Indias Roadmap to an AGI-Enabled Future

Researchers, founders and policy makers. At Chariot, we are proud to be one of the companies mandated to build frontier models for India under the NDIA mission to build sovereign frontier models for the country. But as we embark on this journey, we must recognize a fundamental truth. Building true frontier intelligence from India is a monumental ecosystem play. We cannot simply import models and talents, run them on borrowed infrastructure, and call them our own. If we want to solve India -scale problems at population scale, we must own the power, the hardware, and the talent and the research that drives them. That is the thesis of today’s session, India’s Path to an AGI -Enabling Ecosystem, to bridge the gap between energy, infrastructure, and research.

We have brought together the absolute pioneers of this field. Before we begin, let me quickly share our roadmap for the next one hour. We will start by inviting each of our distinguished speakers to share opening remarks on their respective domains. After that, we will dive into the topic of the next one hour. We will then move into a panel discussion. And then finally we will open the floor for your questions. To guide through this we have assembled the absolute pioneers of energy, compute and research pillars. Today we are joined by Shri Ghanshyam Prasad, Chairperson of Central Electricity Authority. Mr. Shri Tarun Dua , Founder and Managing Director of E2E Networks. Professor Jayadeva, GSV Chair in Formal HOD of Electrical Engineering at IIT Delhi.

And finally my co -founder at Chariot, Mr. Parth Sarthi. To build this ecosystem from ground up starting with the very power that makes this revolution possible, energy. To speak on the sheer scale of this transition and to help us answer critical questions such as what we expect AI’s true energy demand in the country to be and how are we preparing and modelling our national grid to meet it, I would like to introduce a true veteran of the power sector, Shri Ghanshyam Prasad ji. Shri Ghanshyam Prasad ji, presently holds the post of Chairperson of Central Electricity Authority. With an illustrious career spanning over 35 years, his expertise covers generation, transmission and power market development. Having served as part of the G20 Energy Transition Working Group and the first Executive Director of BIMSEC Energy Centre, his global perspective and visionary leadership are ensuring our grid is ready for the AI era.

Please join me in welcoming Shri Ghanshyam Prasad to the dais for his opening remarks.

I think the speaker wanted me to speak on some of the key challenges that is likely to happen in the era when we are transiting from the present situation to AI -enabled or AI -driven power system. We all know that the moment we talk about AI, which means that it is supposed to be data -intensive and it is to be a power juggler, and we are talking about the data centers which will try to enable it as we go along. In India, we are now transiting from smaller data centers. Earlier, we used to have a small data center. We have 10 megawatt, 50 megawatt data centers to now gigawatt scale data centers at many places. particularly in Mumbai, Vizag, Chennai and all other places.

So far, we have a visibility of around 16 gigawatt of such data centers coming in across India. The challenge remains a few, particularly if I see from the perspective of serving a large load which earlier we thought that it is going to be almost like a constant load, but practically it is not. And if it is not, then how do we manage such type of a variable load? So far, we were struggling with only variable sources, that is solar, wind, etc. Now, we are going to have something from the load side as well. A large load getting integrated into the DESCOM system and which is also going to be used for the solar system. To have a nature which is going to be variable.

second is the kind of reliability that is that it demands into the system the reliability is we talk about n plus one plus one now which means that the same data center will have to be supplied from two different sources and they have to be slightly differently located as well second is even if the supply fails then it has to be backed up simultaneously by two I’ll say two steps that is DG sets and each DG set will have to be backed up by another DG sets so we have four layer almost four layer of security of supply it’s definitely challenging for a country like India which is now expanding and growing to provide such kind of a reliability but still we are geared to meet this kind of a challenge.

Some of you who have not seen Adani data centers, I’ll request you. It’s very close to Delhi. It’s in Noida, which is coming up. That is 50 megawatt data center being built by Adani and would be operated by Google. 10 megawatt has already been commissioned and rest 40 megawatt is in the pipeline. You can see the structure that is coming and the kind of challenges it is facing. But still, I must congratulate UP Discom who have been able to provide this kind of a reliable supply to that data centers. But this is 50 megawatt. But very soon you will find a data center coming in Mumbai area, which will be of a thousand megawatt. YJ, which may be even more than thousand megawatt.

Thousand megawatt. But the moment I say it. means the supply that will be required to this particular center will be at least 1 .7 times. That’s the near thumb rule. It may require I think sir is saying that it will be required 2 times at least. But data center which I visited has been designed for 1 .7 times of the data center’s capacity. So the challenge is first is how do I maintain a variable load? How do I meet the N plus 1 plus 1 criteria of supply? Some of you researchers who are sitting here probably must be aware about these kind of things before we try to design this kind of a thing. What is further more going to be more challenging is because these data centers are also planning to go green.

That means had they taken a mixed supply probably the challenge of DISCOM would have been slightly lesser. But if you want to classify yourself exactly green data centers, then that means I need to ensure you that only green power flows into your data center, which means a combination of solar, wind, battery, hydro, hydro pump storage, or any such type of a combination, which we’ll be able to ensure to you. And that means I need to ensure a transmission line from such sources to your place so that at least there is no interruptions in the supply of green power being provided to you. But let me assure you that India is geared up for that kind of a challenge because we have started the journey of energy transition somewhere in 2011 or 2010, wherein we started.

We started with a meager figure of somewhere around 2 gigawatt. Now we are more than 50 % in terms of renewable in the country. surpassing 250 gigawatt and which has majority share coming from the solar and then from wind and storage is now kicking into the system. Last year, we surpassed 30 gigawatt in a single year. This year, in just 10 months, starting from April till January, we have already crossed 40 ,000 megawatt, which means that probably in this particular year, we will have more than 50 gigawatt coming in in a single year. So even if the data centers or the AI -driven systems demand green power, I think the country is geared up to that kind of a challenge. Further, what we are suggesting to the data centers is, please try to have a diversity.

Diversity means don’t have at a single location. Try to be as close to REOs. Try locations as possible. slightly away from the main town and diversified locations. So far we have two landing points in the country. It is Mumbai and Chennai, but we are trying to diversify that as well. So we can have multiple landing points in the country so that the data centers can come at multiple locations and so that at least the challenges of the DISCOMs will get diversified. Coming back to the further stability, and since I have been asked for the international scenario as well, so India is also connected to Nepal, Bhutan, Bangladesh, and to some extent to Myanmar. And we are also promoting to very soon connect with UAE and Saudi Arabia and Singapore.

