CYBER TECH | The terms India has not yet demanded

Artificial intelligence is undergoing a reclassification that most policy commentary has not yet caught up with. It is ceasing to be primarily a software industry and becoming something closer to foundational infrastructure — a layer of the global economy comparable to semiconductors, telecommunications networks, or energy grids. Like those systems, AI is not simply adopted by countries. It is negotiated, positioned within, and — for those with the foresight to act — shaped. The countries that helped define the architecture of earlier infrastructure layers retained structural advantages that compounded for decades. Those that arrived late as consumers found the terms already set.

This is the frame through which India's current moment in AI deserves to be understood.

Most Indian commentary on artificial intelligence still asks a consumer question: how can AI improve our agriculture, healthcare, education, governance? These are legitimate concerns. But they are not a strategy. The strategy question runs in the opposite direction — how does India position itself within the emerging infrastructure itself, and on what terms does it participate?

The answer begins with an observation that domestic policy circles have not fully absorbed. India is already structurally embedded inside the global AI production chain. Indian engineers contribute significantly to the semiconductor design work that underpins frontier AI hardware. India is among the largest suppliers of data annotation and reinforcement learning from human feedback labour — the invisible, iterative human work that makes large language models behave usefully. India possesses large-scale datasets in languages serving populations that frontier labs cannot train on themselves. And India represents one of the largest active user markets for every significant AI platform currently operating at scale.

India is not outside this system. The question is whether its participation generates durable strategic value, or merely serves the system's current operational needs.

Here a structural risk deserves stating plainly. The annotation and human feedback work that constitutes India's deepest current participation sits at the low-margin end of the AI value chain. The workers are numerous; the margins are thin; the work is increasingly at risk of automation by the very systems it helps train. Ownership, compute infrastructure, and intellectual property, meanwhile, are accumulating at the frontier — in the companies, data centres, and research institutions of a handful of countries. This mirrors a pattern familiar from earlier technology waves: nations that participate in assembly without capturing design eventually discover that the value has consolidated upstream, in places and institutions beyond their influence. The risk for India is not exclusion from AI. It is integration on terms that reproduce dependency rather than build leverage.

The compute question makes this structural asymmetry concrete. India currently lacks frontier training infrastructure, advanced semiconductor fabrication, and sovereign control over the large-scale GPU ecosystems that serious AI research requires. Indian institutions depend on foreign cloud providers for capabilities that define who can build what. Attracting data centres to Indian soil addresses this partially — but infrastructure hosting and infrastructure sovereignty are

not the same thing. A data centre built and operated by a foreign company, on foreign cloud architecture, running foreign models, is not a national asset in any meaningful strategic sense.

The most consequential institutional response would be a public compute and data commons — a nationally governed research infrastructure, modelled in spirit on how India built its digital public infrastructure, from Aadhaar to UPI. The UPI architecture succeeded because it was designed as a public layer on which private innovation could build, rather than a government product competing with private ones. A sovereign compute initiative could follow the same logic: open access for Indian research institutions, structured to generate domestic intellectual property, governed transparently, and insulated from short-term political interference through an independent institutional design. The EU's AI factories programme and DARPA's long-horizon research funding offer relevant structural precedents. The goal is not to replicate frontier labs but to prevent permanent dependence on them.

Why has India not already demanded stronger terms? The answer lies in a domestic political economy that rewards visible wins. A data centre announced, a foreign headquarters opened, an investment figure cited in a press release — these are legible achievements in an electoral cycle. Technology transfer conditions, equity participation requirements, and sovereign IP arrangements are harder to communicate, slower to materialise, and more likely to complicate negotiations that governments prefer to conclude smoothly. The incentive structure currently favours India as a welcoming destination over India as a strategic interlocutor. That preference is understandable. It is also, over time, expensive.

History offers a consistent lesson. Countries that fail to shape foundational technologies during the period when their architecture is still being negotiated rarely recover meaningful leverage once standards consolidate and dependencies calcify. The window is not permanently open. The question India faces — not abstractly, but in the specific deals being signed and the specific terms being accepted right now — is whether it intends to be a participant in this infrastructure or merely a location for it.

That distinction, once lost, is rarely recovered.