A Financial Times documentary released last week — India’s AI Gamble, reported by Krishn Kaushik — poses an uncomfortable question with unusual candour. It asks whether India’s much-celebrated position in the global AI economy is genuine participation or elaborate servitude. Factory workers in Tamil Nadu wearing GoPro cameras on their foreheads to generate training data for humanoid robots. Homemakers paid ₹1,000 for three hours of filmed domestic activity that will teach machines to mop floors and fold clothes. A data annotation industry of fifteen years’ standing that has kept the lights on for millions of first-generation graduates in small-town India, while the intellectual property — and the billions — flows entirely westward.

The FT’s framing, in the words of one analyst it interviews, is pointed: India risks positioning itself as a market for the taking — “more akin to the backroom,” with echoes of the BPO era. I think that assessment is broadly correct, and I say so as someone who has watched India’s technology ambitions from close quarters across three decades of public service. But correct diagnoses are only the beginning of the obligation. This article is concerned with what comes next — with prescription rather than prognosis, with the policy architecture India actually needs if it is to be a sovereign actor in the AI age rather than its most sophisticated subcontractor.

The window is not unlimited. Decisions made, or avoided, in the next three to five years will determine India’s position in the AI economy for a generation. I propose to examine five domains: what the Government of India must do; what the established IT industry must do to survive and evolve; how India should engage with global technology majors from a position of strategic clarity; what the startup ecosystem needs to become a genuine engine of AI innovation; and what our education and human resource policy must look like if any of the rest is to be more than aspiration.

I. The Honest Reckoning

Before prescription, a brief, unsentimental accounting of where we actually stand.

India’s IT services sector — the $330–340 billion export machine built over twenty-five years on price arbitrage and engineering talent — faces the most serious structural threat in its history. It is not being threatened by a competitor nation. It is being threatened by the very technology it failed to lead. The sector was profitable enough not to invest in AI when investment was still possible at manageable cost. That complacency has produced a vulnerability that cannot be resolved by press releases announcing AI partnerships with Microsoft or Google. The core work — high-volume, process-intensive, moderately-error-tolerant services in HR, finance, compliance, and software — is precisely what large language models are designed to automate. Where one person was needed, forty will do the work that a hundred once performed. The white-collar displacement this implies is significant, and it is already underway.

At the same time, India has allowed itself to be positioned as the world’s data labourer. The annotation industry, the egocentric video collection, the reinforcement learning through human feedback — this is real work, it generates real income for people who would otherwise have little, and in the context of small-town India, particularly for women who cannot or will not migrate to cities, it has genuine social value. I do not dismiss it. But I insist on naming it accurately: it is the lowest rung of the AI value chain, and a nation of 1.4 billion people with the world’s largest youth population and the second-largest AI workforce cannot afford to make it a strategy.

India is also not without assets. Its linguistic diversity — twenty-two scheduled languages, hundreds of dialects, more than a billion active internet users generating data in tongues that American AI companies cannot easily harvest — is a genuine competitive advantage if treated as such. Its domestic market is vast enough that global technology companies need access to it. Its engineering talent, whatever the current quality debate, remains deep by any international comparison. Its democratic legal system, for all its delays, provides an institutional foundation for AI governance that autocratic competitors cannot claim. The question is whether India will convert these assets into sovereignty, or squander them in the excitement of being courted.

II. What the Government of India Must Do

Choose the battleground. The first failure of India’s current AI posture is strategic diffusion. We are simultaneously claiming to build foundation models, to be the world’s AI use-case capital, to lead in AI talent, to emerge as an AI manufacturing hub, and to champion Global South interests in AI governance. These are not mutually reinforcing goals at India’s current resource level. They are competing demands on finite attention and finite money. Singapore chose financial services and logistics. Israel chose cybersecurity and agricultural technology. South Korea chose semiconductor manufacturing and consumer electronics. India has chosen everything, which is to say it has chosen nothing.

My own view is that India’s defensible moat lies in three areas. First, AI applications for its own uniquely complex conditions — multilingual, multi-religious, climatically diverse, institutionally federal. No American or Chinese company will build these as well as India can, because they do not understand the problem space. Second, AI-enabled public services at population scale — the combination of Aadhaar, UPI, ABDM, and DigiLocker gives India a digital public infrastructure that most nations cannot replicate, and which creates a laboratory for AI-enabled governance that the world will want to study and purchase. Third, AI talent — not as a commodity to be exported, but as a strategic resource to be cultivated and retained.

Build the sovereign compute stack. Semiconductors and sovereign compute are to the AI age what steel was to the industrial age. India’s dependence on foreign cloud providers for AI training is not merely a commercial inconvenience — it is a strategic exposure. A country that cannot train its own AI models on its own infrastructure is not an AI sovereign. It is a licensee. The IndiaAI Mission’s ₹10,372 crore allocation is a start, but it must be insulated from the bureaucratic tendency to mistake expenditure for achievement. Every rupee must be tracked against compute capacity delivered, models trained, and researchers enabled — not against press events attended. In parallel, India must use the scale of its domestic market to negotiate meaningful technology transfer arrangements with semiconductor manufacturers. The Americans, Europeans, and Taiwanese all want access to India’s market and its engineers. That leverage exists. It should be exercised before it depreciates.

