Panel Discussion Data Sovereignty India AI Impact Summit
20 Feb 2026 16:00h - 17:00h
Panel Discussion Data Sovereignty India AI Impact Summit
Summary
The panel discussed what digital sovereignty means for AI, focusing on who sets the rules and where data and compute are located [5-7]. Sunil argued that sovereignty does not require total isolation; instead it involves deciding which parts of the technology stack must be controlled locally and which can be collaborative [21-23]. He emphasized that core compute infrastructure and model training should reside within national borders to serve language-specific needs, noting that a domestically built model can satisfy about 95 % of India’s use cases without needing frontier-scale systems [25-28][34-38][39-42].
Sunil illustrated this with the migration of India’s Pashini language platform from a hyperscale cloud to Yotta’s local data centre, and the open-sourcing of a critical NVIDIA component to keep it under Indian control [136-142][148-154]. Nasubo highlighted that many African countries lack compute capacity (≈1 % of global) but possess rich data and local use cases, so they pursue partnerships while also developing offline-capable solutions for regions with limited connectivity [57-63][166-174][176-179]. Seima stressed that sovereignty is about strategic control rather than full ownership, calling for visibility into supply-chain structures, guardrails to protect digital assets, and trust-engineered designs that are transparent and auditable [77-84][88-98][100-107][109-118].
The moderator summed up that sovereignty requires a balance between building domestic AI compute and engaging trusted global partners, noting that no nation can do everything alone [48-53][130-134]. All speakers agreed that government must provide clear policies and sovereign guardrails while fostering public-private collaboration to ensure security, continuous verification, and industry innovation [184-191][198-205]. They also concurred that the ultimate goal of sovereign AI is to address real-world problems for the most underserved citizens, such as language diversity in India or specific health data for African women [31][60][220-223].
Seima added that treating digital infrastructure as a national asset and providing long-term policy stability encourages private investment in large-scale AI factories [185-196][200-204]. Overall, the panel presented a roadmap where sovereignty is achieved by controlling critical layers, establishing trust frameworks, and aligning market, government, and society to deliver AI solutions that are both secure and locally relevant, ensuring AI benefits the “last person in the line” [215-218][221-223].
Keypoints
Major discussion points
– Sovereignty is fundamentally about who sets the rules and controls the digital infrastructure, not about total isolation.
The moderator frames the core question as “who gets to make the rules?” and notes that sovereignty is usually discussed in terms of data location [5-7]. Sunil stresses that sovereignty means “we do not allow a single country or a single company to define our digital destiny” and that compute must be “within your country, … where your data is getting processed” [25-27][34-36].
– Local (sovereign) compute and storage are essential, but they can incorporate foreign technologies within a ring-fenced environment to avoid lock-in.
Sunil argues that “compute infrastructure … has to be within your country” while also acknowledging the need for collaboration [25-27][34-42]. He illustrates this with the Yotta example: the AI platform was moved from a hyperscale cloud to a locally-controlled data centre, using NVIDIA, Microsoft and Amazon tech inside a “ring-fenced” wall where no third-party can log in [135-154].
– Designing AI for local contexts and building trust are critical; this includes using native language data, respecting supply-chain transparency, and establishing clear governance.
Nasubo highlights that Africa’s advantage is “data and use cases” and stresses designing for lived realities such as breast-cancer diagnostics for African women [57-64]. Seema adds that sovereignty requires “visibility into ownership structures” and “policies, guardrails … to ensure … not compromised externally,” emphasizing transparency, traceability, and engineered trust [78-86][104-110].
– Strategic partnerships, not dependence, are the pragmatic path forward.
Seima notes the rise of “air-gapped, ring-fence environments” and the need for “global partnerships” while keeping strategic ownership [91-98][118-120]. Sunil’s story about open-sourcing a critical NVIDIA component shows how a partnership can be reshaped into sovereign control [139-154].
