EXCLUSIVE: India's First Orbital Edge AI Cloud - Nevecloud's Sovereign Infrastructure Bet

[00:00:00]Hello everybody. Uh thank you so much for joining in. I'm from CubeCon 2026 in Mumbai and joined with me today is Vijay Kumar Aaram Nadar. He is the chief AI officer of new cloud a company that we've spoken about multiple times on IM network. So firstly uh welcome Vijay.

[00:00:16]How are you liking the event at CubeCon today?

[00:00:18]>> Yeah, thanks for welcoming and uh CubeCon is always uh close to me. Uh it's I think second one in uh India what is happening? Earlier it was in Harabad.

[00:00:27]Now we are doing it. We are a landlord sponsor as well. We have taken a booth.

[00:00:31]W is really amazing. Developers uh customers we have the booth is houseful.

[00:00:36]>> It's going good.

[00:00:37]>> Right. And to be sitting with you today in these times it's a very important conversation about sovereignity where we have one layer of players like server, mani and socket building the foundational models but the other layer of sovereignity that ensures these models are built is getting enough comput. Necloud is an important player in the country. So this entire business model uh Vijay uh terms of procuring GPUs and being a neocloud player and helping startups access with compute uh how do you see N cloud's market today and in terms of the customers that you work for so how important or what how do you see the growth of a business of Ncloud today? See so virity uh for me it's like two ways right one is the data where it resides and who controls the data right uh we have seen what happens uh to mythos uh the data access from US perspective >> but from Indian soian perspective what we are seeing is even though the models are open source >> where it runs it matters who is going to control the models right as a providers like us >> uh uh we are providing a platform where the GPU platforms what we are providing where we control the complete end to end stack at we are building from ground up till the application layer we are building everything we have our own software stack we have our own factory where the data is going to run and where the orchestration is going to control the data >> we give the complete uh end to end uh approach for customers where they can control it >> right >> uh that's from a data control perspective on the soilian angle yeah >> right right and in terms of let's say u necloud is backed by you know rack bank data centers the you know the large player out there and terms of securing compute beat Nvidia beat AMD or not just the GPUs all the you know adjacent parts of the infrastructure around it a lot of people say that today there's a fear that models now are restricted to other countries but what if tomorrow uh you know you have GPUs that are going to be restricted for export so in terms of uh uh leading a neo cloud company how do you see that uh you know situation coming >> see the challenges will always be there uh that is why uh it is very important as we see the futuristic approach of how uh we are going to align with the uh GPU access or the chip AI chip access uh GPU's been there for a long time >> right but the gamings and other things have GPUs have been used but when AI AI come into the play we are using GPU >> but there are TPUs are available there are AI chips are coming in >> the market where I see is there there could be a chance of inferencing 80% of the markets will be captured by inferencing who might not require a GPU.

[00:03:13]Also there could be AI chips that is inferencing chips uh which will handle all this inferencing intelligence where the models will provide the intelligence accessibility through the AI chips >> right >> but uh the challenge of procurement and other things uh as we face but we handle that through the partners and uh direct channels uh on based on the demands actually >> well that was about the future scenario today I mean we we see a lot of releases from companies like Nvidia AMD and in terms of procurement one side we have the Indian government through its India emission procuring uh planning to procure 38,000 GPUs but then we also have players like N cloud directly you know purchasing GPUs so if I were to ask you how difficult or easy so if you could to our viewers explain how the procurement process works whether uh Indian neocloud players can you get access to the latest GPUs or is it usually the older ones that are easy to procure so how does the entire process go about >> see process takes a long time uh for example India mission uh they procure GPUs from us actually >> right >> so we are we are part of that and we are progressing towards that from India perspective >> and uh from uh latest GPU procurement part it takes a long time actually in fact we are progressing for B300 for a long time with partners actually >> uh direct purchase is going to take time as well we are going with the partners actually partner enablement securing GPU is always take time consuming meaning uh we secure the customer we secure the procurement with partners and then we go hand in hand from the market perspective >> that's the way that we reach >> uh latest GPU yes we are uh with B300 B200 we'll we'll look for the market where the GPUs are available but in the futuristic approach for example the latest GPUs are available we'll have okay there are secured customers okay we'll go ahead and place the order accordingly >> but we do also have some of the GPUs which is in the market already like uh uh 8000 those kind of GPUs also will uh will keep running in our platform so that there are customers who can use for those GPUs as well as well uh by the time we procure the latest one.

