EXCLUSIVE: India’s Sovereign AI Push Has Silicon Valley Worried | Front Page

Added: 2026-05-09

[00:00:00]In 2023, 815 million Indian health records were exposed in the ICMR breach.

[00:00:07]The largest data breach in Indian history. A year earlier, Ames Delhi was hit by ransomware, patient data, critical systems, weeks of disruption.

[00:00:17]And just last month, Verscell, a major global tech platform, was breached through a third-party AI tool. One of its employees had actually connected it to their corporate account. Every one of these incidents had the same root cause data that left the building. Many Gupal Gangana is the CEO of Langur which I believe is fascinating name for a marketing company which is of course transforming and now deep is in deep conversations with CISOs and CTO's it has never expected to be well talking to. Then we have Angad Alualia who is the COO of Arerinox AI which is also a fascinating name for a company and I'm of course saying that also because I don't want to offend him before I bring him into the conversation. He and his company are building the infrastructure layer that makes sovereign AI something you can actually deploy and not just promise. Together they have announced a very strategic partnership to bring on premise AI to regulated enterprise sectors. Ladies and gentlemen, please put your hands together for both of these brilliant men. Welcome to front page.

[00:01:28]>> Happy to be here.

[00:01:29]>> Wonderful to have the both of you. Hello Angad. Hello Venu.

[00:01:34]>> Very nice to be here. Yeah.

[00:01:36]>> Yes. And we are very excited to have the both of you here and we want to kick off the conversation because there is so much that I would like for you to go ahead and educate us all about because this I believe is impacting all our lives. I want to start with you Angad.

[00:01:50]Every major breach in recent memory as I just mentioned India and globally involved data that left its origin environment. Please make the case for all of us why cloud first AI is actually the wrong model for regulated sectors and what sovereign AI actually changes about that very risk. Mhm. So you know there's been a lot of chatter especially in India about sovereign AI and owning your intelligence owning your data owning your infrastructure and residing it within your country right and in fact to really understand sovereign AI one actually has to understand the various layers of AI right and as Jensen Hong practically put it layer cake right you have at the base you've got your infrastructure or in fact energy below that on top of that you've got your um your uh your layers for inferencing and models and on top of that you've got the application layers right now the application layers are essentially the AI layers in which the consumer or the end user or the client actually play with the touch and feel AI right now sovereign AI initially was about keeping your data in your country and then for a while that evolved into keeping your data within your enterprise >> correct >> but what what but it's not just about the data or the models Right? It's not just about having your hardware.

[00:03:13]>> If you go up one layer, the LLMs, right, also have the propensity for breach.

[00:03:19]>> Your LLMs which are open source and stored in your own infrastructure play a very big role in making sure that your data, your processes, your ownership of your IP is maintained.

[00:03:31]>> Now, let's go one level higher, right?

[00:03:33]Which is the application layer. And this is truly where all the magic happens, right? This is the marketing in a box.

[00:03:39]The the solutions that you build for enterprises. The end user touches this, right?

[00:03:43]>> This is comprised of so much proprietary company data.

[00:03:47]>> Oh yes.

[00:03:48]>> Your own processes, your own uh incident managements, your own SOPs, the learnings from your own team, right?

[00:03:55]That also should not be going outside.

[00:03:57]Absolutely.

[00:03:58]>> When you take sovereign AI as a whole, it's all these layers, right? And then do you want to keep that in a different country? you want to keep it in your country or you want to keep it in your enterprise within your own servers or your own uh four walls. That is why the understanding of sovereigny I should uh is not absolutely uh crystal right it's evolved >> and the last layer the one that I defined where you own right from the infra >> to your IP in terms of the process is where you truly get sovereign AI truly protected AI you you're you still empower your employees to use local tools or AI tools you know the other using cloud inputting information that they shouldn't be >> safely securely with no inhibitions whatsoever. That is true empowerment, right? So that is the solution that one needs to truly leverage AI in a sovereign safe manner. And all those examples that you gave were nerve-wracking. They they they shook the world >> and the sheer amount of the the the number of uh cases for example or the sheer volume of it >> is very very you know disturbing.

