Studiotalks | Enterprise AI Infrastructure & Data Sovereignty with NxtGen | ET Edge CIO&Leader

0:13 Welcome to ET CIO Leader. Today we have with us Chandrashekhar CDTO at ONGC and Abhishek Singh principal engineer at NextGen. Welcome, sir. Data sovereignty residency used to be a compliance checkbox. With agentic AI now processing sensitive enterprise and customer data continuously, has sovereignty become a board-level infrastructure decision rather than a legal one? There's a real trade-off between the scale and maturity of global hyperscalers and the control that comes with sovereign India-based infrastructure. How are you thinking about the balance for your AI workloads today?

0:50 >> We value sovereignty for our data and then we believe that data has to reside within the country. When as far as now with the new laws that are coming in, I mean, it's becoming even legal one, but more of it is sovereignty issue for us than as far as our data is concerned. We expect it to be within the country for all our applications. That's what we are looking at then. All I wouldn't say all, but most of the applications we expect it to reside within the country. It's not just the data remaining here. We also expect the control plane or the metadata or mean what some of them claim that only a few data we send it outside. Even that we are not doing it.

1:34 We want everything to be done within the country. Complete sovereignty for the data as such. That's what we are looking at. >> Exactly like sir said, right? It's not just about data. It's also about the technology that we use and the operations that are supporting the day-to-day. Essentially, you know, we want to keep control of all of these aspects. Of course, there is compliance. You know, there's legal, there is controls that need to be baked in into the infrastructure and the applications.

2:04 You know, our job at NextGen, in fact, you know, anyone providing services in the area is to make sure that we make it easy for our customers and partners to go through that journey. At the end of the day, everyone, you know, who's in this space is either looking at, you know, is presently in some form of a hybrid solution. Going ahead, measures could be taken to, you know, control the percentage of what runs where. And we sort of play a role in, you know, assisting in that transformation and making it easier to go ahead.

2:33 >> Okay. And there's a real trade-off between the scale and maturity of global hyperscalers and that control that comes with sovereign India-based infrastructure. So, how are you thinking about the balance for your AI workloads today? >> Actually, the industry way we work when actually not much of business is there for the local players to develop all these things. When we look at the scientific applications, when if I talk some generic things like maybe equipment maintenance and predictive sort of thing, then we have a lot of talent within the country and we have I mean that maturity is very much there. But when I talk something like the data processing, seismic data processing or the interpretation, where I mean there are only three, four companies within the company. One is our company, one is Oil India, the other is K&R, and Reliance. I mean, a few companies are there, handful of companies. So, we don't find much of the local players gaining that sort of maturity.

3:29 If they gain that maturity, then we'll be we'll be very happy to work with the local ones, but then we need that sort of maturity and that sort of capabilities also to be built. That's what we are looking at. >> Right. I I think that that's an excellent charter for us. The purpose of these conversations and essentially the purpose of the direction that we set in the industry comes from leaders like sir, right? The idea is that we are perhaps catering to a majority right now, but there are amazing breakthroughs happening in niche domains that are, you know, scientific and science-driven. And that's a charter for all of us as well to make sure that the right kind of managed services are available for our partners so that, you know, science and research and development is also given as much importance as perhaps IT system, you know, our CRM, ERP, etc.

4:21 So, you know, this is something that we are putting a lot of work in. In fact, global partners are also allowing, you know, us to run this within our country's infrastructure footprint and make it available for our partners to leverage. It's just that the right kind of feedback cycle needs to be put in so that we can take steps in the right direction and bring it up to that mark. >> AI inference costs behave very differently from traditional compute.

4:45 What surprised you most about the real cost of running AI workloads once you had moved past the pilot stage? >> Yeah, pilot stage when scaling it's really difficult. I mean, that's going to be a big thing. I And when you scale it, can you scale it at the same proportional cost or not? That's another big thing, which we haven't reached that stage yet. We have run several pilots successfully. We got some good results and then we are looking at scaling them up, but then the infrastructure is playing a bit of issue for us.

5:15 But then the real cost of scaling it up, that's still the thing which we need to assess. But then definitely, unless you get that sort of ROI and then the value back to the company. It will definitely be a difficult decision to take when so much of investing and then whether you get the real benefit out of it or not. These are some things which we need to keep assessing and thinking. And probably we'll look at only those solutions which will which we are sure that we are going to get the value.

5:52 That will definitely look at that, but then for the bigger things mean where it's going to be too costly, we have to go very cautiously and that's definitely a thing we keep assessing every now and then. >> If you had to make one infrastructure investment decision today that you'd be judged 18 months from now, capacity, sovereignty or partnership model, which would it be and why? >> Capacity to open as I said mean some capacity we are already building where we think that that is going to serve us for the complete life cycle.

6:23 It's not that we want to invest where mean just for the sake of building up the capacity. That's definitely not the thing and sovereignty is definitely a no go for us. If it's not sovereign, at least for most of most of our applications, we may not be even thinking about that. Sovereignty is definitely a compulsory requirement for us. But we are thinking, but then there are other issues as well mean where we can take a decision about that.

6:52 But definitely when we get somebody who can partner with us mean where it can benefit both of us. Probably we might look at it, but we cannot say for sure that mean for what purpose we may have partner partner should be in a niche area where mean we get benefit and then we also see that mean other people or other mean it's not a generic available sort of thing. It should be definitely for a niche scientific thing mean where we can partner and then get the benefit.

7:25 That's why we are looking at. >> No, I mean I think you know, bundling the two things that sir said, right? It's about if if sovereign is the focus, we need to make sure that whatever we're running on that sovereign infrastructure is giving us a very good price to performance ratio as well, right? If there's a certain workload, it needs to be done in the most optimum way and we need to get the maximum out of that infrastructure and as well, you know, from my experience viewpoint as well.

7:52 So, when it comes to a lot of generative AI is and predictive AI as well, I feel that the the tooling and the frameworks and the industry support is right up there. I would say, you know, cutting edge and very niche research and development is happening. So, it becomes a unique challenge and an interesting one not just for sir, but for the industry supporting sir as well to solve and make sure that if we're putting that compliance pressure, right?

8:19 It It shouldn't just be that, "Okay, I have some sovereign infrastructure and it's not really doing good work for me." We need to bring that up to the level where in it's, you know, a like for like or giving much better value. And, you know, that's that's that's the possible synergy that we could have, you know, going ahead as well. So, I think, you know, sir is is absolutely on the point over there. It's not just about, you know, scaling of the infra. It needs to be for a purpose. Partnerships also need to be for a purpose and if that's, you know, a case where in a partner and the customer are able to bring equal to the table, right? So.

8:52 >> So, thank you, sir. Thank you, Abhishek for this wonderful conversation. >>