Who owns India’s AI memory?

In Part 1, we established that AI memory is not evidence of a learning mind — it is a retrieval architecture whose costs fall disproportionately on the user. Here, the argument acquires a sharper, more consequential edge for India.

Every memory your AI system holds about you sits on physical infrastructure. That infrastructure has a geography, a jurisdiction, an ownership structure, and a government with the legal authority to compel access to what runs on it. When an Indian professional’s work patterns, strategic decisions, professional relationships, and organisational knowledge are stored, processed, and retrieved through AI platforms hosted on foreign-owned infrastructure, the conversation is not merely about subscription economics. It is about where India’s most sensitive cognitive capital resides — and who, ultimately, controls access to it.

This is not hypothetical. China’s launch of the world’s first commercial underwater data centre off Hainan — and its more recent Shanghai Lingang facility, wind-powered and housing nearly 2,000 servers 35 metres below the ocean surface — is not primarily an engineering story. It is a sovereignty story. Beijing understands, with a clarity that most democratic governments have yet to match, that the physical infrastructure of AI memory is strategic terrain. Whoever owns the servers owns the context. And whoever owns the context can, under the right legal or political conditions, look inside it.

India’s Peculiar Position

India sits at a peculiar intersection of this reality. It has one of the world’s largest populations of AI users. It has a knowledge workforce whose professional output has materially fed the training pipelines of global AI systems — contributing, without compensation or consent, to the models now being used to displace them. And it hosts almost none of the infrastructure on which this exchange runs. India’s citizens are generating memory, paying to maintain it, and storing it on servers they cannot audit, in jurisdictions whose data access laws are not aligned with Indian interests.

Consider the scale of what this means.

When four million Indian IT professionals interact daily with AI systems — debugging code, drafting strategy, processing client data, annotating decisions — they are not merely using a tool. They are feeding an intelligence infrastructure owned by others, financed by others, and governed by others. The value flows outward. The cost flows back in the form of token charges and subscription fees. The memory — the accumulated professional intelligence of India’s knowledge class — remains on foreign soil.

There is a word for an economic arrangement where raw material flows out and the finished product flows back at a markup: colonialism. Indigo grown and processed in India once travelled to American mills to dye the denim that came back to Indian shelves as branded jeans, sold at a markup Indian growers never saw a share of. The AI economy is building a digital variant of precisely that structure, with cognitive labour replacing physical commodities and cloud infrastructure replacing the trading post. The difference is that this time, the extraction is invisible, frictionless, and dressed in the language of convenience.

A Familiar Pattern

This is not, it must be said, without precedent. Since the year 1990, Indian knowledge workers on H-1B visas have contributed an estimated billion dollars annually into American social security systems — mandatory deductions, non-negotiable, non-portable — for decades, without the reciprocal benefit portability that the United States extends to over thirty other nations through totalisation agreements it has simply never found the urgency to sign with India. The worker pays in. The system retains the benefit. The worker departs with nothing. The AI economy works the same way, just faster and bigger — instead of taking a cut of wages, it now takes people’s ideas and knowledge. The system has
changed. The unfairness has not.

What India Must Demand

The policy response required here is not protectionism. It is parity.

India should be negotiating, at every bilateral and multilateral trade forum, a framework for cognitive labour attribution — a recognition that the intellectual output of a nation’s workforce, when it feeds AI training pipelines, constitutes a form of data contribution that carries reciprocal economic and legal obligations. This is the same logic that underlies data localisation regulation; it simply needs to be extended from personal data to professional knowledge output.

India should be mandating AI memory sovereignty standards for platforms serving Indian users at scale — requiring that the infrastructure processing and storing Indian users’ context meet minimum data residency, auditability, and access-restriction requirements. Not because foreign AI is hostile, but because the risk is real and growing: memory is no longer merely a technical feature. It is an economic and, increasingly, a geopolitical product — and treating it as anything less would be a costly mistake.

And India should be investing, with the same strategic urgency it brings to naval infrastructure in the Andamans or satellite telemetry in the Indian Ocean, in the domestic AI infrastructure required to give Indian users a genuine alternative — one where the costs of memory are transparent, the jurisdiction is domestic, and the accumulated cognitive capital of India’s workforce stays within reach of Indian laws.

The data centres being built under the ocean off Shanghai are not cooling servers. They are anchoring a new kind of strategic geography — one defined not by territorial waters but by whose infrastructure your thoughts run through. India, which has spent a decade building a physical arc of presence across the Indian Ocean, cannot afford to give up the digital equivalent of that terrain by default.

A Question Worth Asking

The public debate has focused intensely on whether AI should remember us. Perhaps the more important question is whether the economics and geopolitics of that memory should continue to remain largely invisible to the very people financing it.

As artificial intelligence becomes more deeply woven into our personal and professional lives, transparency should extend beyond algorithms and privacy policies to include the commercial architecture of memory itself — and the sovereign geography of the infrastructure on which it runs. Because memory is no longer merely a technical feature. It is becoming an economic product, a geopolitical asset, and a question of national interest.

And every remembered conversation is a quiet reminder that someone, somewhere — in a server room you will never see, in a jurisdiction you did not choose — is paying to keep it alive. The question India must now answer is whether it intends to remain a tenant in that arrangement, or build a room of its own.

END OF ARTICLE