[00:05:07]>> Yeah. And and and that's the fear right and then you associate that fear with AI. It's not that it's the governance of it. It's the controls of it. It is >> Thank you so much for giving us distinguishing that ang. It's very important for you to have you know brought that to light. Now of course I'd like to bring uh both of you uh hello Venu and please before we get on I have to ask you what inspired you to uh you know moniker or give christen your company with that name. It is a fascinating >> I'm a I'm a I'm a tech kind of data guy trying to act like I'm a creative and a marketing guy. Uh sometimes I call it stupidity pays off right. So I thought outside of human beings if there is anything that is uh uh smart but as retained curiosity is a monkey. So we we would like to call ourselves a bunch of people who are curious who are trying to marry creativity with data with tech and that's where the name origin from.

[00:06:02]>> Wonderful. I I'm so happy because I completely agree with you and of course uh now I want to bring Venu. I want to start with you. This question is to the both of you Venu and Angar. AI in a box an entire AI system which kind of runs on hardware inside your own data center.

[00:06:18]We've established that Ang right no external servers or third party cloud.

[00:06:22]What makes this different gentleman from the platform enterprises are already evaluating and where does it typical kind of face challenges when it hits let's say a real bank's IT and compliance stack >> okay maybe I'll take the real use case of that right so in my in in in our experience um the air in a box I think is overly used term across enterprises so let me demystify it a Enterprises today go and implement AI platforms by evaluating multiple things, multiple tools. They stitch vendors, they stitch systems, they stitch integrators and they implement something which is very horizontal for an enterprise, right?

[00:07:07]>> Okay, that in itself is a problem, right? Okay.

[00:07:11]>> Uh what is different in a AI in a box which we are hitting the market >> sorry why do you say it's a problem, Benu? because you you have not designed it for an AI native organization. Ah okay fing right over the years you have tools now we are adding more vendors systems processes integrators that I think that is the problem of a large enterprise not going going to go away right so what we are trying to do is we are trying to scope it we are instead of an horizontal AI platform pretending to do marketing >> we are saying let's deploy a sovereign deployment specifically built around enterprise regulated marketing workflows very simple people customer engagements, personalizations, campaign operations.

[00:07:56]This is about running the entire infrastructure that satisfies the enterprise compliance questions.

[00:08:02]>> Okay, >> that's AI in a box where it breaks when you try and take it to a bank, when I take it to a life science, when we take it to an insurance company is >> it rarely breaks at the technology layer. Okay, because that exists infrastructure, right? It breaks at whether the enterprise is data ready.

[00:08:24]Step number one, right? Is the data ready for AI? Okay. Second is is the process already from a baggage perspective. Is the process built in correctly to implement AI in a box? And the third one is of course the marketing teams, change management and people. I know it's a overly used term of data, people, and process, but it breaks there. No, but it makes sense what you're talking about. Absolutely.

[00:08:51]>> Yeah. It doesn't break at the technology and the infrastructure layer like everybody gets scared about it, but it doesn't.

[00:08:57]>> Okay. So Angad, I want you to weigh in on this and tell me this um especially I mean you know Venu just mentioned very highly regulated sectors >> and I think one of the biggest challenges that these sectors are facing is actually at the data layer as Venu very rightly pointed out and they are also struggling right because probably that's why the implementation etc is taking more time so how is it that you are probably trying to overcome that challenge you've established okay that's the challenge how do you overcome Okay, >> I'll I'll give you an example. Okay, you know, actually, let me just touch on AI in a box. It's one of the most used abuse terms because it's, you know, it just rolls out the tongue so easy. So, people tend to use it a lot.

[00:09:39]>> But AI in a box actually basically means a plug-and-play solution on a platform >> that allow you to adopt AI for your company on your data privately at the get-go. Right? So what that means is this so-called platform has again Jensen's all layers the hardware the chips >> the inferencing layers the uh the foundational model layers the AI application layers and then when you go deeper into the application layers and personalize it for a company is when you get >> marketing AI in a box right these are toolkits that are vertical built on a horizontal platform >> now these vertical tools now every organization ation even if it's marketing or it's insurance or whatever has unique processes >> right and so a horizontal platform allows you to build vertical solutions that are tailored to that organization >> right and that is AI in a box right so you take it you plug it in and you're good to go >> a lot of that depends on two things right one is the subject matter expertise of the of the of the person who's building that last layer >> and two the data that gets into it, right? Which is >> that data that gets into it, >> right? And because there's no such thing as one solution fits all for the enterprise. True, >> right?