And even Sri Lanka. So if the moment we are going towards the other countries, that means we will try to have both. not only powered, but we’ll try to have the communication network as well. And there are situations which is emerging that maybe these countries will like to have the data centers in India and get supported through that kind of a systems. But all these definitely as we go along will require huge amount of balancing power and storage capacity. So right now we are depending on two major sources. Either it is hydro pump storage or coming from the battery energy storage systems. Recently we launched a report which gives us a confidence that we’ll try to have somewhere around 100 gigawatt of hydro pump storage coming within next 10 years, which is going to be a very good support to meeting or meeting the 24 hours supply to, or the supply to these data centers.

So we’re going to have to wait and see what happens. similarly because we have to cut down our carbon footprint so we are also trying to have a roadmap for 100 gigawatt of nuclear which is targeted to 2047 but there is a visibility even as of now that we go in a fleet mode and we will be trying to achieve somewhere around 22 gigawatt by 2032 or 2034 and then moving up further with more technology kicking in and more expertise being gained particularly from the private sectors and all other sectors so holistically if I see we have huge amount of challenges but to meet those kind of a challenge we have good mix of resources in the country and the country is geared up to meet those kind of challenges the country is also able to make the transmission lines in a record time as compared to anywhere in the world if you see we are able to provide connectivity in 24 months to 36 months time frame in the country as compared to anywhere if you take US etc they take around 10 years to give grant a connectivity that’s the kind of waiting list that they have if you take even European countries they also take more than 5 years for building the transmission lines so at the end I will only say that we are totally geared up for any kind of innovations all the youngsters are welcome from across the world to set up their systems here and I can assure you that the country is fully equipped and fully geared up to support you thank you so much applause

Thank you sir for setting the stage with those vital insights on our energy readiness I just quickly ask the panel to get together for a group photograph applause Thank you. Thank you. Thank you. By delivering enterprise -grade cloud infrastructure at significantly lower costs, he is democratizing AI and empowering over 10 ,000 innovators with advanced H100, H200, and B200 GPU infrastructure. His work is building the foundational infrastructure that enables our sovereign AI ambitions. I would now like to invite Tarun to the dais for his opening

Yeah, thanks, SUvrat. So, like, thank you for this opportunity to be a part of this August panel. So, building infrastructure is something we have been doing since 2009. So, 2009, when we began our journey as E2E Networks, like, most of the… So, there is an incident even 10 years before that. So, when… Yeah. We had a startup plan, like I think somewhere around 2006 era or something like that, or even before that. So, we were discussing, three or four friends who were working in the IT industry, oh, we are going to make a website, and this is what the website is going to do, and this is how the website will make money. So, the fourth guy asked the question, but who is paying for building the website?

So, the idea was that, like, it is always someone in the West who is outsourcing the development of the website to you guys, and you are building the website for them, not for India. So, that was the era, once upon a time in India, where we used to do everything for the world, not for ourselves. So, the second stage was when we started doing things for ourselves. Now, the third stage is what we are doing today as a country. We are saying that, like, Like not only are we going to do things for ourselves, we are going to do things for the world. And we become the innovation hub and the innovation capital of building cloud infrastructure for the world.

So with that, I would like to kind of like once again hand over the stage back to Suvrat.

Thank you so much, Tarun, for sharing how E2E got started and the vision behind starting E2E networks. But raw compute and energy are just untapped potential without human ingenuity and the mathematical rigor to harness them. It is my deep honor to introduce Professor Jayadeva, the GSV Chair Professor and former Head of Department of Electrical Engineering at IIT Delhi. An alumnus of the same department, Professor Jayadeva. Jayadeva is a trailblazer whose internationally recognized work bridges theoretical mathematics and practical AI. His group was amongst the first to fabricate an SVM -based AD converter on chip. His recent work on minimal complexity machines provides astounding model size reductions of up to 300 times. His contributions to optimization and machine learning are vital to building highly efficient indigenous models.

Sir, we would look forward to your opening thoughts on how India can contribute to the research and talent pipeline for building artificial intelligence models from the country.

VLSI and as it turns out there are a host of issues that need if you ask me serious discussion and brainstorming. Primary among them is the issue of manpower. The entire development at one time if you remember Silicon Valley the word IC used to actually jokingly be referred to as Indians and Chinese. So the intellectual innovation that came to build Silicon Valley and most of the entities there that are known today came out from Indian universities, came out from the IITs and all a few decades back. Question is what would it take for example to build that same kind of ecosystem here and you need to have a critical mass of very smart researchers doing work within the country.

And we have to supply the reasons for them wanting to do that. first of that first amongst that is what’s the career connect for a student wanting to pursue his or her PhD or any other research degree for that matter at a university here most of our so I should just put a disclaimer all of my comments are my own personal comments and not representative of IIT Delhi before I continue further but if you ask a student today well a lot of them come to us for recommendation letters and in most cases the first choice wanting to do a research degree would be a university outside that has to change and it is changing but it’s changing slowly what are the reasons for that think of a student who decides they are excited by research wants to do it within the country wants to do it at university here what’s the career after that that connect is directly visible if you look at a research lab you look at a university research lab in the US or elsewhere that connect is missing in most places not because there are no industry driven projects or so on but the nature of those projects is different in many of the successful examples that I can discuss when we have that panel itself the instances are where the university has embedded their researchers within let’s say university along with other students along with other researchers and those who are working for their PhD are already for example employees working in the university environment the scalability of research is very difficult within the industry it’s expensive to explore ideas because maybe out of 10 or even 50 ideas that you explore it’s very difficult to explore one becomes successful and ends up returning revenue to you it is far cheaper to do that exploration within the university environment we have to find models that allow universities and industry to work together but also to find ways so that the biggest bottleneck of IPR sharing which is really the bone of contention or really the key point in most MOUs that you sign this particular aspect is handled more seamlessly and in a simplified fashion the other difficulty is of course with regard to the way the entire ecosystem is configured there is a deterrent amongst many Indian parents from their parents to the children in a sense why don’t you finish your current degree first join a job and worry about a higher degree or PhD later the difficulty with research is it is best done when people are not doing the research when people are in their prime when they are overflowing with new ideas Because once they’re in a job, they get saddled with other responsibilities, you know, familial, others, and so on.