– For low-resource regions, offline capability and community-driven innovation are necessary to achieve sovereign AI.
Nasubo points out that only ~50 % of Africa has reliable connectivity and that “offline access” is a key design goal, with local innovators using compute provided by Kala [166-173][176-179]. The moderator links this to India’s Aadhaar offline verification as a precedent for the Global South [180-182].
Overall purpose / goal of the discussion
The panel was convened to move the concept of “data sovereignty” from a slogan to concrete practice in the AI era. Participants shared how governments, industry, and civil society can jointly define rule-making, secure local compute and data, design culturally-relevant AI, and establish trusted partnerships so that every country-especially those with limited resources-can reap AI benefits without surrendering strategic control.
Overall tone and its evolution
The conversation began with a definitional, analytical tone, focusing on what sovereignty means and why it matters. It then shifted to a pragmatic, solution-oriented tone, with detailed examples of infrastructure, open-source workarounds, and governance frameworks. By the end, the tone became optimistic and inclusive, emphasizing collaboration, trust, and the moral imperative to serve “the last person in the line.” Throughout, the speakers remained collaborative and constructive, moving from abstract framing to actionable, real-world strategies.
Speakers
– Speaker 1 – Moderator/host (appears to be the event moderator) [S11]
– Sunil – Panelist (no specific role or title mentioned) [S2]
– Nasubo – Panelist (no specific role or title mentioned) [S1]
– Seema – Chief Executive Officer, L&T Vioma [S8][S9]
– Speaker 5 – Panelist (no specific role or title mentioned) [S5]
Additional speakers:
– (none identified beyond the listed speakers)
1. Opening & framing (Speaker 1) – The moderator opened the session by defining digital sovereignty as “who gets to make the rules” and noting that the debate usually centres on the location of data and compute [5-12]. He then asked Sunil whether a “sovereign yet connected” model was realistic [8-12].
2. Sunil’s view – Sunil rejected the idea that sovereignty means total isolation, explaining that it is often confused with a “do-everything-ourselves” mindset but in practice requires recognising inter-dependencies across the global technology stack [13-27]. He argued that a country must retain control over the strategically essential parts of the stack-compute infrastructure where data is processed, stored and models are trained-while collaborating on the rest [25-27][34-42]. Using India’s linguistic diversity, he highlighted the need for native-language voice AI that can handle regional slang and deliver real-time responses, a capability achievable with domestically hosted models of 20-100 billion parameters rather than frontier trillion-parameter systems [28-42].
To illustrate operationalisation, Sunil described migrating the national AI language platform Pashini from a hyperscale public cloud to Yotta’s locally-controlled data centre [135-142]. After the migration the team built roughly 30-40 different technology components and deployed them on virtual machines inside the data centre [143-150]. The only remaining foreign dependency was the NVIDIA NVCF library, which was open-sourced and brought in-house, eliminating the external reliance [151-154]. He concluded that the best foreign technologies-NVIDIA, Microsoft, Amazon-can be used provided they run inside a sovereign, access-controlled (ring-fenced) compute stack [144-152][144-154].
3. Design layer (Speaker 1 → Nasubo) – The moderator asked Nasubo to discuss design for local realities. Nasubo noted that Africa possesses only about 1 % of global compute capacity [57-58] but has rich data and concrete use-cases. He cited the development of breast-cancer diagnostic models that reflect the specific tissue composition of African women [58-64]. Because roughly half of Africa’s population lacks reliable internet, he argued that offline-capable AI is essential; Kala is therefore building compute resources that can operate without constant connectivity and offering them to innovators at the AI Village [165-174][170-179]. He stressed that sovereignty must be pursued through partnerships that provide compute while allowing African stakeholders to define the rules, rather than accepting externally dictated solutions [166-176].