[00:11:01]>> Uh it was true for traditional software ERP, right? You have to follow those those those guardrails and the whole company >> maybe minor tweaks here and there, >> minor customizations, yes, but you find that the entire organization >> pivots or changes. There's so much change management. you adopt a a vertical technology. People change, processes change, software change, people get fired in two years have gone by and it's probably one of the most painful experiences for a transformation manager.

[00:11:29]>> True.

[00:11:29]>> With an inbox, it's actually very different. The software, the II bench to you.

[00:11:34]>> If I have these 10 processes in my organization, >> then the software bends to you and says, I will do that as for you. So change management is zero. But the only way for that to happen >> is the data availability of that organization.

[00:11:48]>> Right? So you can the data that you use to train one insurance company's flows versus another completely different. The way you test >> risk is different from another one.

[00:11:59]Right? So data is um data availability is probably the biggest roadblock. And when it comes to highly regulated industries like banks and IT, they've got legacy systems that last 20 years.

[00:12:10]>> Yes. And that's the reason why they haven't changed their systems >> because if you software B exactly AI makes it very very different. AI can use unstructured data, >> ingest it, use another agent to organize and structurize that data and make sense of it and create actionable insights. So all those regulated industries that have not changed in 20 years actually have the opportunity to transform themselves completely with AI because AI is doing that work for you, right? Right. And it's opened up so much uh avenue for for invigorating uh banks processes, banks engagements with customers, analyzing their own data, presenting in a more better way, better, making better decisions, right? So I think what you claim was a weakness or what we thought was a weakness in terms of legacy data is actually a huge strength for AI and enter step in for 10 seconds a practical example of how this is implemented in one of the largest banks in India.

[00:13:10]>> Sure.

[00:13:12]>> I think we >> one of the largest or the largest >> the largest okay let me say it that way.

[00:13:18]Okay. So uh I think as technologists when we talk to CIOS and CISOs and CTO's right I think everybody loves this conversation data business data preparedness the CDPs you know from last 30 years donkeyy's years we are all trying to address that lot of people have not matured into that everybody knows it correct so that is the elephant in the room okay now how do you go around it okay when you put the box >> by knowing the functional and banking operating systems. We have built-in accelerators within this box.

[00:13:53]>> What it means is when you switch on the box, automatically an accelerator wakes up and marries your unstructured data without you going into a data readiness program and marries a customer behavioral data with the marketing data automatically in a click of button.

[00:14:08]>> Right?

[00:14:08]>> If I give that then the CTO says, "Oh, it's not another 18 months program. I can switch on my value tomorrow." So where did this accelator came from? It came from experience. Where did this flows come from? It comes from the agency doing this regulated business for last 10. Now you put that into the box.

[00:14:29]Then the value just taps out. Uh so that's like that in this box over 40 50 accelerator sits >> and switches off the value.

[00:14:39]>> Makes sense. And and and and Venu Yeah, I agree with you. I mean because the sheer scale that you just gave that example I think just gives us all the the you know advantage that is coming to the table. Very quickly before I move on on the data residency bit I wanted to ask you is this do you feel that there's a fear still you know because these are very highly regulated sectors do you still feel there's a fear in the mindset maybe because you know these are legacy systems etc. And if you gentlemen have faced these kind of fears, how do you kind of overcome them for the customer?

[00:15:16]I'm purely talking from the customer.

[00:15:19]>> So I think the fear is real. Uh but I think it literally in the last 6 to 8 months that has evaporated because the embracement of that has happened. Okay.