And it never ends up being the same story, let’s say, a few years down the line. This particularly hits women candidates harder because there’s also pressure, you know, although I don’t want to make it a generic statement, but there’s a pressure amongst many of them from their parents to get settled early. So we find, as a consequence, fewer women in research, in engineering research, let’s say, particularly, as compared, let’s say, to male candidates. And finally, the incentive in terms of what people get if they join a research career and eventually join industry or elsewhere, that incentive needs to be made far sharper and far clearer today. Okay. If a student joins an industry today after their undergraduate…

Thank you. degree and works there for a while. Many of them continue doing research in the industrial setting. But as I said, exploration is costly within the industry itself. And so unless the student has a clear -cut motivation to do outstanding research early on so that the industry or whatever career option offers them a significant incentive to do that, I think the scalability will be missing. So I’ll stop. I think I probably have taken more time than I should have, but we can discuss.

Thank you, Professor. I think your vision for preparing the next generation of researchers and what it takes to incentivize them is exactly what this ecosystem needs to thrive. Finally, I’d like to introduce my co -founder at Chariot, Mr. Parth Sarthi. Parth Sarthi went to Stanford to do his PhD in engineering and he was a professor at Stanford. He did his undergraduate and master’s degree in computer science. and more recently was working at the Google Gemini DeepMind team on the DeepThink project. He was the inventor of Raptor, which is currently the state -of -the -art technique in retrieval augmented generation based on which all retrieval augmented generation pipelines today operate on. I’d love for Parth to speak on what it takes to build sovereign frontier models and the differences that he has seen building these models out in the West versus what it takes to build these models from India.

Thank you.

Thank you. India under the India mission has 38 ,000 I think scaling to more than 50 ,000 GPUs which is so much more than you know what we had a year ago two years ago thanks to the India mission and I’m sure the scaling up will continue have many more GPUs now but if you look at the West you know their companies with much more GPUs with deals for many Blackwell and ruin chips coming in right so I was at Google DeepMind I worked there on Gemini deep think the reasoning capabilities of one of the most research resource which yeah labs in the world and and that this number is of GPUs is going to go up but why does any of this matter right why is there a GPU race after all why can’t we just write better algorithms and make better models so the answer to this is in my opinion one of the most important empirical discoveries in the history of computer science is scaling laws so the GPT papers were impressive you know GPT -2 could write paragraphs GPT -3 could write essays they were really good work but the GPT papers were the tinder.

The match, the thing that actually started this whole AI revolution and lived in the entire industry were scaling laws. So in January of 2020 Jared Camplin and some colleagues at OpenAI including Dario who went down to start Anthropic published a paper called Scaling Laws for Neural Language Models and what they found was really simple. So if you take a neural network’s loss, its error rate how wrong it is against the amount of compute used to train on it on a log -log scale you basically get a straight line. A very clean smooth power law. A straight line that spans 7 orders of magnitude. What that means in really simple languages, every time you 10x your compute, your model gets measurably, predictably better.

Not randomly, not sometimes, every single time. The exponent they found was roughly 0 .07 so which means for every doubling of the parameters you see the loss drop by 5%. This sounds small but at a log loss scale across many many doublings if these, you know, these gains compound enormously. GPT -2 to 3 was a 100x increase, 3 to 4 was another 100x, and each jump, you know, produced a leap in capability, right? And then the DeepMind’s Chinchilla paper, which corrected it, said you need to roughly scale your data and compute equally. So the reason why this was so consequential was that, you know, this turned intelligence into an engineering problem, right? Not a science problem. You don’t really need a breakthrough.

You need, you know, more GPUs, more data, more electricity. You need money and whoever has the most money, right? So you could call the race right there. You know, if the scaling laws hold, they have held for five orders of magnitude. So then there’s a spending competition, right? And this was the dark picture. A lot of people would ask me, you know, why are you leaving DeepMind? I come back to India to build against this kind of backdrop. And the reason is this. So about a year and a half ago, something changed. We had the reasoning revolution that hit. You know, there was the O1 model, the O3 models, and they showed there was a difference.

So there was a different way to actually make these models smarter. So, you know, this word reasoning gets thrown out a lot. Let me explain what it is in some simple language. In the old paradigm, you would pre -train these models by making them bigger and training it on more data, which is pre -training. And, you know, the models will see the strillions of text, and at inference time, they would just generate it really fast by, you know, just one at a time with no ability to sort of correct for its mistakes. And these reasoning models, they started working differently. They could, you could give it a problem, a math problem, a coding challenge, a logic puzzle, and you could let it think.

So it would generate a long chain of thought. It would think for a bit, and then it would try an approach, maybe, you know, backtrack to a different approach and eventually, you know, reach a final answer. So this result was a new scaling law and where you could actually, you know, spend more RL training compute. And now we’re even seeing that this new type of RL compute is actually even exceeding the amount of compute spent during pre -training. so this was a reset and if you look at, and let me explain why so if you look at RL training, right the majority of your compute is not actually in gradient impedance it’s not actually in the training, it’s in this models trying different things out in different rollouts, and this is basically inference, and this is this doesn’t really need to happen on your you know, top of the line 100 ,000 GPUs in one building with NVLink and InfiniBand this RL inference, you know the sampling can be synchronous, you can generate asynchronous, so you can generate rollouts on one set of machines, collect them you can make them distributed and so on you can make them run on older GPUs on, across multiple locations and now we have hundreds of, you know, techniques coming out every day to make this work, right and just doing RL is one step, the other part, and this is the, I think the main thing why I do think, you know, India will succeed is environments, RL environments are where majority of the training happen, you know, a math environment has math problems a coding environment has coding problems where the math the model tries, gets feedback, and improves.

And the key observation is that these environments, you know, it can scale with humans and CPUs and not necessarily GPUs. And GPUs are important, but they’re not the most important thing, right? So building a math environment requires mathematicians. Building a coding environment requires software engineers. Building a medical environment, you know, could require doctors defining clinical scenarios. And this is human expertise, right? It scales with people and ordinary compute, which we have a lot of in this country, right? So this is the bet I made. You know, India has 1 .4 billion people. We have domain experts in every field, medicine, law, agriculture, finance, education. We can work in so many languages. We can build environments for problems that a lot of labs in the West don’t even know exist, like agricultural loan assessment in Tamil, legal aid reasoning in Hindi, and so on.

