What does sovereignty mean in AI?
When we hear the word sovereignty, we think of independence, especially in India, where it is tied to our history of self-governance. So sovereign AI sounds like building everything ourselves: our own models, infrastructure and technology stack. But is that what sovereignty requires?
The dictionary defines sovereignty as the authority to govern oneself without outside interference. It says nothing about having to build or own everything. To govern AI is to control how it is used, audit what it does and inspect how it works.
That authority can sit at three levels: a country, a state or an enterprise. The question is the same at each: who decides how AI is deployed, managed and changed? For enterprises, the answer has to be the enterprise itself.
Sovereignty does not mean self-sufficiency
AI depends on chips, computing infrastructure, energy, data and talent from different parts of the world. Few countries, and no enterprise, can manage all of it alone. A Brookings analysis calls the practical alternative as managed interdependence: retaining the ability to make independent decisions while accepting external dependencies.
Researchers who examine four pillars, data, compute, models and norms, find India stronger in data, compute and norms but less autonomous over foundational models.

Depending on an external provider is not necessarily a problem. Not understanding that dependence, or having no alternative when circumstances change, is.
Why are we discussing this now?
For decades, enterprises bought software as a product and ran it on their own terms. Their data stayed ring-fenced and secure, and nothing flowed back to the maker other than subscription costs. Consumer platforms were different: millions of people used them with no idea how their data might be used.
AI brings that uncertainty into the enterprise. It processes customer records, financial data, contracts and intellectual property. Once that information enters an AI model, who can access it? Is the data retained or is it used for training?
Where your data sits is not who controls it
Residency tells us where data is stored. Localisation is a legal requirement to keep it within a jurisdiction. Sovereignty goes further: who controls access, holds the encryption keys and has operational authority?

Under Section 16 of India’s DPDP Act, personal data can be transferred abroad unless the government restricts specific countries (DSCI’s analysis). Storing data in India, therefore, does not mean an enterprise controls it.
Every enterprise should know which dependencies it is accepting, and which layers it cannot afford to lose control over.
Three things to take away:
Sovereignty means the authority to govern, not the ability to build.
AI makes data control more complex because data can shape models, outputs and decisions, not just be stored or processed.
Storing data in a country is not the same as controlling it. Ask who can access it, use it and govern it.

That is a more practical starting point for AI sovereignty.What part of AI sovereignty do you want us to uncover next? Let us know your thoughts in the comments.
Sources
Is AI sovereignty possible? Balancing autonomy and interdependence, Brookings, Feb 2026
Sovereign AI: Rethinking Autonomy in the Age of Global Interdependence, arXiv, Nov 2025
MIT, The GenAI Divide: State of AI in Business 2025, via To AI or Not to AI
Cross-border flow of data under the DPDP Act and Rules 2025, DSCI

Comments