Give data governance real teeth. The Digital Personal Data Protection Act 2023 is framework legislation. What India needs, urgently, is sector-specific regulation that treats Indian data as a national resource rather than a global commons. The footage being recorded in Tamil Nadu factories, the household activity data being collected by robotics companies, the behavioural data being harvested by American platforms — none of this currently generates any sovereign dividend for India. At minimum, companies collecting training data from Indian citizens and Indian environments must be required to maintain data residency, share model-access rights with Indian public institutions on fair terms, and compensate data generators beyond the poverty wages currently offered. This is not protectionism. It is the elementary assertion of property rights that every functioning economy upholds.

Create a serious labour transition architecture. AI-driven displacement of white-collar workers in India is not a future risk. It is a present reality. The government’s response so far has been to note, correctly, that AI also creates jobs, and to leave it there. That is not a policy. What is needed is an AI and Automation Labour Transition Fund — financed through a levy on companies above a threshold size that deploy AI systems to replace domestic workforce — that funds intensive, sector-specific retraining linked to actual employment outcomes. The PMKVY framework, as currently structured, cannot do this work. It is calibrated for skill certification, not for the kind of deep professional reorientation that a software engineer whose job has been automated genuinely requires.

Regulate intelligently, not reactively. India’s regulatory instinct toward transformative technology has historically oscillated between indifference and over-reach, often in rapid sequence. The AI regulatory framework must be built around sectoral sandboxes — defined domains where AI applications can be tested and evaluated under light-touch oversight — with hard statutory rules reserved for genuinely high-risk domains: AI in criminal justice, AI in credit decisions, AI in medical diagnosis, AI in political advertising. The European Union’s AI Act, for all its imperfections, provides a risk-tiering architecture that India would do well to adapt rather than ignore or replicate wholesale.

III. The Established IT Industry: Adapt or Become Irrelevant

Let me be direct with the leadership of India’s IT services majors: the business model that made you is not the business model that will sustain you. Acknowledging this is not pessimism. Refusing to acknowledge it is.

The arbitrage proposition — Indian engineers at a fraction of Western cost doing work of acceptable quality — is being eroded from below by AI automation and from above by client capability building. Global corporations are building their own AI-enabled capability centres. The work they once outsourced to Bengaluru, they are increasingly doing themselves, with smaller teams and AI tools. The IT services industry’s share of that work will shrink. The only question is how fast, and what replaces it.

From labour arbitrage to intelligence arbitrage. Indian IT companies have accumulated something of extraordinary latent value over twenty-five years: deep operational knowledge of their clients’ industries, and in many cases, contractual access to vast repositories of enterprise data. A company that has run a global bank’s risk systems for fifteen years understands that bank’s risk logic better than any consultant. A company that has managed a retailer’s supply chain across twelve countries understands that supply chain’s failure modes at a level of granularity that no LLM trained on public data can match. This contextual intelligence — combined with proprietary fine-tuned AI models — is the new arbitrage. The companies that recognise this and invest accordingly will survive. Those that continue to optimise headcount and billing rates on legacy contracts will not.

Vertical depth over horizontal breadth. The era of the generalist IT services company is ending. Companies that attempt to be competitive across every vertical and every service line will be outcompeted in each by specialists and by AI systems themselves. Every major Indian IT firm should be asking itself: in which two or three industry verticals do we understand the problem space deeply enough to build AI solutions that clients cannot replicate with a general-purpose model? The answer to that question should drive portfolio decisions, acquisition strategy, and talent investment. Companies that cannot answer it clearly are already behind.

Invest in genuine AI research, not AI marketing. The announcements of AI partnerships, AI centres of excellence, and AI transformation programmes from India’s IT majors have far outpaced any visible investment in original AI research capability. Building a wrapper around a foreign LLM API and calling it an AI product is not a sustainable competitive position — it is a margin-compression exercise with a good communications strategy. India’s IT industry needs to fund actual research: model development, domain-specific fine-tuning, AI safety and evaluation, and human-AI workflow design.

A pre-competitive consortium for foundational AI infrastructure. No single Indian IT company has the capital to build frontier AI capabilities alone, nor should it try. The industry should form a pre-competitive consortium — analogous to what SEMATECH did for the American semiconductor industry in the 1980s — to jointly fund foundational AI research, shared compute infrastructure, common benchmarks, and interoperability standards. NASSCOM can convene it, but the initiative and the funding must come from the companies themselves.

IV. Engaging Global Technology Majors: Partnership Without Subordination

Google, Microsoft, Amazon, Meta, and OpenAI are all investing in India — data centres, research facilities, developer programmes, talent pipelines. India’s government and industry welcome this investment, often effusively. I do not question the investment’s value. I question whether India is negotiating the terms of engagement from a position of strategic clarity or from the giddiness of being courted.

