The sovereignty debate: Looking beyond data residency

Across boardrooms, regulatory bodies, and policy circles, India’s conversation about AI sovereignty has converged on a familiar concern: where does the data live? Keep it onshore, the argument goes, and sovereignty is secured. It is a reasonable starting point. But if it becomes the endpoint, India risks answering a 2015 question for a 2030 problem.

AI sovereignty, in its fullest sense, is the ability of a nation or organisation to control its entire AI stack: infrastructure, data, models, and the operations that run across them. By that definition, data residency is necessary, but not sufficient.

The harder questions remain underexplored. Who controls the models that process the data? Who owns the encryption keys that govern access? Who operates the control plane that orchestrates every service?

Having spent decades working across AI research and enterprise deployments, I have seen this shift first-hand. The systems being built today are fundamentally different from those that came before, and the frameworks used to govern them must evolve accordingly.

Research from the IBM-IndiaAI study underscores this demand—77% of Indian enterprise leaders are calling for secure, affordable cloud infrastructure. Yet many organisations still face gaps in governance, data readiness, and integration that slow meaningful AI adoption. Sovereignty cannot be retrofitted. It must be built into the architecture from the start.

Sovereign by design, not by contract

India’s next phase of AI governance must be anchored in a simple principle: Sovereign by design. Sovereignty should be enforced architecturally, continuously, and with full runtime visibility not asserted contractually and audited after the fact.

Consider what a contractual sovereignty commitment actually means in practice. If a vendor accesses your data without authorisation, you find out later, engage lawyers, attempt to prove the breach, and seek remedy in a jurisdiction that may or may not be your own. By then, the data has already been accessed. The model has already processed it. Whatever was inferred is already known.

An architecture that makes unauthorised access technically impossible operates on a different logic. There is no breach to discover, no court to approach, no remedy to seek—because the action could not have occurred in the first place.

This is not a stronger version of trust. It is a different category: provable control. For sectors such as finance, defence, and public infrastructure, that distinction is critical. Sovereignty must be demonstrable—by design and in real time—not assumed through agreements.

When AI stops advising and starts acting

This shift becomes more urgent as AI moves from advising to acting.

For years, enterprise AI has operated with a simple safeguard: the model advises, the human decides. Imperfect, but it preserves accountability.

Agentic AI changes that equation. AI agents do not just recommend; they execute. They access systems, trigger workflows, invoke APIs, and complete transactions autonomously, often at machine speed. By the time review is possible, the action is complete—and sometimes irreversible.

This is already emerging across sectors such as financial services, healthcare, and enterprise operations. The sovereignty challenge it exposes is not about where data resides, but who governs what the AI does.

As Indian enterprises scale AI adoption, the risk is clear. Without strong governance, organisations do not build sovereign AI. They build automated systems of liability.

The critical question is no longer just whether we can build AI, but who controls its actions and how that control can be proven.

What India must do next

India is not starting from scratch. The DPDP Act, the India AI Mission, and sustained investment in digital public infrastructure signal strong intent. AI could contribute over $500 billion to the economy by 2030, making sovereignty not just a regulatory priority, but an economic one.

Realising that potential now requires coordinated action on three fronts.

For enterprises: The more useful question is whether you control the environment in which data is processed. Do you control the inference layer, the encryption keys, the audit trail, and the control plane? If the answer to any of those is “our vendor does,” that is the gap to close.

For regulators: The DPDP Act provides a strong foundation but must be complemented with clearer frameworks for AI model governance, agent execution environments, and continuous auditability, particularly in critical sectors.

For the technology ecosystem: Sovereign by Design must be treated as a product capability, not a compliance layer. Organisations that embed provable control into their architectures will be more trusted and more resilient as regulatory and geopolitical environments evolve.

India has a rare opportunity. Few countries combine regulatory intent, engineering depth, and market scale in the same way. This positions India to not just adopt AI but define how sovereignty in AI is built and enforced.

But ambition is not architecture. The decisions being made today will determine whether India participates in the AI era as a sovereign actor or as a capable tenant on someone else’s infrastructure.

Sovereignty in the AI era will not be defined by where data resides. India has every reason to lead. The question is how quickly intent becomes architecture.

(The author is general manager, IBM Software India and Software Innovation Lab. Views are personal.)