You know, these are problems that affect hundreds of millions of people. Then we can build RL environments for them that don’t exist anywhere in the world, right? But with the, you know, India emission grant, we have a lot of compute to actually build this frontier if we’re smart, smarter about these environments, right? And if you look at India, India is a voice -first country. And that’s why at Charity, we’re building a voice -native speech reasoning model, right? Reasoning over speech for all the reasons I just described are in -train, environment -driven, and in print scale. So I think, you know, the race to AGI sort of has begun. We have the right environments, the right algorithms, the right focus, and this distributed setup.

Now, I think, with the support of a mission that’s already scaling up so many GPUs, I think we can go ahead and

Thank you so much, Parth, for sharing what you think is the roadmap for building intelligence from India. With this, our distinguished speakers now assembled. I think let’s dive straight into the panel discussion. Thank you. Okay So I have a few set of questions that I’ve prepared for all our panelists but people please feel free to interrupt and if somebody can go around with a mic asking questions please do so So I’ll ask my first question to Tarun I think we’ve all spoken about large GPU clusters of how they’re growing in size I would love to understand your perspective of where you see India’s compute requirements are today, where do you forecast them going to be and where do you think the demand for the same is coming from?

Sure So a number of things So like if we just look at the compute requirements of say top 15 or 1000 So If we just look at the compute requirements of say top 1500 or 2000 or 2500 or even 5000 organisations, so are there enough teams that can utilise say 16 to 128 GPUs? Just looking at top 1000 organisations and say that like do they need at least 128 new GPUs every year? I think the answer is most likely yes. More likely the answer is that initially we need 128 GPUs and eventually we are going to use at least 1000 GPUs where there are multiple teams within an organisation trying to solve multiple problems and so it’s not just that GPUs are used only for training and inference, they are also used for data cataloguing, they are also used for like many different types of inference which is like available straight out of the box.

I think it’s a good question. I think it’s a good question. I think it’s a good question. I think it’s a good question. So net net the compute environment required by each of these organizations is going to be of the size of at least 1024 and that’s the representative of like the mid segment and the SME and the higher education and research and like literally there are so many different types of organizations apart from like for -profit companies. So net net if we were just to look at like say thousand organizations wanting 128 GPUs each you’re looking at like India needing at least 128 ,000 GPUs and we are not there yet. So which means that like there is a journey ahead of us in terms of building the infrastructure and having faith and the confidence that yes like India may be lagging maybe 18 months behind the rest of the world but that lag will keep coming down and at some point of time we leapfrog.

Like we did with 5G and 4G. So when that leapfrog happens, those compute requirements would explode even further. So I think it is safe to say that like India is a country with 20 % of world’s population and currently having capacity of processing about 3 % of world’s data will sometime in the future leapfrog to processing not 20 but like maybe 40 -50 % of world’s data by becoming the data center and the compute capital of the world. So those are my thoughts around that.

No, absolutely. Thank you so much, Tarunn, for sharing that. So I think at a bare minimum what you’re saying is like the 128 ,000 GPU infrastructure that we.

That’s today’s requirement just in India alone. And we just don’t serve India alone. Like when we build compute infrastructure, we serve the whole world because this is not a super latency sensitive like a website or a CDN kind of an environment. So reasoning models, they think. And when you add another 200 milliseconds to the thinking process, it does not like really kind of like add a whole lot of latency to what the people are experiencing. So in that sense, we can actually serve the compute for the world. So which means that we can build a lot more than what just India needs alone.

Makes sense. And so that’s an excellent segue to my next question which I’ll direct to Mr. Shri Gansham Prasadji. Sir, when we talk about these kinds of compute infrastructure that is needed for the country, how do you forecast like what the energy consumption of modern day data centers would be compared to our overall energy requirements for the country? And like how does our country, for example, be prepared to meet that over a 12, 24, 36 month time horizon?

See, we have already, as I mentioned in my opening remark, we have already factored it right now demand equivalent to 16 gigawatt which we are projecting for the data centers. But the philosophy of planning we have changed. It’s now in India. And we are trying to upgrade our systems and planning systems every year. It has been made dynamic. Earlier you used to hear something like five year plans, right? Those days have gone. So we are upgrading our transmission every six months, that plans. And the resource adequacy plans is being upgraded every year. And even when I was speaking in Singapore where US and all other regulators were there and they said how are you able to really manage this in six months and one year.

So I said it’s a computing environment that has gone in India and we have really cashed up and we are able to do this. And that has really helped. If you see whatever error we make in the planning process or the projections we are able to correct it within no time. And that has led us to do a course correction immediately whenever we have this kind of plan. Second is the growing demand that the country has right now, which is phenomenal. I will say it is much much higher than any other countries across the world most of the European countries you will find that they are growing at either they are stagnant or growing at 1 % or 1 .5 % or 2 % at the most we are growing at 7 to 8 % and some year we have even grown at rate of around 10 % so meeting that kind of demand unless you are resilient and you are able to do it in real time frame probably you will not be able to sustain that kind of a thing and the kind of further expectations that is there with the customers probably you need to have that kind of jump.

And sir I think like one follow up question which I have actually both to you and Tarun is that like these modern data centers like the energy densities are hitting quite high levels right so one thing that we hear is that do we move data centers close to where the energy generation is happening so when we talk about this new upcoming like data center hubs you talked about you know like sort of Mumbai being one of the hubs for where these data centers are being created but like according to my naive opinion there’s a lot of energy production that’s happening in states like Rajasthan like how do you foresee this environment that you see data centers moving close to where the energy hubs are like would you be interested in building like the center close to these regions I would love to hear both of your perspective on the same.

so I am really looking forward to like the SMR nuclear reactors being made available as quickly as possible in the data center campuses and see nuclear power is again like I am shilling for nuclear power for no reason so nuclear power is like also very reliable so you can actually run it for like all together for like 8 years 10 years now several advantages to that is like you are not transporting on the grid so you don’t have to pay the transit fee which is very very reasonable in India but like again every cost saved is that savings can be passed on to the end customer and similarly you also don’t need diesel generators to be there on site you can just have a slightly larger battery energy storage systems along with nuclear and you can build a data centers of the future so that is something that i’m really looking forward to but i think like it could be like three to five years away so those are my thoughts about like wherever you are putting data centers you can put the power over there as long as there is availability of sufficient amount of land because nuclear power requires like some free land around that facility and another advantage of nuclear power is that once you have set up like some land for nuclear power you can like modularly increase the size so let’s say you start with 220 megawatts then you can add like in chunks of 220 megawatts which is the most dominant design of the smr or like even the bsr designs that are there so that’s what i think about it.