4. Critical systems (Speaker 1 → Seima) – The moderator shifted to critical systems and invited Seima. Seima asserted that sovereignty is about strategic control and visibility rather than outright ownership of every component [77-85]. She called for clear policies and guard-rails that treat digital infrastructure as a national asset, ensuring transparent supply-chain structures that cannot be compromised by external geopolitical leverage [88-98][121-128]. She introduced an air-gapped, ring-fenced sovereign-infrastructure model within commercial infrastructure [100-107]. Trust, she argued, must be engineered and continuously verified; designs should be transparent, traceable and auditable, and partnerships with global vendors must be built on mutually-established trust rather than dependence [109-118][122-124]. Seima also highlighted the need for a public-private partnership model in which the government provides stable, long-term policy guard-rails while industry focuses on innovation, scale and time-to-value [184-196][200-204], and she stressed continuous verification rather than point-in-time checks [198-205].
5. Moderator synthesis – The moderator summarised that sovereign compute is both desirable and feasible [48-49] and that sovereignty also involves who decides how systems are designed [52-53]. He reiterated the consensus that sovereignty requires a balance between domestic control and trusted global collaboration [130-134] and that co-accountability among market (bazaar), government and civil society (samaj) is essential for implementing sovereign AI [214-218]. Continuous security verification was identified as a key governance practice [198-205].
6. Consolidated trust-partnership insight – Across the panel, speakers agreed that foreign hardware and software can be leveraged safely when deployed inside sovereign, ring-fenced environments [144-154][100-107][109-118], providing the pragmatic path forward while avoiding lock-in.
7. Actionable takeaways
– Anchor sovereignty in rule-making authority and control of critical compute and data [25-27][48-49].
– Leverage foreign technologies inside sovereign, ring-fenced environments [144-154].
– Design AI around indigenous data, multilingual needs and offline capability where connectivity is sparse [28-31][165-174].
– Engineer trust through transparent, auditable supply chains and continuous verification [100-107][109-118][198-205].
– Establish stable government guard-rails and public-private partnerships for long-term AI infrastructure investment [184-196][200-204].
– Treat digital infrastructure as a national asset to ensure consistent security, oversight and accountability [201-208].
8. Moral framing – The moderator quoted Gandhi’s call to consider “the last person in the line” and stressed that AI, even in a technocratic age, must serve the most underserved citizens [221-223].
9. Session close (Speaker 5) – Speaker 5 thanked the participants and asked the audience to wait for the next session [220-222].
been used almost as much as AI in this session, the last three days, it’s been sovereignty. So I think it’s good that we get 24 minutes and 47 seconds to discuss what sovereignty is about. So I’ll jump straight in. We’ve got a great panel. And I think the key question of sovereignty is a question of who gets to make the rules. And the way in which sovereignty has been discussed is in terms of where data is stored. So we have a variety of viewpoints here, and I look to get some opening remarks from each of you. So Sunil, I’ll start with you. You’re running some very large and very impressive data centers in India. One term that we’ve often heard is sovereign yet connected.
So we want to be sovereign but connected. Is that realistic?
No, as you said, there are different ideas, different theories, different narratives going on in sovereign. Everybody has their own take on sovereignty. And so many times, sovereignty is also confused with we will do everything ourselves. We’ll start looking inwards, we’ll isolate ourselves from the rest of the world and everything is done by us also. I think let’s understand any and every technology stack, AI is now the latest one, you will always have interconnectedness, interdependencies across the world. Somebody will be good at making chips, somebody will be making raw material for the chips like gases and maybe rare earths, somebody will be making models, somebody will have great data sets, somebody will be very good in making applications, agentic AI.
You will have, and of course capital flows, somebody will have lots of capital and somebody will be waiting for that capital and somebody will have talent. We all know where India is good at and where any other country is good at. So, sovereignty for sure does not mean we become isolated and just try to do everything ourselves. It is a matter of what is the thing we need to control and what is the thing where we need to collaborate. For sure, it definitely means that as a country, we do not allow a single country or a single company to define our digital destiny for future. Answering your second question, there are certain things which are fundamental.
