business is a winner takes all or a winner takes most business right I mean if I can get my product right I can sell it to everybody and everybody's going to use SAP right services and managed services are actually very fragmented businesses historically with very very limited pricing bar so if you look at the outsourcing boom of the 2000s there's a trillion dollars of IT services spend in the world and the top five companies account for only 10% of it because there is no pricing power no competitive mode and you need engineers deployed in a physical form at the client site right I think AI implementation will take a very very similar sort of sort of direction as well you will essentially have not one but a thousand deploy co's globally who will basically be the new template for managed services the question for Indian IT is not whether they can be defeated or saved by deploy co but effectively can we acquire buy or replicate this this new age managed services provider who somehow looks a little bit like Deployco who looks a little bit like this new age MSPs and I think that's the game for us right. So I think it's not either or I think it's about whether we can replicate this new template of IT services. Quick sidetrack to the point that Kashyap made about Google's graveyard of products. I'm going to read out a tweet which is just a few days old. We'll link to it in the show notes as well as this person called Nathan Clark. It's in Gemini just created in AI studio. Oh that's for your personal Google one account. For workspace you need Gemini business. No not Gemini advance. That's AI pro now unless you need AI Ultra. Oh agents you do that in Spark. Actually no not Gemini API managed agents. That's different. For coding use jewels unless you mend the agentic ID that's anti-gravity. No that's the old anti-gravity download the new one. Actually Gemini CLI has been deprecated used anti-gravity CLI. No the flash model is smarter than the pro model unless you need pro if it's video use flow no flow uses VO no nano banana is images actually that's in Gemini now unless you're in search then it's AI mode no no research is notebook. Anyway it's all very simple. So Google is absolutely still killing it if anything Google has super charged it's branding and naming strategy using AI it is super confusing I use Google AI products and I can tell you it is super confusing right so So what do they say you ship your art chart? Yeah and this is after Google has done all those cleanses of its art chart right seriously but to be fair to Google I think Google has done a tremendous job after the initial body blows by OpenAI right and this is actually killing it on so many fronts. Closing statement gentleman look at the next 12 months specifically from the lens of HClT's server I'm not asking you to comment about this specific deal but when we look back a year from now at this announcement what might we think? Yeah I think this should be seen as an indication of its L.T. willing to disrupt its own book of business you know when it's not into new sort of areas and maybe just ask questions of itself as well as to what can the next sort of you know monetizing opportunity be and it may not be server more it may be server but I think for me the direction is very encouraging we do need these acquisitions and the industry will have to make these sort of these overtures to sort of these new age companies because that's the only way to kind of come out and sell out of this face. So three quick points one is I'll broadly look to what happens in fell to 18 months like in the follow on drums so that'll be quite instinctive to see how the thesis has played out and I'm also looking forward to similar deals from our other IT services companies because this now has given a template because the discussion so far has been should we build a front tier model or not but now it seems like you don't have to build you can actually invest so that that's interesting but this seems like a market of one right like I mean because there's frankly no other Indian model that any other company can invest in. We'll see and there are some waiting in the wings so they can scale it they use the opportunity. The last thing really is like for Hitching Tech and for other services and consulting companies not just investments and frontier companies but to realize on this execution opportunity there are so many other adjoining categories and orcitation platforms etc. orchestration platforms and whatnot. So will it spur investments in those things because that may be sort of a more traditional investments are not worried about is this foundation model play won't it be profitable or not but investments in software tools that help you deploy AI that help you validate AI security AI etc. It's sort of interesting to see. That's that's very interesting if I'm you know I think one of the companies which I would want to kind of mention in that space might be the vibe the vibe coding platform emergent right because that's the only other company that I can think which sort of is adjacent to IT services company but who knows because I think they also raised if I'm not mistaken a pretty large round in Jan 2026 from SoftBankins course 11 cheers. But thank you gentlemen wonderful conversation and so many interesting analogies including you know the layer cake and Abhishek you want to share with us that chart which is energy of favorite we'll include it in the show notes of course and thank you Kasha thank you Abhishek you thanks you