[00:15:30]uh on one side there's a typical fear of that is coming from the baggage not from the reality of what the tech can do actually speaking and that fear is existing subconsciously but when we really have the conversations already people have gone into solution mode we say like hey I know the fear is real let's do this right let's figure it out because there are many ways to circumvent this right that's what we have seen in our experience >> which is encouraging of course now I want to come to you Ang data residency essentially mandates that data never physically leaves your country or building as you had very clearly mentioned earlier a legal requirement across banking healthcare and government correct so it is like extremely critical so are enterprises coming to you for AI outcomes Angut or for compliance comfort and what is that pure business case which is beyond regulation >> so they're actually coming for both I think I've probably spoken to close to about a thousand enterprise in the last one and a half years.

[00:16:32]>> Wow.

[00:16:33]>> And uh the conversations have really evolved right initially when we would approach them, it was what's an LM, what's an agent, what can it do?

[00:16:41]>> Today most of them have built out PC's, right? They built out working models, they built out uh easy to connect solutions, right? So these PCs are typically cloud-based where they can where they connect five or six different tools, >> right?

[00:16:55]>> Uh they open up to 20 privileged users.

[00:16:58]They test them out, they see great results. Uh it's built on limited data because it's a proof of concept.

[00:17:04]>> And so they're able to demonate the productivity of it, right?

[00:17:10]Completely. But when they say, "Okay, now let's productionize this and open this up to my entire organization."

[00:17:15]>> Um finance says, "No, I can't give you that data." CRM says, "No, I can't give you that data."

[00:17:20]>> Uh it says, "Oh, we can't open that to everyone else because the token costs are so high." Sure.

[00:17:25]>> Right. And so this raises the compliance question. So now all these customers have actually done this entire journey more recently as opposed to a year ago.

[00:17:33]They built out the shown the efficiencies and the outcomes but they were not able to transit into compliant requirements. So the conversations today are all about compliance.

[00:17:44]>> AI outcomes are proven at least for the ones who have dipped their feet in right and this is the fundamental shift in mindset from a year ago till today.

[00:17:52]Right? The first thing we ask you is it's great. It's absolutely great.

[00:17:56]Right? Because >> uh I mean it's really no fun explaining people about what is AI, how it works, what they need agent as opposed to tell them what it can do for you. Right.

[00:18:05]Right.

[00:18:05]>> And uh and anyone today who is using cloud-based frontier models to do their work will watch for it. Right? Is it is it right to do it in an organizational setting? Probably not. But will they be as efficient without it? Also probably not, right? So I think now compliance has become the most important uh >> yes >> conversation uh simply because they have proven outcomes through their PC's >> and we help them basically bridge that gap from PC to production to live scalable solutions because we're not handling compliance uh discussions >> from a marketing point of view. uh venue I need to ask you this what Angit just mentioned right that it is encouraging people have moved on now from when you let's say go ahead and and and speak with a customer in nowadays let me give you an example um BFSI let's say as a sector right uh over the last 3 years now there is this entire conversation which has been going on regarding BFSI which is the fact that you know PC's as angel mentioned these ones of course are positive because they've worked and you know people are seeing it and the results etc. So they're em embracing it >> but there is this entire thing which is being said right that uh BFSI is struggling especially the big IT companies are not being able to give them the kind of solutions they want and it's called death by a thousand PC's you've heard that term right?

[00:19:30]>> Yeah.

[00:19:30]>> So how would you like to address that?

[00:19:40]Is this for Wenu or >> that's for Venu? That's for Venu.

[00:19:45]>> I think he's frozen a little bit if I see. Uh >> did I freeze Venu with that question?

[00:19:52]>> Okay, maybe he's back. Okay, he can take that one.

[00:19:56]>> So just repeat the last uh five six.

[00:19:59]>> You've heard this term, right, Venu?

[00:20:01]That you know the BFI sector is suffering because of this entire thing called death by a thousand PCs, right?

[00:20:08]Okay.

[00:20:09]>> And it's been 3 years.

[00:20:10]>> So I mean here's Angad who's talked about PC's in a very positive manner.

[00:20:14]It's working etc. But the so-called tier one companies are not being able to do it. What are your thoughts on it and how does the customer seem to understand it?

[00:20:24]Let's say when you're facing the customer from a marketing point of view.

[00:20:28]>> Okay. Very simple. So the the good thing is uh uh in in in in Langore 45% of our business is BFSI. So I said one of the largest we work with the eight of the top 10 banks in India. Right? So this is a conversation which I have every day with the CMOS. Two things have changed.