Compute infrastructure, I strongly believe, has to be within your country, has to be within your control. That is where your data is getting processed, that is where data is getting stored, that is where your models are being made. Your needs as a country, forget control, your needs as a country are unique. You want to create a voice -based AI because majority of population may not be comfortable speaking in English or writing in English, but they’ll be very, very comfortable talking in their own native language. We all are very comfortable talking in native language. We have a mix of Hindi, English, Malayalam, Kannada, whatever native languages, and we mix up with English. So if we are able to talk to a device in my own native language with my own slang, and the device does all the processing and gives me my answer in real time in my own language, my slang, that is where the real benefit, that is where population scale benefits comes in.
Maybe the model builders of any other country may have a different viewpoint of how they want to adopt AI at a global level. So frontier models are good for those use cases, but for India use cases, possibly I need the focus to go on my use cases which can benefit masses at a larger scale. So both from control point of view that nobody else should tomorrow just switch off my access to digital infrastructure, also from the point of view that my priorities for my citizens can be different, I would rather like to have sovereign compute, right? And some of the models which are taken care of, let’s say, as Minister I think said in the last three, four days, in Devas also, that 95 % of the use cases which India requires can possibly be handled.
So I think that’s the goal. by having models which are 20 billion to let’s say 100 billion parameters. You don’t need to necessarily go for frontier models, trillions of parameters. So we build our compute here. We store our data here. We allow controlled data flow outside. We build the models which are satisfying 95 % of my need. That is what I need to do. But what we can do, and I give you our own example as Yotta. While on one side I’m building…
So Sunil, we’ll just come back to that. We’ll just get everybody else in and then we’ll speak about your examples. I’m just mindful of time. So I think the takeaway is that as far as the infrastructure layer is concerned, as in sovereignty in compute is not only desirable but perhaps possible. And as far as control is concerned, and we should try to have control, but I’ll take that to you, Nasubu. Let’s look at the design layer. I mean, Sunil gave what the infrastructure is about. Sovereignty is also about who makes the rules in terms of how things are designed. And what Sunil said, we work for a very large country like India where there are lots of buildings.
There are lots of builders. But how does it translate to the rest of the world? Maybe some experiences from Kala as well.
Excellent, thank you very much for this when you look at or when you think about let’s say Africa sometimes it’s we are disadvantaged in that we don’t have compute when you look at the computing capacity it’s like at 1 % so already we are at disadvantage before we even leap forward and get ahead but the one thing we have is we have data, we have use cases so when it comes to use cases how are we able to design for our lived realities because we as he said that the language the different things that we are looking at for example when you look at when you look at the local needs, what are the things that we want that we can adopt for example if I look at the use case of health if you look at the at how People in other sectors have been looking at health, for example.
We’ve done a lot of work on designing for our needs in terms of breast cancer. We were able to get data sets from our lived context, knowing that when you look at the composition of the breast tissue for African women, it’s different. So those are the use cases that we need to look at. Because we can be confident and say that, yes, we don’t have compute, but we have the use cases, and that’s the important bit that we need to put into place. That in as much as we are disadvantaged, we have use cases, we have the people who are able to build. That’s the one component that we never talk about. We always talk about, you know, we are getting the data there, someone else is defining the rules.
But we can define the rules by building the tools that actually work for the people in our context and being confident that, you know, once it works for our context, that people are going to use.
That’s right. I think that’s a really powerful statement. because at the end of the day, it’s only local people who have skin in the game who will build for local problems. And I think that’s where actually the opportunity also lies. So I think that’s a very critical intervention. And I’ll take that to you, Seema. So we’ve discussed the infrastructure layer, the design layer. And I think it would be good to get a holistic perspective as far as critical systems and sovereignty in critical systems is concerned, especially because, as Sunil was saying, that while we can certainly try to build, compute, and store it locally, locally, it’s, again, a pipe dream to think that any country can do everything itself.
