From AI Momentum to Sovereign AI Method: What Emerged at the Mint Sovereign AI Summit 2026

India's AI conversation has changed shape. The question is no longer whether AI can work. It is whether AI can hold up once it moves into real systems: public services, enterprise workflows, and national-scale infrastructure. When AI becomes usable at scale, responsibility stops being a side constraint and becomes part of performance.

That was the starting point for the Mint Sovereign AI Summit 2026, presented by Dell Technologies, held in New Delhi on January 23 as an officially aligned pre-summit forum ahead of the India AI Impact Summit. The afternoon was designed less as a showcase and more as a working session: a concentrated attempt to clarify what sovereign AI requires when the ambition is population-scale deployment.

Alokesh Bhattacharya, Deputy Managing Editor, Mint, set the scene in opening remarks by naming the shift that has brought these questions into the foreground: “After going through several cycles of promise and disappointment, something fundamental changed over the last few years. AI became usable at scale. It moved out of research labs into everyday life and work into hospitals, classrooms, farms, enterprises, public services, government and media, especially across social and digital. Needless to say, that shift brings enormous opportunity. But it also raises harder questions.”

Those harder questions - trust, governance, inclusion, and control - shaped the rest of the day.

Sovereign AI as execution, not aspiration

In the welcome keynote, Manish Gupta, President and Managing Director at Dell Technologies India, framed sovereign AI as a test of execution rather than intent: whether India can convert its current momentum into systems that endure once they are exposed to real-world complexity. His argument rested on a simple premise. If AI is to be designed for India's realities, it has to clear a much higher bar than experimentation.

“The intent is to turn India's AI momentum into something durable, into something scalable and something that's designed for India's realities. When we say AI by India, for India, it has to mean three things in particular. First and foremost, it has to work at India scale.”

That scale, he emphasised, cannot be partial. It has to function across population-sized deployments, not just controlled environments. But scale alone is insufficient. The second requirement is trust - not as a slogan, but as an architectural choice, embedded into infrastructure and foundational models. That means clear accountability for how data is handled, how models are used responsibly, and how systems remain secure while producing outcomes that can be relied upon over time.

The third requirement, he argued, is inclusion. India's users are not a single, homogeneous segment. Systems have to work across languages, access conditions, and uneven levels of digital literacy if they are to function as public capability rather than gated advantage.

The consequences of getting that wrong are structural. Systems that fail to account for how people actually live, work, and access technology do not widen capability; they concentrate it. “If AI can't meet people where they are, it won't really bring in true public capability. It will continue to remain a privilege. From our perspective as Dell, this is where the conversation on sovereign AI becomes very real. Sovereignty is not just about where a model is built. It's about whether an organisation or a nation can run AI with control, with resilience and with confidence.”

That definition - sovereignty as operational control and resilience - became a reference point that echoed through later sessions.

Sovereign AI on India's public rails

Akhil Kumar, Managing Director and Chief Executive Officer, Digital India Corporation (MeitY), used the special address to connect sovereign AI to the institutional arc already underway: Digital Public Infrastructure, the IndiaAI Mission, and the governance frameworks that are now being built around data and deployment.

He framed sovereign AI in continuity with national self-reliance thinking, shaped by supply chain disruption and geopolitical volatility, while also resisting the idea that "sovereign" implies closed. “When we may talk about anything sovereign that doesn't mean it will be exclusively for Indians. It will be for the world. It will be a global good which will be inclusive, accessible to all.”

That mattered because it reframed the day's premise: sovereign AI isn't an inward turn. It is a capability play - built for Indian realities, but exportable precisely because it works under constraint.

What breaks when AI meets real scale

The plenary conversation - From Adoption to Advantage: Scaling AI in India with Trust, Speed and Inclusion - was where the day moved from framing to friction. What surfaced repeatedly was a shared diagnosis: model capability is no longer the only gate. The bigger constraint is whether systems can absorb AI without losing reliability, trust, or control.

Deepak Bagla, Mission Director, Atal Innovation Mission, NITI Aayog, Government of India, argued that India's advantage lies in adoption velocity: when systems work, the country scales them fast. "I think the biggest strength of India and every Indian as an individual is the ability to adopt and adapt to new technology. We adopt and we adapt. I mean look at the real time transactions. 2021, you did 41% of the world's transactions. In 2024, you did 188 billion. So India moves exceptionally fast." But he was clear that speed alone does not guarantee advantage. It only compounds value when innovation pathways remain broad-based rather than concentrated in a few pockets.

Suresh Khadakbhavi, CEO, Digiyatra, offered a grounded example of what "trust as design" looks like when deployed at scale. Digiyatra's architecture, he explained, avoids central data accumulation altogether. "We do not have any of the 20 million user bases, passengers, data - all the data rests with you in your own phone. If we don't have anything to be hacked, what will you hack? If somebody has to hack into Digiyatra personally identifiable information of users they have to hack the 20 million users' phones because that's where the credentials lie!" The point was not only privacy, but governance: designing systems so that failure modes are contained rather than amplified.

Tarun Dua, Founder, E2E Networks Ltd., brought the infrastructure layer into view: sovereign AI, he argued, must be buildable by learners and developers, not only consumed by large enterprises. "AI labs as a service is like a thin layer which we have created for any kind of learner or developer to be able to launch their Jupyter notebooks and start learning AI very, very quickly without having to scrounge for GPUs or building workstations." His broader principle was that compute strategy has to start with usage reality. “AI computing is not about what computers want. AI computing is all about what the users want.”