[00:05:20]>> Right. Right. So as you mentioned you know the India mission largely is dependent on players like yourself like new cloud in terms of getting access to GPUs for the mission.

[00:05:29]>> [snorts] >> So today if you were to explain to our viewers the entire you know the kind of players that exists in India's you know comput you have new clides like yourself then you on the higher level you have hyperscalers or uh like like place like cifi you know who provide data centers and spaces for hyperscalers and then you have yota you have e2 network so how is the compute market structured today vij which >> okay so I'll split this uh into two because I I was also I have worked in sephi >> oh okay 2010 when CFI cloud infinite launched infinite cloud I was part of that okay >> so I know the way that we are creating today the new players how we are creating the EI stack the way that we designed uh earlier in uh 2010s the cloud stack >> uh the bringing up the rack system itself is the cluster size the capacity planning everything has changed actually >> uh initially in the cloud era we we were like three uh uh uh it's it's a very small clusters we plan now we are planning for a higher clusters actually the AI systems >> uh considering ncloud perspective how I'm planning I mean how I'm differentiating is uh the way that we layer the intelligent or orchestrator >> uh the full stack solution there are players who are using a proprietary softwares uh from others where they are licensing it and they are building the platform but at new cloud we have our own proprietary software which is developed we have a strong engineering team which has been built. The control plane what we have it has been completely in-house we have built the intelligence though we are using open source components but the orchestration the heart of new cloud is being completely proprietary what we are using uh that way we are trying to see uh the differentiator in the market as we go with other players >> I can give you an example uh for example if someone is using uh Nvidia uh software Nvidia has to patch it Nvidia has to send Nvidia holds a uh good control >> good control of it but if you are using an open source if you are contributing to it we can reach the customer demand way faster than we are waiting for uh the patch release from Nvidia so it's like a software with hardware hardware we are already using uh Nvidia hardware but we are also completely contributing to open source and we are consuming open source products as well so it's like a differentiator if you are asking we are building an end toend full stack solution which we control all the layers of uh the orchestration >> so just to uh summarize and confirm my understanding today. If a startup wants compute, not only does NEC cloud provide them compute but also some of the layers where they can effectively deploy and orchestrate their compute around their uh software and the back end is cloud offers.

[00:08:12]>> Yeah. Uh today we are uh uh launching uh the AI agentic studio.

[00:08:18]>> Okay.

[00:08:18]>> Uh which is basically the sandbox environment where the developers where the AI de AI application developers can use this sandbox environment. Okay, >> that is one of the offering that we are launching today.

[00:08:30]>> They can run their own applications into the sandbox environment because the demands what we see is >> uh developers don't want to use it completely, right? They can use it as and when it's required. For example, during this uh session we need to use this room after that it's free. So like that the sandbox environment will come up and utilize utilize the compute capacity and exit the door. So it's like that level of agents can run this complete orchestration can be run from agent perspective open cloud code all that we are giving it as an agent actually so we're making the developers life easy to access the compute >> right right >> and uh Vijay so a good part of I'm I'm assuming uh rightly that new cloud's expenditure is on GPUs and compute but today as I speak to many developers they're spinning up very capable applications on smaller language models they are getting more and more capable like The Gemma 4 is much more capable and so tomorrow do you see when more and more uh you know startups are running less resource inensive models on NeoCloud clouds like yourself and obviously that means that your compute utilization is a bit lower and the startup is getting the results that they uh want. So do you see the rise of small language models and more capable ones will lead to better cost efficiency for players like yourself and you can just split more existing compute to more and more players and you also spend less is a net net benefit.

[00:09:53]>> Yeah. So future is going to be there. I feel uh there will be lot of small vertical intelligence will be built.