So there are, of course, questions of supply chains, trusted supply chains, who’s supplying what, and how that control is going to be exercised. So maybe a little bit from your experiences as to what sovereignty means for you, building a large data center, many large data centers now in India, but the rest of the world as well.
So first of all, thank you. I’ll just keep it. I’ve answered this in two parts, and real quickly. So. So. critical question at the critical moment I think it’s very important it’s like an important question for this decade what first question is can you be connected and sovereign yes, I don’t see a problem at all of being sovereign and connected I think over there what is important to understand is basically the strategic control that you need needs to be sovereign and it remains sovereign, I think that’s the definition more on sovereign so you don’t need to really build everything yourself so if you want me to just elaborate around the three what does really government look from its services, so it’s like public services, critical citizen data, financial networks AI systems, unlimited amount, so we’re not talking of outsourcing right over here what we’re talking of is basically critical national infrastructure, I think it’s very important to define not in general but in specific right, what it means so let’s look at three things one one is ownership.
Is ownership very important across all the components in the supply chain and in the critical infrastructure for government? Not really. Not really. I think we need to define the extent to which you want to have ownership. Second is visibility into ownership structures. And third, I think for most important for all countries, whatever, developed, underdeveloped, developing, whatever it might be. I think it’s important for all of us to treat like our digital assets like any other precious asset. And therefore, you have to have policies, guardrails that ensure whatever you have in a sovereign or semi -sovereign infrastructure is not compromised externally and you have a degree of assurance. Where you don’t have geopolitical leverage. I think that is important.
That defines sovereignty to a great extent. So what does it mean for industry? Industry, we have seen some some really good models come up, right? So there is like sovereign infrastructure model. I’ve seen some real good air -gapped, some kind of ring fence environments within the commercial infrastructure, which has been very interesting. And of course, the public -private, which still remains. What does it all mean? It means no national… We are not trying… The goal is not to nationalize. I think the goal is assurance, which is most important. That’s number one. That’s your strategic ownership question. The second is operational efficiency. I think over here, yes, degree of sovereignty does matter. It goes well beyond a few definitions of infrastructure that we have.
I think what is important here is to ensure the extent of operational control, look at efficiencies of operational control, the components within operational control that can be sovereign. And I think that’s what we’re trying to do. is to ensure the extent of operational control, and to operate So what does it mean for industry? We need to build things that are transparent, traceable, and also observable. I think that is the code to your design. That is sovereign design. Then you decide how you want to implement it. So the second thing, what does it mean? It means trust. So trust is not paper -based. Trust can only be engineered, and it needs to be verified, in my opinion.
Okay, I’m quickly coming to the third question. I think you had so many things. Supply chain trust, absolutely. Today, if you look at data sovereignty, it goes well beyond data, digital data. It goes into hardware, chipsets, network components, AI provenance, a whole lot of stuff. So in this case, I think industry needs to basically, you can’t isolate yourself. I do not believe in that. You need to forge very good technology global partnerships. It is important. Again, another degree of trust. The second thing is, of course, you can have some guardrails around it by the government, and you can govern that. I think what is most important in this case is to build some sovereign capacity.
By domestic, which is because in the age of AI, I strongly believe that the sovereign AI compute infrastructure has become a global leverage. So it is important, right? So these are my take. And basically, what I also believe in this is national digital infrastructure for any countries, like a national infrastructure, which could be like a power grid port. or a telecom. So you treat it with that level of whatever you need to do for it. Secondly, a very good guardrails from the government to safeguard sovereignty and govern it. Industry should focus on innovation and not worry too much, whatever you can, not try to own everything, because it slows down your transition and your aspiration of growth.