Kapil Bardeja, CEO and Founder, Vehant Technologies, spoke from the perspective of AI systems deployed into environments that do not tolerate fragility. At Vehant's scale, AI is judged by throughput and uptime, not elegance. "The architecture which you have for, let's say, a 5,000 camera system - and when you scale it to 100,000 cameras or more - the scale of architecture is very challenging. So we do ensure that complexity of the system is taken care of in the design stage." His point was succinctly made: once AI systems reach this level of deployment, architectural shortcuts taken early become liabilities later.

Neeraj Arora, Field CTO, Conglomerates, Dell Technologies, distilled the system challenge into a single constraint that cuts across technology and governance. "For sovereign AI, from technology point of view, the first and foremost important point is to think about sovereign controlled data pipelines. Data cannot just flow like a wild river, it has to be channelized properly." At scale, sovereignty is exercised through control over how data moves, where it is allowed to accumulate, and how tightly it is governed as systems evolve.

Taken together, the plenary made one thing clear: the moment AI leaves pilots and enters production, progress is no longer driven by models alone. The real work becomes orchestration - aligning stakeholders, data, infrastructure, and guardrails so that systems can scale without breaking trust.

What DPI teaches about trust at scale

The fireside conversation - Building Sovereign AI: What Happens Next - brought the discussion closer to the centre of national infrastructure, using UIDAI as a reference point for what trusted scale looks like in practice.

Bhuvnesh Kumar, CEO, UIDAI, challenged the assumption that population-scale systems require expansive data capture. Aadhaar, he argued, is large in reach but deliberately lean in what it holds. “All 140 crore people in India have an Aadhaar. But in terms of data, we are a very, very thin organisation. If you have your Aadhaar, then you will see that the Aadhaar is generated on the basis of your biometrics. They are captured once for generating and deduplicating with others. And it is kept secure forever. Other than that there are four fields: name, date of birth, gender, address. Mobile number and email are optional. And that's all data that we have. This is all that we have.”

The implication was subtle but important. At scale, trust is often achieved not by collecting more information, but by limiting what must be collected, securing it tightly, and keeping the system intelligible to those who depend on it.

In the same session, Manish Gupta returned to the operational definition of sovereign AI, this time laying out the underlying conditions required to make it durable. “What are the key tenets of sovereign AI? It's got to be within your borders - infrastructure, data - so that all of that is within the sovereignty. Number two, it's got to align with the local ethos, local values and, more importantly, the local laws and governance rather than the global by default governance.”

He linked those tenets directly to execution risk, arguing that sovereign AI cannot be built on outdated foundations. "Legacy systems are not going to help you when you have to bring everything on the same platform." Modernising the digital backbone and clarifying governance early, he suggested, is what allows systems to scale without repeated resets once dependencies form.

Together, the fireside sharpened a theme that had been building through the day: sovereignty is not only about scale or control, but about sequencing. The choices made early - around data minimisation, infrastructure modernisation and governance alignment - determine whether AI systems can move faster later, or whether they stall under their own weight.

Sovereign AI as an operating discipline

While Masterclass Part I - Building India's Compute Backbone - focused on infrastructure choices, Masterclass Part II turned to a different question: what it takes to operate AI systems once they are deployed, evolving, and relied upon.

Rishi Bal, CEO, BharatGen, framed sovereign AI not as a label, but as a set of guarantees that only matter over time. "What is sovereignty? Sovereignty is the guarantee that whenever you need AI, it will be available to you. Sovereignty is that you should know everything that has gone into your AI. And then the third part is that you should have the guarantee that you can service your own AI." In this context, sovereignty isn't tested at launch, but when systems need to be maintained, upgraded, audited, or adapted to new conditions.

That emphasis on serviceability led directly to the question of capability. For Rishi Bal, sovereign AI cannot be sustained if a country remains dependent on external expertise to understand or modify its own systems. "So we're not just interested in creating the AI, we are interested in creating the creators," he said. “Because when these hundred kids graduate from their respective colleges, they will already be trained in how to create AI.”

The point was not workforce development for its own sake. It was about continuity. Systems can only remain transparent and governable if there are people inside the ecosystem who know how they were built and how they behave.

The masterclass reinforced that at scale, trust is not static. It has to be operated - through availability, visibility into system components, and the ability to service AI without external dependence. Those guarantees rest as much on human capability as on infrastructure and policy.

Turning AI momentum into method

The most useful outcome of the afternoon didn't take the shape of a single announcement, but rather, a clearer, more disciplined understanding of what sovereign AI actually demands once ambition gives way to execution.

The opening premise held: AI is now usable at scale, and that changes the nature of the challenge. By the end of the day, the conversation had narrowed into something more exacting. Sovereign AI is not a model milestone or a policy declaration. It is an operating discipline - control over how data moves, infrastructure choices that match real workloads, governance embedded early enough to accelerate rather than stall, and inclusion treated as a design constraint rather than a downstream fix.

What the summit made visible is that these choices are already being made, whether explicitly or by default. Systems are being deployed. Dependencies are forming. Architectural decisions are hardening into long-term paths.

If India's sovereign AI story is to be written as capability rather than aspiration, the decisive factor will not be momentum alone, but method. Alignment between public rails and enterprise execution. Between speed and trust. Between the scale India can reach - and the systems it can sustain once it gets there.

Note to the Reader: This article has been produced on behalf of the brand by HT Brand Studio and does not have journalistic/editorial involvement of Mint.