[00:09:59]That's what we call uh small language models.

[00:10:02]>> Uh that is going to the players like us will distribute the routing. For example, not everyone requires trillions of model, right? I have seen one customer was using 1 billion model or 500 10 million model which is completely trained for their own specific use case actually right in that case they don't want a large language model.

[00:10:23]>> What like players we do is we understand the context of the request prompt and we distribute to across this model that is the orchestration that we are building right. uh this in this case what happens is you don't want a larger GPU memory to run this large model we'll run this vertical intelligence and based on customer request we route through that okay from our intelligent orchestration that is that is the advantage that we are building >> but uh certainly there will be lot of uh SLMs will be coming in >> right >> so [snorts] and that way because of your intelligent routing and orchestration you also ensure that they that your customers aren't spending needlessly on models >> exactly exactly because uh see if I wanted to run a trillion model then I need a larger capacity of GPU. I need to put the cost >> to customers. Now I'll run smaller language models the costing of GPUs will be lesser but I'm giving the same level of intelligence what the customer want actually it's more of a vertical intelligence we are calling right what they want we are giving based on the context actually right now the way the industry is moving is context aware valuedriven tokens >> uh earlier it was more of a token per cost now it is more of value the question what you ask I have to give the answer to if I'm giving different answer it's it's meaningless. So we have to be a valuedriven token cost.

[00:11:41]>> So I strongly believe it will be a context aware uh request from prompt and value driven by cost.

[00:11:49]>> If I'm giving the right value to right answers, my token will be paid for the cost.

[00:11:53]>> Right. Right. Right. And also uh one other question is with regards to sovereignty we see recently uh you know they raised a lot of funding. they also have models and when we spoke to Vive Ragavan he said that for me sovereignity is having a model where I can audit everything so today you might say that you can build on top of an open source model but I can't audit the data that is trained down now let's say if someone let's say if server only wants to deploy their model on your uh cloud infrastructure or GPUs do you think there is going to be a benefit for you as well because server knows in and out of their model they can sit with you they can ensure that they're not spending unnecessary comput resources they provide an optimized experience for your customer and everyone is winning in terms of the cost that is spent. Do you see that as an advantage because server owns every layer of the model? Certainly uh it there's a lot of advantage actually because they are from uh India and surv build a fantastic model actually >> uh from soian angle the data traceability is what most of the important thing what we see as a provider >> uh today as the agent in agent uh workflow automations increases the traceability of who does what is what is very important agents agents are coming up with its own roles now serv has a model the task what the agents are going to execute. how we trace this logger >> uh that is that's we call like observability and traceability trace back the actions what the agents has done >> that is very important and ser with the their own models and with the new cloud it's it's a fantastic uh >> uh go to market and synergy is what I would say because both of us are soian by nature they have built their own local models and we are soian by nature in the data center and the software perspective and we go well and also we are also planning to deploy ser opensource models in our uh uh Ne cloud to make it available for our customers as an endpoint.