I think what Tarun said is very right, but the visibility that I see in SMR may not be 3 to 4 years. So maybe slightly longer period, I am not very sure about it. And because I have been talking to most of the people who are going to be into the business of nuclear, because so far we have only NPCIL, Nuclear Power Corporation of India Limited. But all others are also slightly apprehensive, that probably that may take slightly longer time. But again, what is required, what he rightly said is you will be requiring a containment zone. And that containment zone vary anywhere between 1 kilometer to 5 kilometers, depending on the capacity that you are going to have in the nuclear space.

That means again you will be moving away from the main crowded places, right? Because you require a containment zone wherein no habitations are allowed. second is you rightly mentioned that we are trying to say that you should go to as close to the resource center as possible because you need green power if you really need green power then you should have that kind of a closest because if you take let’s say if your target is somewhere from Rajasthan or Gujarat we require a huge amount of transmission lines and we are trying to optimize on the transmission system itself so let it be at the generation place and Maharashtra I mean good thing for India is we have 8 to 9 states which are very rich in renewables starting from Gujarat Rajasthan, Maharashtra, Karnataka, Telangana, Andhra I mean so all these and so you have multiple choices it’s not that you have only one choice where you need to put it similarly if you see the IT hubs which is getting created they are also scattered around the country so that and last point I said is we are trying to have multiple landing points again so the moment you have a multiple landing for example for Singapore Vizag or Paradeep or Gopalpur could be another choice so we are looking for an alternative and Singapore probably is likely to be connected with Vizag so similarly for the western side as well so you need to have the diversity of this and that is how you will be able to successfully meet your demand

No, makes sense. Thank you so much for sharing those points of view I will move on sir to Professor Jayadeva when we talk about the talent pipeline for the country we would love to hear your perspective sir on what you think is sort of the undergraduate readiness of our workforce for training and deploying these AI workloads and what is your point of view on a lot of people in our country moving abroad to do higher education or moving abroad to do higher for better work opportunities compared to sort of the PhD education system in the country thank you like what would your perspective be on you know empowering more of our children to sort of continue PhD opportunities to continue grad school opportunities in India versus sort of doing that in other institutions around the world ?

Were actually employees of a firm working full time in the department. This company had stationed them in the department and said, well, work for your PhD, but you have to work on areas or these problems that are relevant. They were, of course, discussing with many other students in the department who were also in that lab. And then, of course, there were professors part of that team. That kind of success story is, you know, I would say rare. And if one finds a way to replicate those examples in numbers, I think the story will change dramatically. It takes a leap of faith. Most HR managers are averse to letting their employees work full time at a university.

Well, if you’re working there, you know, you’re not on site and therefore you’re on some kind of leap. In this case, we created a way so that they could logon. So VPN and work as if they were on site. So it’s kind of. site for themselves. The other problem of course is people have to join research careers early. They have to take that plunge early on. That’s when they are most productive. That’s when they can churn out new ideas quickly. And I think while the government is doing a great deal to make that happen I think we need more examples from the industry trying to do that, trying to bridge that gap. So if that happens in my view, the story will dramatically change.

How do students today look at PhD as a career path right out of college outside of the other opportunities they may have?

So the duration of PhD is the primary deterrent. It’s 5 years. And so there’s a social deterrent as well. I have heard from students you know when they get back home PhD student some neighbor will make a comment well you are still studying is it because you are still at college still at university haven’t got out aren’t in a job so it’s you know that mindset will change in my view only if one you get paid more I mean if they are actually employees let’s say working that changes the fellowships I think are far more lucrative and that can only happen with industry help in my view and but there is a there is a via media there is a path in between we have something called MS research which is like a research degree that takes about 2 years plus numbers there have actually tripled in the last 3 to 4 years so number of PhD enrollments I would say is now static it dropped after COVID but in this MS research degree those numbers have actually and you can get a job and you can get a job and you can get a job rippled in the last 3 years I am saying for our department so I think we have to find you know we have to really brainstorm I think that that dialogue hasn’t happened in sufficient measure to be able to answer your question.

If

f I may supplement I think professor is saying what is the practical case but government is slightly thinking in a different manner now and you must have heard about ANRF that is Anusandan foundation that has been created with an outlay of 1 lakh crore rupees and this is going to be across the country across all the segments all the sectors which will be almost under the principal scientific advisor of the country very recently we also had a meeting with him and you And very recently you must have seen that we have got something like 20 ,000 crores under CCUS, carbon capture and utilization in storage sequestration. So these are some of the projects which are now being identified.

What are the gaps that India has in terms of technological things which other countries have or can we surpass them? So with that objective, this fund has been created and it is likely that the industry and this kind of an organization and even what we are thinking in the power sector is can we have a university or maybe a cluster of such this thing. Already one has been experimented in Gandhinagar which is doing a good job. That all those people who are trying to do something. Can they do some kind of innovations? Can they be supported through some kind of a fund? And then… the industry takes over. So the gap that earlier used to be there that a PhD he does a paper or a professor he does a paper or even his promotion is linked to the paper publication.

So that kind of a situation will have to be slightly modified and you need to really take whatever you do, whatever PhD that you do or whatever research you do, it has to be taken forward from there so that what we are thinking is that it’s a concept to commercialization. So you have to take it to that level and then only it has to flow. Very recently we also had a good competition of AI use case in power sector. I think only two months back and we have identified few companies who are really trying to have that kind of an ideas and we have already assigned them some tasks that okay you do it on a nomination basis.

So that’s the kind of you find that. So there are a lot of good changes. that is that the change in mindset of the government and trying to support this kind of activities that is going to happen.

Sir I would like to add something over here. So these are great ideas that like research should be promoted and supported in India. Now academia does a very good job of identifying pure problems which need to be solved which advances the human knowledge. We in industry see the build versus buy decisions like almost every day. And also we kind of like look at all the road maps of okay what needs to be done and what amount of time. So give you a few examples like basically like if you just look at say things like optoelectronic networks co -packaging of optics with electronics. So those kind of problems are very well known. So to go from 100 Gbps to 1 .6 Gbps there is a certain time frame in which it has to be done.

And at a certain volume of production that it has to be done. So, which means that resources have to be deployed in a manner that it produces goal directed research in a certain time frame. So, what is considered as like a good outcome is something that we in industry can help define but most of the time we don’t always have the kind of money to deploy behind those goal directed research and also we do not have our own use cases for kind of like selling out that much to be able to support that volume of research. So, that’s my suggestion that task people like us who make build versus buy decisions to at least create the roadmaps that okay this would be good to have if we can do it in this much time frame.