[00:20:47]>> Uh death by P is real but everybody has become very smart to do hybrid PC's. Uh what I mean by that is hey I will do a P. Uh I will not create a drama to do P within my first party data. I'll do using third party data. I'll put couple of compliance guard rails to prove that first. Okay. Then I'm going to slowly bring one layer of first party data into it and one layer of compliance. So nobody's trying to build the whole house of cards. Okay. So hence the quality of PC's even while there's death by many PC's the qualities of PC's have improved in the matured players. Okay.

[00:21:28]>> Okay.

[00:21:29]>> That's point number one. Point number two is they they figured out where to do and what to do the PC's >> in in 6 12 months back this was all open-ended PC's to be honest solve the world while doing this while doing that but right now it's very clear that like hey I know exactly what I need to solve for example I need to get my new to bank users improvement by 10% because the entire banking insurance banking sector.

[00:22:01]So the stays on only two use cases new to bank existing to bank.

[00:22:05]>> Sure.

[00:22:06]>> NTB ETB that's all marketing is only that right. So now when I in this P how many new users will come is only a question. So very simple answers to that. So P is fantastic hybrid mature to which use case.

[00:22:24]>> Hence our jobs have become little bit more easier. The second biggest shift is I I think the marketeers and the CMOs have realized I cannot make the decision.

[00:22:36]>> Let me bring the room together.

[00:22:37]>> Okay. When I say >> when I say that they're saying they're already proactively bringing the co and CIO and CISO >> nice >> into the room.

[00:22:48]>> So at least uh uh they're not >> legitimately thinking two smartest people in marketing are not sitting and talking in the room. A CMO and a agency always thinks that these are the two smartest people right the room is making the decision which is fantastic that is working for them >> which is encouraging and of course always is encouraging when you have a wonderful partnership with Angad where he gets his PC's bang on every time I'd always I'm sure >> it's it's it's solving larger problem right so while the PC's are good if the infrastructure is set and secure and compliance and so on that mind share is just getting ticked and ticked and ticked saying that oh when need I can switch it on but I'm not going to use it for every case and every conversation wherever applicable I'm going to use so that's why infrastructure with marketing knowledge when it comes together the so AI with enterprise use cases becomes real >> mythos what are your thoughts I mean you know the entire community everyone seems to have gone into a tizzy uh you know for concerns in banking for its ability to autonomously detect and exploit software vulnerabilities this This is of course for both Venu and Angut regulators and CISOs are already briefing at the highest level. So air gap we just discussed does a model capable of autonomous vulnerability gentlemen exploitation change the sovereign AI conversation from let's say it being a preference to now becoming an urgent necessity because that's what well seems to be the order of the day.

[00:24:23]I mean let me I will share my view uh added to what Angad said. So it's I think as as as marketeers who understand technology right my job is to simplify stuff because I cannot call these uh uh uh jarens uh to get the point home right so air gap in my mind is very simple it's like knowledge that is out there which is learning >> and thinking is two different stuff >> okay >> get we get we get super carried away with the knowledge uh okay so the the in that context I may have great knowledge coming from LLMs whether I refresh it by once in 3 months or 6 months the the change and the impact is only one degree two degree after one after a certain level of learning maturity >> the biggest thing is about the thinking maturity in very simple terms right so once I got the learning into the system how is your thinking and reasoning maturity and what have you built in those flows and that's when the mythos all of these are coming into the picture right in the AI moization I can find 20,000 things that there is a problem if you give the complete control into an agentic flow to make those decisions that is disasters waiting to happen correct so you need to still make those decisions you need to still be in control of that thinking leave the learn uh learning and knowledge to the missions. The thinking is what you need to that's that's how I kind of translate and make it practical.