[00:13:46]>> Right. Right. And also my uh one question is so in India how are you seeing the demand for services like Neve cloud and the overall neo cloud service that you provide. Do you see the growth of startups coming in and saying hey you know what today everyone's obviously pivoting to AI every company wants to run an AIdriven service or an AIdriven product today. So how do you see the demand for name cloud's products and services increase in this day and >> see I'm seeing uh there are two two ways because I've been in this industry for 20 years now uh there are serious startups there are starting startups in the very very initial stage initial stage they are still going to hypervisors using the free credits and uh they are spending times there but uh serious startups who are coming really they wanted to make a business we are seeing a conversation with them because uh the accessibility is what they are also saying we are from Indian players the accessibility of the comput resources what we provide and they know this is where it's running actually so in that way we are seeing some of the voice agents the latency part of it >> all right the accessibility from the Indian region the voice agents that is bringing getting developed uh voice perspective we are seeing lot of conversations lot of customers are getting onboarded because of the latency perspective how the turnaround time from text to speech how the conversation happens right so we we have an upper age towards that actually being an Indian player as well for the Indian customers from the startups but serious startups are uh in conversations and they wanted to mature with this >> right and of course Ncloud's you know ambitions as a player has reached a sky high and beyond quite figuratively so with the recent announcement of you know with Agnikul so uh so how would you what is your thoughts on the ambition that the company has today because today there's also So a a notion that okay uh India is not fully AI you know it's not as mature as a country like USA but for a company like new cloud yes GPU deployment comput deployment is there but we also want to focus on higher goals like you know uh extraterrestrial sort of a thing terrestrial data centers so what's the overall ambition of the company today >> see from uh uh company perspective we have a long-term vision and we focus on deep tech as well cloud deep tech part of it uh this space-based data centers coming as part of Nikloud deep tech uh but uh when we wanted to announce this partnership also we were there is a purpose to do that we don't because these conversations were happening for years actually now we felt there is a purpose to do that what it means is how we can make the latency low latency which is ultra low latency in fact uh the challenge is what we have today we build a data center we need a land most importantly the power >> right and we need a proper critical usage of customers. Actually what we found is customers are there you are but how we can reach those data how we can reach the low ultra low latency provided intelligence to accessibility is very important [clears throat] that's where we found building an uh edge orbital data center uh it's not a full-fledged data center which I'm thinking it's more of an orbital edge data center which will be a compute capacity of that uh that is also in the time being we are thinking about the yeah inferencing part of it Right.

[00:17:06]>> Right. Uh there during that announcement itself we got to one uh conversation with government agriculture government where they wanted to use that where they wanted a geographical in uh India perspective where they wanted to use a satellite right so there is a serious business to it if you are really providing the value to them >> which uh which which is completely an ultra low latency like 10 to 15 m uh subs you can provide reachability right So a robots which which does the operations it's required actually there is a serious business towards that where we have wanted to make it it's an ambitious project but it's a doable one because already Nvidia servers have been certified for the space uh climatic conditions and of course uh the source of energy is free course we need to see how to emit the energy we need to see how to store the energy uh because when we reach the eclipse region we need to see how to uh emit the energy to store it using the battery capacity And when we reach SA now we can make the most out of the comput capacity. So there are a lot of orchestration and logics but it's a doable one futuristic one but there is a serious customer base uh where we have.

[00:18:17]>> Now one question I'd like to ask you as the chief a officer of new cloud. Today a lot of people are saying okay India might not have enough purchasing power for compute but also you have players like yourself building a full stack solution. So how much value do you see in okay we're going to build software optimizations make most of the existing GPUs now of course there's a conversation of CPU based workloads coming in for agentic AI applications and workflows so Vijay what how do how much of value do you see in let's build more software optimizations let make let's make innovation in the orchestration layer and let's not worry too much about purchasing raw compute but providing maximized value to our customers through the existing power that we have and how much of that is a focus for N cloud That is a topmost priority I would say right because we are building a logic how we can utilize the ideal capacity of the GPU as we all know that GPUs cost expensive.

[00:19:10]>> Uh >> we have seen customers who are bought the GPU and utilization is 30 to 40%.

[00:19:16]Right.

[00:19:16]>> We have seen the GPUs are idle.

[00:19:20]>> We have seen the time that GPUs are completely 100% utiliz utilized. For example, the day times are GPUs are utilized. The night times are not GPUs are right.

[00:19:29]>> So what the plan to approach the solution is uh to answer your question this is topmost priority. That is what the intelligent orchestration layer what we are building will handle all the placement logic of the jobs. For example, you are running a training sessions and you have a training that you have to do in the daytime, night time what the GPU does. We'll have an offline jobs which will be scheduled.