If we don’t do it in this much time frame somebody else in the world will go and do it. So, that is something we can help with.

Yeah, absolutely. And this is the basic idea. This is the basic idea with which we are trying to have this. Just I’ll give you one example. In fact, we are facing huge amount of challenge in research. It’s what you see right now in the country. we have only two companies in the world and they are really taking us on ride in terms of supply chain in terms of prices etc etc then we said ok nothing doing let’s can we have our own industry coming up in India so we have now lined up L &T and Power Grid Corporation of India both of them are contributing 300 crores each to go in for that the gaps so you will find this kind of situation we have already tried to identify something like 76 elements in the power sector which needs immediate attention so you need to go aggressive now on this kind of thing similarly other sectors as well IT sectors, METI is trying to do in mining areas, in critical mineral area so you will find all the ministries have now waken up to take up this kind of a challenge.

Thank you.

I just wanted to react in a different way to some of these comments so it’s not always that you know research needs to be abnegated issue. A lot of research is applied, a lot of research that happens within universities, IIT and so on. A lot of it actually is with industry. But more often than not, the industry funded projects tend to be kind of at arm’s length. It’s like kind of saying, look, here is the problem and if you can find a solution. Sir, we need both types of research. If we only do goal directed, we will never innovate really well. No, I just wanted to say it differently. So, the point I was trying to emphasize is not about either necessarily short term or long term or medium term.

You need to have a mix of all three. Certainly new ideas come forth at all possible levels. Okay. The difference that you know eventually an idea makes is well ideas don’t make money companies make money or you know organizations make money so the key is translation it is difficult to create an ecosystem within a university that’s efficient and let’s say I would just say efficient at translation on the industry side translation is much simpler they’re geared up for production as an example if you ask a student to write production level code it’s not going to happen it’s not feasible and that’s why one has to rethink the nature of this partnership it’s not about funding it’s about trying to work on these problems together like I gave you an example what happens or used to happen say at Berkeley Cadence labs Cadence and you know set up a lab at Berkeley and they had researchers from both sides working together and they had researchers from both sides working together and they had researchers from both sides working together and they had researchers from both sides working together and they had place.

Now it might be a new idea comes, you know you come across a new idea, might be something that is groundbreaking, will take time to scale and you want to look at that separately. There are problems that would give an edge, would give an edge to a company today and they need to be solved in the next six months. Those are also problems that people need to work on and look at. And sometimes there are things that come simply out of the discussion, something a company has been doing for the last ten years, turns out as a far more efficient way that you could deploy in the next six months. So all three happen. Right now I would say the dialogue is at arm’s length.

And that if it changes, I would say funding is less the key than really that, you know, making that dialogue happen because when that starts happening, you will also see excited students wanting to say, look, I know that I will find a career. That doesn’t take any money from the government. It’s fully sponsored supported by industry or sponsored by us. It’s at IT Delhi. It’s called the VLSI Design Tools and Technology Program. It was started in 1996. And till today all the students are sponsored either by projects or by industry. And many of them have led to patents and other things going on. Two of the gold medalists of that program decided to forego all their placement offers.

They had like three or four offers in hand off campus, on campus. And these gold medalists decided to stay back and continue a PhD because they realized all these companies want them. They are really good at what they do. They will get a job and they wanted to see that chip come out. They wanted to see that develop. They wanted to test it out and see the outcome of that. That level of excitement really happens when these are live projects. with involvement from the industry or whoever else, it could be even a government entity, public sector but you need the end users enmeshed with problem discussion and solutions.

No, absolutely. Thank you so much for sharing that sir and everybody. I’d like to invite Parth to share a personal story something that sir just said about people who move to the US don’t often come back and then also on the same side that you know like while you were studying you decided to sort of take a break and sort of join Google DeepMind part time. So what was that thought process like? It was very similar to what sir described as a project, as the passion of working on a life project and what was sort of your reasons for moving back?

Thanks Avrith. I think I think the thought process there was you know I was doing my undergrad and my masters and at some point I wanted to go on also and do a PhD and perhaps be in academia that was definitely one of the considerations I had because I got into research pretty early on even in my undergrad career. I think the excitement around AI and sort of showing that even a lot of PhDs and professors at my university were going on and then building out companies and showing that this research that’s been done for so many years now is actually starting in the 80s but now is actually paying off these dividends and leading to this new technological revolution as I think Professor actually said sometimes a lot of these ideas take a while to actually materialize and we were seeing that materialization sort of happen in the Bay Area there and then so that was at that time you know again AI is one thing that required a lot of computes a lot of these big industrial labs had that compute which you know universities had some of it but didn’t at this scale and you know as I said scaling laws were happening so you wanted that scale that was my reason to sort of be a deep mind to see that scale and then but really I mean we need that sort of same infrastructure in India and we need the same research and people in India so that was what sort of drove me back to here because now with the mission support we have similar compute in India and actually we were seeing that you know these scaling laws show you can scale up but you know there are new innovations that India sort of needs and there are I mean there are so many smart people here so now that we have the compute we have the people for me just made a lot of sense to be back here and you know build the same thing from India.

No thanks for sharing that Parth. So I’ll open the floor for questions. There are mics here if people in the audience want to ask. Hi,

My name is Pradeep Subramaniam. I come from the physical world. So, AI, I have been recently building an agentic AI, but I come from the physical world, R &D, technology, etc. So, my question is to Ganshyamji and to Parth, actually. So, if you build any infrastructure, the physical layer, right, in terms of IoT sensors and the one which is collecting data is the most important part. So, what I was finding in this whole discussion was data centers, infrastructure, but nobody talked about the IoT part or the physical collection of data, right? For example, the electricity plants that you have, whether it is at the power generation, at the distribution, transmission. hardly any IoT based systems or SCADA legacy systems, right?

They are not connected end to end in terms of building a digital twin of this electric system, right? We built something like this for the Haryana government, but it’s not scaled to the full extent, right? So where is the role of India building the ecosystem for the physical layer, which can generate so much amount of data, which can help build this AGI, right? So while infrastructure is good, how do we create? China does that, right? China has used cases which are full of physical layers, which are there. We in India tend to, for example, UPI we build, it did not require much of a physical layer, so we could easily build, right? I think the catch is building the physical layer.