And this
That’s great. And I think one underlying point that you made across these three is of trust, because at the end of the day, you can’t build everything yourself. Sovereign nations don’t do things themselves, even in a non -AI analog world, so it’s not that you’re going to do everything yourself. But sovereignty is only partly what we say, but more importantly, I think is what we do. And so I want to take that to each of you in terms of what you are doing in your own domains, in your own companies, and where that line, that what am I going to do? myself, what am I going to do with somebody else and if so how will I ensure that this person is trusted and I have control.
So Sunil you were saying about Yotta and what you do briefly so that we can get the others in.
Yeah sure. So I’ll just go by the actual example which we have sort of done in the last two years or so. Last week we inaugurated and made open to the world that India’s AI language platform which I think every government entity is using, Pashini, we actually migrated that from a hyperscale cloud operator to our cloud. It’s a combination of a whole lot of general compute services and AI and GPUs and all on which all those language models are working which are giving real -time translation services. Now their purpose considering that it is a digital public infrastructure they were very very clear that at no point of time we want to be dependent on the platform service of a hyperscale operator because that makes it make stickiness that you cannot come out of that platform.
Whether it is a hyperscale platform or for that matter Yotta’s platform they don’t want to remain dependent on only one entity. They want it to be independent. They wanted a choice. we ended up not only giving them the physical infrastructure which was obviously local in my data center but we ended up creating almost I can say 30 or 40 different technologies we developed put it on their virtual machines in their environment in my data center and brought them into their control they were not using PaaS anymore. At the last when everything was going live we suddenly realized there is one component NVCF which is a NVIDIA’s software tool which was still running on NVIDIA’s platform somewhere in US and it was not running in India and then they said even though it is all fine NVIDIA is my biggest technology supplier giving me GPU software, everything but they said no this cannot go this software component is very critical for this whole structure but it has to come within your control into my environment so what we have done after that after everything was done was also to and NVIDIA agreed to open source that part of the software brought the software into our environment and now it is available and it is running within my control this example is telling.
I’m just in same breath, I’m telling you same thing. I’m using the best of the foreign technologies. I’m using Nvidia’s technologies. I’m using, of course, open source technologies. I’m using Microsoft technology. We have great partnership with Azure. I’m using Amazon’s technologies. But I’m not using these technologies in the public cloud. I’m using their technology stack within my ring -fenced walls, within my GPU and CPU compute infrastructure. The access control of these technologies firmly lies with me. No third party is able to log into my system and control or dictate what will be running and what will not be running. And that, I think, is the real balance, that you use the best technologies. These guys have spent hundreds of years, put billions of dollars in creating great technologies.
We must benefit from that. But you use these technologies within your environment, within your control.
That’s right. So I think partnership and not dependence. What I’m interested in, Nasubo, and I’ll come to you on this, is what you’re doing at Kala. Because what Sunil is saying may work, say. In a setting like India, where… tell NVIDIA perhaps that some part can be stored on something locally. But I’m thinking of Malawi, I’m thinking of Lesotho or Eswatini, I’m thinking of smaller Southern African countries, which also will want to use AI for solving local problems. And so what does it look like from your perspective, having done so much work in Southern Africa?
When you look at, let’s just first ground what Africa has, right? How are we going to use compute that also allows for offline? That is one of the use cases we are looking at, because in as much as digital connectivity is everywhere in Africa, it’s up to like 50%. So how do we also ensure that people are able to use? So one of the ways that we do this is, one, we are working with global partners to give us compute, but at the same time we also want to buy compute, for ourselves, because the… conversations that he’s talking about in the rules, creating the rules and the structure can only be done once you also understand what is happening.