[00:19:52]For example, if you are selling a GPU for $4 per hour, this offline job will be scheduled when the GPUs are ideal and that time the schedule job will use the GPU and that job will be like $2 per hour, >> right? So the but the whole idea here is how we can utilize the GPU when the GPU is in ideal state and if a job can wait till night time because not always all the customers need immediate GPU their job can do later also that's what we call baduling job >> right baduling job can run night also because they need the output in the next day morning so we are there are multiple approaches we are taking one is how we can improvise our placement logic to make sure we placing the jobs uh to avoid the idle GPU and the other part of it how we can do an offline jobbing when the GPU is not really utilized how we can utilize the GPU for offline batching and inferencing logic is also coming in how we can use the ideal GPU so our end goal is how we can utilize the GPU 100% to make sure if you use GPU 100% we can reduce the cost from the customer perspective also if not this 30% utilization rest of the 60% as well to go to customer action But if we as a provider we are utilizing the GPU in effective way the cost will reduce that's why cloud we wanted to be more affordable that's why the intelligence what we are building the software layer is very very very important >> right right my last question to you uh Vijay is when we talk about sovereignity one is open source we're at an open source conference today open source is one of the biggest pieces of sovereignity because you develop something you get to own it traceability auditing everything >> so today sovereignity is spoken like it's a responsib possibility of few players like Saram, Neve cloud, Y or anybody but the way I see it is sovereignity everyone can contribute towards sovereignity through open source today we have hundreds of thousands of opensource developers today so what is your message to so many engineers I mean the huge community in India to turn to open-source as their contribution towards sovereign tech maybe if anyone has let's say an idea towards better GPUuling or better GPU orchestration or better uh you know open-source optimizations to make GPU more efficient. So I'm sure there are tons of people with ideas like that. So what is your message for you know students, engineers and veterans in the industry where you see sovereignity as a responsibility of everybody and that responsibility can be fulfilled through open source contributions.

[00:22:22]>> See certainly open source contribute uh we do open source contribution. We also consume open source product actually uh because uh if if a product is open source there are a lot of hands to it >> to develop to make more idea innovations will come diversification will come right uh the way that their contributions will come ideas will be discussed that's why open source is very very important and that's why we are in coupon when cubernetes 10 years back when it was announced as an open source from Google uh we didn't expect this big uh the cubernetus community as it is grown because of the community because of the contribution the open source community cube cubernetes as a stack it's going and today we are using cubernetes as well >> uh everyone has to focus on the deep tech actually uh if I wanted to give uh the recent uh college grads or the startups who are coming uh instead of focusing on the application layer of using the intelligence where the models already built focus on the deep tech how you can crack these kind of sol these kind of problems what the new cloud players as well we are facing. There are places where the ideal GPUs are there.

[00:23:29]There are places where the GPUs are completely utilized also. So how these kind of balanced approach can be taken but focus on the deep tech where you solve the real problem. For example, the traceability problem what you're talking right who does what? Thousands of agents are going to do one job. Which agent there's no people involved right? uh there are the agents will be named the thousands of agents will do lot of things who will take that responsibility accountability we cannot ask the it's a responsibility of the person who is designing the agent who is going to execute the task and then the observability point of it how we are going to trace who does the job this is also comes as part of the governance actually right where the data resides one of the other challenge which we are seeing is the agent gets access to the root system agent will also perform which it's not supposed to do that that's why we do guard rails of that uh policies so we need to have a governance we need to have a policy so all this is part of the soian perspective also so I add one more thing who controls the data is also comes under the soianity because that's how I position actually soity is not just the jurisdiction where the data lives it also the person or the organization that is going to control the data So that is also part of the soian angle which we need to see in ncloud we I see that like a three layer approach one the place where the data lives even the software stack what we use right uh the that's why the inner software stack how we are going to use the software stack and the next layer is the uh the data how we are going to completely control and orchestrate the data so it's like a triple layer of soanianity we follow and that's why the governance control guardrails, traceability, observability, everything has to put together to make a successful platform where Nikloud is approaching towards it as well.

[00:25:27]>> No, thank you so much for your insights uh Vijay. Uh I think it's a good time that we're speaking about sovereignty when there's so much going on >> and excited to see how Neocloud you know charts this journey uh already a very ambitious one but yeah looking forward to the future. Thank you so much everybody for watching. For more coverage from CubeCon, stay tuned to Aim Network.

[00:25:47]>> Thank you.

[00:25:48]>> Thank you. Thank you so much. Wonderful.

[00:25:50]Thank you.

[00:25:53][music]