What are we doing for that? For example, in your area, sir.

Yeah, thank you. Thank you for raising this particular. concerns of the industry and this is definitely an issue and let me be honest on this. We have very good infrastructure particularly coming from the generation side and till transmission and going up to the low dispatch centers. Till that absolutely we are at par with the world but when it comes to the actual concentration and link with the customers that means the distribution and the customer link probably we are still lacking behind. So that is the physical and this is the practical situation wherein we are at present and you all must be hearing about the issues of the distribution licenses and their financial viability. So until this they are financially viable probably they will not kick in into the area of automation.

My question is why is the government not supporting to help create this data?

I am coming there only. So we realize this right It’s not that we didn’t realize in the government That this particular segment Of the entire value chain of the power sector Requires some kind of a support We had been supporting this particular segment Earlier as well And in the recent one it is the RDSS Program that has kicked in And this is This is a program which is reform link program So if you are able to Achieve certain goals you will be given the money Or else you will not be given the money And this is supporting in two ways Two very very important ways One is the infrastructure that is required For ensuring reliability of supply And second is the automation systems That means We need smart meters Until this you have a communicable meters You will not be able to do that kind of a smartness Into the entire value chain Of the product So as I said we had this missing link We had come up to the State load dispatch centers But going from the state load dispatch Centers and connecting with the Customers you needed this kind of a smartness and that is how we introduce smart meters and it has rolled out and I think so far more than around 3 crores of meters have already been installed in the country with 25 crores already in the pipeline.

So hopefully we will be able to reach this kind of a number in next say 2 years time frame or maybe in 3 years time frame. What that it leads to? Then it leads to the SCADA system being developed in this particular segment as well. Isn’t it? And right now we do have the SCADA system but it is coming from the other side of the fence. So we have shortlisted a few companies and we are trying to work with them so that we have our own indigenous SCADA systems which is supporting the entire value chain. You all are knowing about the cyber security concerns and we do in a similar manner. And so we want this kind of things to be developed in India.

as well. Now what does this mean? The moment you have the automation in this particular segment, use amount of data is long going to be generated. How do you use this data? So that is how I said that we already had one round of discussion with the startups and some of the AI and driven companies and let me tell you their enthusiasm level and they say, sir give me one year time frame I am going to map all your assets across the country. I mean that is the kind of enthusiasm in these youngsters and we really salute this particular group and that is how my distribution team in ministry, they are working with this kind of people and so that how quickly we are able to take their supports, map them and try to really go further.

Further what, whenever you have this data, then the data has not only to be used only for the billing purpose, right? It has to be used for your planning, planning of network, planning of optimization of resources. I mean you can define any number of use cases the moment you have all this. So this is in pipeline. I’m really thankful to you for reminding me this. And we are trying to.

So my point was that, for example, geo tagging of all the assets of your, you know, right from the power generation to the end point to the consumer. It’s not done end to end today. Right. It’s also a security risk for the country. If some other, you know, server is hosting all that data, it should all be hosted in India in the data centers, all every platform at the back end, including the LLM, which is managing that should be completely in India. Right. So I’m saying that the end to end deployment of AGI will happen only when we have the real physical layer generating enormous amount of secure data, which is not hosted in outside India and lying within the sovereignty.

Data centers of India. I mean, that’s the kind of thought that government needs to think. then we can become so that’s why I wanted my second question to Parth that what are we doing to build that kind of data which will help us set up the AGI part right so AGI doesn’t come simply from some small use case right you need trillions and trillions of tokens and data for that right and you need a domain expertise and knowledge to build that how do we do that what’s the question?

before part takes in differently these youngsters have written insights than me but this gap that you rightly just now mentioned about the data being hosted elsewhere in fact this we came to know the moment we started rolling out these smart meters in our systems and we found that the suppliers are having their resources somewhere going out you immediately we took that action and we said that nothing doing all the data has to be housed in the country itself so right now whatever smart meters that we are placing in the country their data doesn’t go out and it has to be in India so wherever we are able to plug I think we are trying to do that and trying to create that physical layer so that we are cyber secure that is very very important for the power sector.

Parth I think you will take over now.

Thank you sir I think just to echo some of sir’s thought I think I mean a lot of work is actually being done on this layer right so if you look at data sets we need indigenous data sets you have AI kosh by the India mission which is solving for this right Indian data sets for Indian companies to build these frontier models if you look at compute as you said you need you know we need compute in India so if you look at the budget policy we have this data centers you have till 2047 tax people so you will see a lot of these data centers come in there are already a lot of data centers being built you know we have you know Tarunji who is building E2E right see for India infrastructure so that this compute you know the frontier models of India can be can be hosted in India and it can all be done on local compute.

So I think the GPU infrastructure that is being supported by the India mission is actually solving for the exact case that you’re seeing. And already over the last two years, you’ve scaled up our GPU so much for this. So I do think a lot of work has already been done and this work is just going to continue to solve for this.

Sir, I would like to kind of like take a stab at trying to answer your question. I think we are still having some gaps in terms of being able to harness the impatience of the youth to build physical stuff. So unlike software, the physical stuff actually costs money and the cycle time today is very high. So you need to be able to reach the nearest 3D printer to be able to prototype. You need to be able to kind of like design the chips. You need to be able to solve for all the physics problems. I think what the LLMs will do for us and the frontier models will do for us. is to reduce the cycle time of the thinking part like say you have to do the actual physical world calculations you have to do the digital twin part that part is used to take a lot of time that gets solved faster but what we still need to solve is that we need to do the prototyping that is the part we still need to solve but i think like having spoken to a few companies who do who used to do physical prototyping they have done away with a lot of physical prototyping all together and they are just doing it on top of the digital twins now so so i think somewhere we will converge so that’s my hope.

o I am saying that imagine the next upi innovation is say the agentic ai for the vending machines i am working on that but physical layer right of the sensors which pull the data for a vending machine the back end of the vending machine is the back end of the that hardly any vending machine is connected to iot doesn’t have any physical layer. It is just used like a dabba, right? So the point I am saying is that why the government is not enabling the instruments which help the connectivity of this data to the AI and the data centers and then the intelligence can be built to automate, to create more jobs and you know, it’s very counter intuitive.