So at Kala, we are also offering compute to different innovators. And if you go to our stand in Hall 14, you are able to interact with different African innovators from the AI village who are building AI innovations. And part of those innovations are innovations that allow for offline access. That is the one thing that we need to be cognizant of. We need to understand how we need to work practically. That’s something that Kala is actively building, actively championing for. So that when we’re having even conversations with government, we are going to them and saying, yes, compute is something that we may not have. But if you approach, let’s say, big tech and you’re talking about offering compute, offering us…
being sovereign, this is what it means. So we are also having conversations with different African governments to talk about what we are learning, what people are building, and now having once they have their understanding now we can continue ensuring that our use cases are well represented. Because if we just take things that are dictated to us without having like a perspective it means that we are building for exclusion. And for us we want to ensure that all voices are well represented, including the people who are offline, who want to use AI for solving use cases in our sectors.
That’s right, and I think this is resonating greatly with the fact that you’re building for offline, because when we were doing Aadhaar in India and the legal framework for Aadhaar, which underlies all our DPIs, one of the key game changers was for moving from online authentication to offline verification, because we realized that that was a big need. So this is where the global south, I think, needs to learn from . each other because these problems are somewhere similar. Last word perhaps to you Seema on this in terms of your actual experiences in ensuring that you have control over whatever is within your ring fence but what’s outside is something that you trust and you think will further the goal of sovereignty and sovereign AI as you mentioned.
Something from your practical experiences.
Let me just give you what most of us are doing and why it is pertinent and it’s important. We are building currently we are building for demand so it’s like a gigawatt AI factories huge amount of compute huge amount of data centers it has to be done responsibly in all ways and a lot of money. I think what’s important is how does this work between the government and the enterprise. I think that is the recipe for success. So there are three, four things which I have a take. One, of course, is basically the policies need to evolve along with the infrastructure. They are not based at the same. So that, I think, is important. The second thing is government must lay the sovereign guardrails.
It’s all spoken about, but you don’t have them. So it’s very difficult. Third, I think what is also important in every country to help the industry to build that capacity is also give, you know, not only have a long -term stability of your policy, but also look at commitment so that private enterprises, private industry is confident of building that huge capacity for you. I think that’s very, very key. And last but not the least is definitely look at security and regulatory, not at point -in -time checks, but move it to a continuous verification process. But this… This will ensure your sovereignty is implementable. You can also, you know, kind of enforce it. and get the best results out of it in terms of outcome.
My closing remarks. One, of course, I did speak about before in terms of how you treat this asset. You’ve got to treat it like any other national asset. Second is government needs to extend that hand in becoming an absolute sovereign partner or a public -private partner to the industry. And third is industry needs to really focus on innovation, innovation, scale, time to value, time to market. I think that’s where your soul energy should go. And last but not the least, this is a core accountability for every country. It can’t be one over the other, right? And that will ensure you safeguard your national interest and also do scale and progress without compromising your transformation times and things like that.
So you’re not left behind. See, AI is a journey where we don’t want any country to be left behind. One, lack of… resources, lack of definitions, security, sovereignty, access. I think we need to have that. I really like the theme. It says welfare for all and happiness of all and that should really be the case if it is so very transforming in nature.
That’s right and I think if we were to quickly wrap up with some takeaways because we see that the purpose of this session was that data sovereignty shouldn’t be just theoretical, a slogan. It has to work in practice and what I took away from the three of you who are actually walking the talk on data sovereignty is A, in terms of the role of the market is essentially to build sovereign AI in whichever country you may be in and build it yourself, store locally and ensure that you have trusted partners when you are partnering with someone because it’s obviously futile to try to even think about doing anything yourself. As far as governments are concerned, again, this is a, I like the word you used, co -accountability, this is a partnership.
And I think government has to build guardrails, but hand in hand with both the bazaar and the samaj, that’s the society and the market. And as far as the samaj is concerned, the society, and Kala mentioned that about, at the end of the day, we mustn’t forget that what we are trying to do is solve real problems for real people. So like she mentioned that the breast tissue in an African woman is different in somewhere else, that’s the person whom we are trying to serve. And I think that that is what is imperative for all of us to do. And I think it’s appropriate to end with what Gandhiji said, that we must think about the last person in the line.