[00:26:03]>> Fair point, Angad. I'd like to bring that to you. But I'd like to add this uh to what Venu just mentioned. The sovereignity aspect seems to of course then well you know become front and center as far as any discussion is concerned and maybe what Venu said that could be another reason why Anthropic has probably released it in a controlled manner because of that entire glass wing project etc. So yes, what Ben Wenu said makes sense. It's a controlled release, so let's understand it. But even the controlled release seems to have well that's not the issue, but the issue was the access to it. They didn't even think about that. They talked about the release, but they did not think about the access where some hackers got into the entire excitement of trying to go ahead and say, "Okay, well, let's say hello, what's behind the glass?"

[00:26:50]>> But sovereignity then takes front and center. No. With something like this happening.

[00:26:55]>> No, absolutely. Right. I mean if you look at uh well mythos is supposed to be a groundbreaking LLM right and >> again like what Venu pointed out it's again about access and control right now open flaw right I mean we've all played with it we we've heard about it we've seen it do wonderful things but that is because openlaw is a consumer product people like you and me using it at home and we are giving it access to control our computers to to control our terminals and as a result it can infiltrate a lot of things, a lot of folders and files and therefore your passwords and therefore your >> sure >> downstream uh assets >> because of that permission that we have given it as retail consumers as users.

[00:27:38]Now when you take the context of an enterprise one you don't have that kind of access you don't give that kind of control right so mythos one being a state-of-the-art cloud-based model will be used very restrictively now what made news was that yes >> um mythos was unleashed on the codebase of Microsoft and a couple of other softwares right >> um >> that's again a question of access you give it access to the codebase and say figure it out uh obviously right when you've got your 15y old code uh being modified as you go along then yes you're going to find vulnerabilities right so it's again about control and access you don't give it access that's again a case of sovereign AI >> fair point makes sense >> the good thing is last point Jensen is being championing this cost so long story short sovereign AI will be the sharp end of the stick >> to make decisions rest everything has to be managed we believe 27 28 That would be the main play. That would be the play. From country level, it has gone into an enterprise level. From enterprise level, it will go into the functions level.

[00:28:46]>> That's all it is.

[00:28:48]>> Well put Venu. Honestly, very well put.

[00:28:51]>> So, as far as the next aspect of things is concerned again, I'd want to start with you Venu. Autonomous workflows, right? Which we are actually seeing AI run entire campaigns from audience selection to delivery without human approval at each step. And a lot of organizations are already live with this, right? So which parts of marketing get there first?

[00:29:13]>> And what still requires a human in the room? You'd mentioned the room is now taking the decision, but elaborate on that.

[00:29:22]>> You know what uh there's lot of uh stuff people there's a smart talk crap in this space to be honest. M >> uh uh there's a lot of people claiming that agentic is done and implemented Sudhi but we work with last year we worked with uh 380 largest brands in India that's our portfolio >> wonderful congratulations to the both of you honestly >> you name you name any brand we have tested largest brands right >> agentic marketing use cases is still a long way to go okay >> there are very small elements in that which have made agentic What I mean by that is customer segmentation, >> sure, >> research work, um, uh, persona, category mapping, social first, lot of those stuff have become agentic, but there's nothing agentic in it, right? There's no intervention. It's just pulling, pulling, pulling.

[00:30:17]>> The the the the agentic in full form on marketing use cases in our experience is very early, >> very very early. the platforms are becoming very very stronger. There is still 80% of the case there is human interventions in the play.

[00:30:36]>> Yeah. Yeah. But it will sound like it's agentic >> but agent human in the play is there. So that's >> it could be just a ploy also to sound cooler right is it's agentic.

[00:30:48]>> That's that's what the market reality is. I it's it's that's what we see.

[00:30:53]Okay. But the velocity of that push the reduction of human in the loop >> that is the ROC is going crazy on that si that's why we see that like probably in 6 to 12 months that is when it becomes truly truly agentic right that's >> that's about it that's the window you're looking at venu 6 to 12 months >> for sure for sure in 6 to 12 months whoever is ready who are able to make this shift they're there because now meta announced direct MCP with claw Correct.

[00:31:23]>> I think a >> performance agencies are 80% debt >> if you truly look at it. So >> imagine that happening right from Google to everybody. So I think whoever is ready to reshape this hence brings the larger question of not AI but how are you thinking marketing as an organization in 2028 29.

[00:31:45]>> Whoever thinks that are the guys who will win not agentic not use cases anymore. Evolution is the name of the game. Angut, I'd like to ask you this.