We say that we will build agentic AI, we are going to reduce people. No. Actually the work is going to increase because the vending machine infrastructure will go 10 times or 20 times it will become like Japan, right? You will have more vending machines. But I do not think that that kind of an infrastructure, private industries, I am from the private industry, I cannot build it. For me, day to day running a vending machine business is I get cheap labor. I cannot use sensors, right? So this is the catch -20 -t kind of situation where most of the infrastructure that we have in India, we have cheap labor, we still manage with that. We cannot take the next leap.

How do we take the next leap by getting the platforms like UPI to build with physical layers that was the question.

I think some of the answers will be given by the India Energy Stack I think you must have heard about that and I am also a member of that committee and we are deliberating on all these use cases and where that gaps are so definitely I think we will take care of that.

Thank you sir for asking that question I would like to circulate the mic in the audience if other people would like to ask questions people can just raise their hands if they have any I think there is one in the back.

Hello, good morning to everyone there are three things to develop any industry first is primary sector, second sector and third sector your AI impact summit is always talking about business model what about the management manufacturing sector because if any unit is made like semiconductors are not developed in our country we take Chinese companies we take Chinese companies we take Chinese companies we take Chinese companies we take Chinese companies we take Chinese companies Although the industry is being built now, six units of semiconductor industry are being built. But what about the 9 gigawatt industry which will be built for data centers by 2032, by 2032, what about the manufacturing sector? Until that is not developed, will we keep working on business…

All these things are interconnected. So nothing is to the exclusion of another. Whatever sector you are working in, eventually that will feed into the other sectors. So as long as the intent is there to be Indian and by Indian, when there is intent, then automatically all the problems will be solved together. If there is intent that, we will work together to solve all the problems, I will move forward but the rest will stay behind. then we will not move forward. How are we developing that?

There is no framework. Nothing comes first or later. Everything goes in parallel. Microprocessor units. Microprocessors. Because for AI, the most basic unit is microprocessor. And for data centers, the most micro unit is microprocessor. So what about microprocessor? Will we keep buying from China? No doubt that in 2025, six units are being made in India and all. But so far, there is no prominent result.

It takes a little time to reach that level.

Sir, actually, if you look at microprocessors, either mobile phone or server or desktop, on the whole motherboard, there are about a couple of hundred pieces of intellectual property. Now, all that intellectual property, if you look at it, a lot of intellectual property is made in India. It is made by the people of India. So licensing India ke through nahi hoti Because the IP is getting developed with foreign money Toh yaha pe hum R &D karte hain, IP develop karte hain Koi usko aur commercialize karte hain Toh I think that gap has to be fulfilled By having volumes which are domestically available So jab domestically volume available hoga Toh jo system on chip IP jo develop kiya gaya Usko add karne ke liye jo log kaam kar rahe hain They will see a domestic market So uske baad ye saari cheezme automatically honi shuru ho jayengi Toh like I think jo bahuti important part abhi ho raha hai Is to move the country forward Build a large market which is interconnected with the world Once you have large markets interconnected with the world Then youngsters who are very impatient to go and build things And say that okay isko commercialize kia ja sakte They will go and achieve the success So like I said Kuch bhi can’t be serialized.

You can’t say that we will do this first. We will go back to the planning era. The communist states who used to plan first we build fundamentals then we will build something else on top of it. So free market allows you to work on all these things in parallel and it throws up the opportunities. So if we fix our economics, all these things will be fixed on their own. Thank you.

I will say one thing. Basically we are moving gradually in the manufacturing sector. If we talk about the power sector, I don’t have much information about METI and other areas. Here a lot of equipment is almost 100 % indigenous. There are certain which is ranging from 50 % to 80%. They are being targeted to see that the domestic content of that equipment also goes to 100 % in a given time frame. There are still equipments which are yet to take off. Those which are 78, 76, which we were telling you about. . We are trying to reach almost 20 % to 100%. So there are different stages of indigenization. But definitely we are targeting that all these equipments must be manufactured in India.

It is same that in primary sector, there will be a lot of silicon.

Absolutely. In power sector, we use a lot of electronics. For example, I gave you a small example of IGBT. IGBT is again an electronic equipment, which is right now we are taking from outside. We had the challenge here and we said nothing doing. Now there are Indian companies are going to manufacture this. So now we have given them the task for two years and they need to develop in two years and commercialize. So similarly, we are taking it up. Thank you.

Professor wanted to make a comment.

So let me divide up that answer into multiple parts. So, the word microprocessor of course is no longer used, right, we do not really talk about microprocessors, the most of the current AIML all runs on GPUs, the architecture is very different from traditional microprocessors. So there is for at least space and some of the other sectors, we have a fairly successful, you know, operation running at semiconductor complex limited Chandigarh, right, Mohali. So the plant at SCL produces some earlier generation microprocessors and produces chips for a variety of other things. There are of course similar other entities around the country, but most of the effort of what you are putting in the VLSI design space, the chip design space, so to speak, most of it is design.

The manufacturing, most countries, I am not saying only India, most countries, India, countries, Europe, US, in fact many of the earlier semiconductor manufacturing plants in the US shut down and are now producing solar panels, right. So the most of the efforts around the world are fabulous design houses and India leads in that. So if you look at, you know, Bangalore, Pune, Hyderabad and to some extent some in Noida, Canada, very significant fraction of designs done for many of the smartphones are actually done within the country almost 100%. In fact some are actually really 100%, complete design. That design is a major component of the cost of developing a new design. It’s actually the manufacturing is there but most of the cost is really in the initial stage really the manufacturing, sorry the design cost.

That’s happening in the country. Scaling up the, you know, the semiconductor design plant itself will take time but you can see it’s already rapidly happening. In the case of memories flash memories and so on there’s already very large investment that’s happened by a fairly prominent multinational in Gujarat and elsewhere that’s already taken off very well. There are similar efforts that probably you will start hearing about their outcomes and outputs in the next 2 or 3 years or even less. So I think as far as the you know that space is concerned the Indian engineers have it almost entirely covered so I don’t think that’s a cause for worry. I think the interlinking of these parts will happen if you ask me organically because everything exists in one.

Thank you.

With that, I would like to thank all of our panelists for spending so much time and answering everyone’s questions. I would like to thank the organizers for letting us go 30 minutes over the time. And thank you so much. I’d like to invite the Indian Air Mission delegates to facilitate the panelists. Thank you so much, everybody. Thank you. Thank you to Suvarath for all the moderation as well. Thank you. Thank you, folks. Thank you, everybody. Thank you. Thank you.