And I think when we are talking about AI, just because we are in a kind of modern technocratic age, we shouldn’t forget that it’s that last person, the man or woman in the queue. The man. The most unfortunate who we must think about. because at the end of the day that is for whom AI is built and that is for whom we are talking about sovereignty so we leave it there thank you very much ladies and gentlemen and thank you to my panelists for a wonderful session thank you
please could you all please wait for a second I’ll just hand over your memory I request everyone to please settle down we will be bringing the next session very soon thank you
“The moderator opened the session by defining digital sovereignty as “who gets to make the rules” and noting that the debate usually centres on the location of data and compute.”
The knowledge base records the moderator Arghya Sengupta framing the central question as “who gets to make the rules” rather than where data is stored, confirming the report’s description.
“Sunil described migrating the national AI language platform Pashini from a hyperscale public cloud to Yotta’s locally‑controlled data centre.”
Sunil Gupta is reported to have migrated the Bhashini language platform from a public cloud to a sovereign data centre, which aligns with the report’s description of moving a national AI language platform to a locally-controlled facility [S84].
“Sunil argued that sovereignty does not require total isolation but control over strategically essential compute infrastructure.”
The discussion notes that sovereignty in compute is desirable and possible, reflecting Sunil’s view that control, not isolation, is the goal [S21].
“India’s linguistic diversity creates a need for native‑language voice AI capable of handling regional slang and real‑time responses.”
Other sources highlight India’s multilingual landscape and the importance of voice-first, multilingual AI for health and broader applications, supporting the claim about linguistic diversity driving AI needs [S81] and [S83].
“Using foreign technologies like NVIDIA, Microsoft, Amazon is acceptable if they run inside a sovereign, ring‑fenced compute stack.”
The edge-cloud discussion emphasizes distributing compute and keeping critical workloads within sovereign infrastructure, providing context for using foreign tech inside a ring-fenced stack [S86].
The panel shows strong consensus that digital sovereignty is about rule‑making and control, not full self‑sufficiency; it must be realized through collaborative governance, locally hosted compute, culturally relevant AI, and trusted partnerships with transparent supply chains. All speakers align on these pillars, indicating a shared vision for policy and implementation.
High consensus across all speakers, suggesting that future initiatives should prioritize domestic compute infrastructure, co‑accountability frameworks, local data/model development, and engineered trust mechanisms to achieve practical sovereignty.
The panel largely agrees that digital sovereignty hinges on rule‑making, trusted partnerships and locally relevant AI. The main points of contention revolve around how much physical ownership and domestic compute are required versus how much strategic control and external collaboration suffice, and whether low‑compute regions can achieve sovereignty primarily through data and partnerships. These disagreements highlight the need for nuanced policy frameworks that balance domestic infrastructure investment with open, secure global partnerships.
Moderate – while there is broad consensus on the principles of sovereignty, the speakers diverge on implementation specifics (ownership vs strategic control, extent of local compute, and the role of nationalisation). The implications are that policy makers must craft flexible strategies that accommodate differing national capacities and avoid a one‑size‑fits‑all approach.
The discussion was shaped by a series of pivotal insights that moved it from a vague slogan of ‘data sovereignty’ to a concrete, multi‑layered framework. Sunil’s opening definition anchored the debate in the necessity of sovereign compute, while Nasubo’s emphasis on local data and use‑cases reframed sovereignty as a function of contextual relevance rather than raw hardware. Seema added depth by distinguishing strategic control and trust from outright ownership, introducing governance mechanisms. Sunil’s real‑world migration example demonstrated how these principles can be operationalized, and Nasubo’s focus on offline capability highlighted practical challenges in low‑connectivity regions. Finally, Seima’s call for evolving policies and continuous verification tied the technical and design considerations back to sustainable governance. Collectively, these comments redirected the conversation toward actionable strategies—balancing partnership with control, leveraging global technology within national boundaries, and ensuring inclusive, trustworthy AI deployment.
Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.
Related event