[00:31:56]Google's reported Remy project describes a 24 by7 personal agent that monitors, acts, and executes on behalf of users.

[00:32:04]Fair.

[00:32:05]>> Mhm.

[00:32:05]>> So, if consumers delegate decisions to AI agents, how do brands prepare for a world where their marketing reaches a machine and not a person? And that's why I wanted to come to you instead of >> they probably use a machine as well.

[00:32:23]>> No, I think that's the reality of it, right? I mean, uh the the the brain that powers these machines is an LM, right?

[00:32:30]Uh the LM triggers actions. Uh well, in the case through Remy or through open clock, right now marketing targeting also we'll have to evaluate uh will also evolve sorry on how uh you talk to these machines, right? And the best way to do that is to use another machine.

[00:32:46]>> You use common uh a common denominator which is the LLM which is the intelligence that comes from uh you know uh the very same LLM that is powering the purchase behaviors.

[00:32:57]>> Uh the same machines also learn about you right. So when you are using uh cloud-based LLMs they are actually responding to you in a manner that is suited to you right. So the LM has or or your your not the owner of the LM but the company that owns the LM has enough data about you >> and as a result of that they can target you damn that good. It's >> well sorry I have I keep adding 10 seconds that's all. No, please. Those are the those are the magic moments.

[00:33:31]>> All all as marketeers we have been taught to optimize for human attention, right? So that's about it. That's what we all uh live and die for. Uh optimize for optimize for time, optimize for emotional resonance and everything. Now when I remove the intelligence uh uh part of it, now I my brand agent is talking to Sudi's agent.

[00:33:56]>> Okay.

[00:33:57]>> I still don't have one but yeah. Anyway, >> I'm not dying to I'm not designing this agent to take your attention. I'm designing this agent to take your intelligence and judgment out.

[00:34:10]>> Okay. That means me and my brand agent should make better judgment than only focusing on attention. That's the big shift. So then comes everything right.

[00:34:20]If I my brand agentic AI has to have judgment. It needs to know me my brand.

[00:34:26]It needs to be machine readable. Then comes the tech all of the grand things, right? How do I teach my AI the judgment so that it talks to Sudi's AI and presents that judgments and get him into the brand house not by the way of attention by through judgment that's where we are going to head to >> I'd now like to give you 10 seconds to add something to that >> uh I think what's what's I think what's important is again understanding the uh the outcomes that you get from from these machines, right?

[00:35:03]>> Uh the fact that they actually are delivering results, >> right? They're they're learning so much about me that when I use these machines, I mean AI to do my work, I'm so efficient. I'm so quick and I'm so prompt. I am taking that benefit of that, right? And on the other side, uh marketeers who are using the same machines to target me are also getting those >> correct >> outcomes, right? Because >> I am hearing things I want to hear. I'm seeing things that I want to hear and it's making me more more you know more increases my affinity towards those those conversations. Sure.

[00:35:36]>> Uh having said that right this is all consumer AI right this is >> you know I'm being targeted um at the same time that doesn't happen the compliance wall that enterprises create will protect you as well also not from just just the the targeting but how you use it as well and then that's the that's the that's my 10 seconds on on enterprise compliance uh that can create a little bit of a wall between between those uh those uh forces.

[00:36:07]>> I love these 10 seconds balance.

[00:36:11]>> Of course. Um I I would like to now come to a point gentlemen where we need to obviously uh sum up whatever we have discussed because there is a certain amount of a positive of time but I have honestly there's so much more I would like to talk to you about but the fact is that uh we want to get you back for this because this is honestly something which is touching a lot of people's lives in a very real way >> and I want you to leave us all with this thought till we bring you back and This I want to start with Venu. Sovereign AI initiatives are already underway globally. Venu EU, Gulf states, Southeast Asia and you know the governments and hyperscalers are building their own versions all simultaneously.

[00:36:57]>> So what do you feel wenu and then I'll come to Angut with the same. What real traction exists outside of India today and how does this compete with what is already being built elsewhere?

[00:37:12]It's the sound AI from an infrastructure perspective from the expertise perspective is built by the big boys.

[00:37:21]Okay, it's all over, right? So those are all acting the infrastructure is acting like a tailwind for us. Very simple.

[00:37:30]>> Now when you add the functional depth of marketing into it, Sudhi, >> the whole world is a market.

[00:37:35]>> Okay.

[00:37:36]>> Which means that most of the global sovereign AI or I will keep going back is horizontal.

[00:37:44]>> Anybody who attaches the vertical expertise of it is the plane. So specific functions, specific workflows, specific enterprise solving if we do as I said as the example implementing in the largest Indian bank gives us the right to implement in the largest Saudi bank to a largest bank in US that will never never existed for us.

[00:38:10]>> True.

[00:38:10]>> Okay. Now it's there. So I keep calling Sonai AI as an horizontal and the T of the vertical comes from the functional.

[00:38:20]That's where you take the tailwinds and run towards it. Very few people are going super deep into that with this TX, right? That's that's that's my uh key giveaway.

[00:38:33]>> That's the nail on the head. What about you Ang? Your final thoughts as far as this is concerned.

[00:38:38]>> Yeah. Sovereign AI in different markets you know have slightly different interpretations. I think in the US sovereign AI basically see the US market is actually governed by the hyperscalers.

[00:38:50]>> Yes.

[00:38:50]>> Right. The Googles, the Azur, the So for them >> the the popularized version of sovereign AI is within my patriotic borders.

[00:38:59]Right. So anything that stays in that country works. It doesn't matter if it's in your organization or not. It doesn't matter if you're sending across or not.

[00:39:05]It's in >> the US. So it's true that's the definition and you have to work within those definitions right >> the Middle East on the other hand uh of course given recent uh events right that damaged a lot of those data centers now what sovereignity is yes absolutely right and that has required or moved sovereignity within the four walls of the enterprise >> right and it also begs to question who should own these data centers right should they be local companies or should they be companies from outside?

[00:39:39]>> Now, India is actually very interesting, right? Because one from an infrastructure perspective, we've only got about 40,000 GPUs as opposed to the million that are sitting in the US and China, right? So, we >> of course encouragingly there are more and there is an entire method which is that by the end of 2026 1 million >> that's the plan.

[00:39:57]>> Yeah. No, that's that's what we need, right? I mean, that is that is mandatory for true sovereignity across all the layers, right? Correct.

[00:40:05]>> But the truth is our foundations are you know in terms of infrastructure are not as strong as the US or China's.

[00:40:12]>> Fair point.

[00:40:13]>> We are not leaders in frontier models either.

[00:40:16]>> So where's the opportunity right is the application layer.

[00:40:19]>> Own your IP, own your process, >> own everything that you built your company on which is human capital and knowledge, right? Use that to build out solutions that actually impress, >> right? And be a leader in that. And we've done this for years, right? I mean we were IT exportal to the world right >> uh >> we have we have it right and I think let's you know we grab the the the cover the hands and make sure that you know we are leader on one of those layers right I think the application layer marketing AI in a box XY Z AI in a box is where we can actually make a mark >> wonderful well honestly with uh the way you guys are leading the charge on this it's been truly of Of course fascinating having this conversation with you both Venu and Angad and wishing you all of the best and may that list of 380 customers that Venu mentioned keep growing. Honestly, we wish you all the positive energy moving forward because you've been able to simplify a lot of the things and you've been able to kind of bring in a larger audience into this conversation to kind of at some level or the other give them a thought or a or a way to understand that you know what they are also a part of this. It is impacting their lives but it's not through jargon you're simplifying.

[00:41:35]>> Thank you so much.

[00:41:36]>> Absolutely. So we thank you for having us. It was just a pleasure having a chat with you. Thank you.

[00:41:42]>> Wonderful. Well, there you have it ladies and gentlemen. Uh Venu and Angad we'll of course be bringing them back because this is a very ongoing dynamic conversation and we'd want to keep uh understanding how things progress. But till then I would please request you to please share your uh comments uh and your thoughts in the comments below.

[00:42:01]This ladies and gentlemen is front page by the IM network. Like, share, subscribe and always remember think AI.

[00:42:07]I think yeah I am.