India has more leverage in these relationships than it typically exercises. A domestic market of 1.4 billion people, with rapidly growing internet penetration and a young demographic that will be the world’s largest consumer of digital services for the next three decades, is not a dispensable relationship for any global technology company. The Indian government should negotiate accordingly.

Technology transfer as a condition, not a request. Every significant infrastructure investment by a global technology company in India — data centre, cloud region, AI research facility — should carry binding technology transfer obligations as a condition of approval. This is standard practice in aerospace and defence. A data centre that employs security guards and facility managers while all intellectual property flows to Seattle is not a partnership. It is a real estate transaction with geopolitical branding.

Intellectual property on Indian-origin data. Global AI companies that train their models on Indian-language text, Indian behavioural data, and Indian-environment video footage must be required to share model-access rights with Indian public institutions on fair and transparent terms. The large language models now generating billions in revenue for Western companies were built substantially on knowledge generated in the Global South. India’s negotiating position on this question is stronger than it appears, if the government chooses to advance it seriously.

Joint ventures in sensitive sectors. AI applications that process sensitive Indian data — in healthcare, agriculture, financial services, and public administration — should be required to operate through joint ventures with Indian entities holding meaningful stakes in both the data assets and the model IP. This is not the blanket forced joint venture regime that China applies to all sectors; it is a targeted requirement applied where data sovereignty and national security genuinely intersect.

Champion the Global South in AI governance. India has a platform at the G20, at the UN, and in multilateral technology forums that it has not fully used to advance a coherent position on global AI governance. The issues at stake — equitable access to foundational AI models, fair compensation frameworks for training data, AI safety standards that do not entrench the incumbency of frontier labs — are precisely the issues where India’s voice would carry weight and where its interests align with those of the majority of the world’s nations. This is soft power of the most substantive kind.

V. Startups and the Entrepreneurial Ecosystem: Building What India Actually Needs

India’s AI startup ecosystem has genuine energy and genuine gaps. The energy is visible in the quality of founders, the ambition of the ideas, and the depth of domain knowledge that Indian entrepreneurs bring to problems that American counterparts cannot even properly frame. The gaps are in capital, in risk appetite, in regulatory clarity, and in access to the kind of first-customer relationships that allow a startup to build a proof of concept into a scalable product.

Patient capital through public co-investment. The global AI investment wave has largely bypassed Indian startups. The government must intervene with a Fund of Funds — structured through SIDBI or a dedicated AI investment authority — that takes first-loss positions alongside private venture capital, de-risking the asset class sufficiently to draw domestic institutional capital currently sitting in government securities. The Israeli Innovation Authority model, with its patient milestone-based co-investment, is more appropriate to India’s conditions than the Silicon Valley venture model.

Open government data as startup fuel. The single most powerful thing the government can do for the AI startup ecosystem costs very little money. It is to open its data. India’s public sector holds datasets of extraordinary potential — land records, agricultural yield data across decades and districts, health registries, judicial records, weather and hydrology data, transport movement data — that could power AI applications of enormous social and commercial value. A National Open Data Architecture with standardised APIs and privacy-compliant anonymisation would create a foundational resource that no American startup can access because the data does not exist outside India.

Government procurement as a market-creation tool. The government is India’s largest buyer of technology products and services. It has not used that purchasing power strategically to build the domestic AI industry. A dedicated category on the Government e-Marketplace for certified Indian AI products — with transparent evaluation criteria and reasonable price preference — would give Indian AI startups the first-customer relationships they need. What American defence procurement did for Silicon Valley, Indian government procurement could do for the Indian AI startup ecosystem.

Domain-specific incubation, not generic tech parks. Generic AI incubators produce generic AI companies that replicate what American competitors are already building. India needs domain-specific incubation: an AgriAI hub in Punjab, Haryana, or Andhra Pradesh; a HealthAI hub linked to AIIMS; a JusticeAI hub working with the e-Courts project. The problems India needs AI to solve are specific to India. The incubation infrastructure should reflect that specificity.

Make failure survivable. India’s insolvency framework has improved since the IBC of 2016. But for AI entrepreneurs who may fail twice before succeeding once, the legal and financial consequences of failure remain disproportionately heavy. A specific fast-track resolution process for technology startups below a defined threshold — with personal liability carve-outs for good-faith founders and timelines measured in months rather than years — is within the government’s power to create.

VI. Education and Human Resources: The Foundation of Everything

Every prescription I have advanced in this article rests, ultimately, on whether India produces enough people who understand AI deeply, deploy it wisely, and can govern its consequences honestly. The education system we have was designed for an economy that no longer exists. The human resource policies of our industries were designed for a workforce that AI is now recomposing. Neither can be left unchanged.

AI literacy as a civic and curricular right. From the middle school years onward, every Indian student must understand what AI systems are, how they make decisions, what they can and cannot reliably do, and what their social implications are. This is not vocational training in coding, though coding matters. It is the kind of foundational literacy that allows a citizen to evaluate an AI-generated medical recommendation, a voter to recognise a synthetic political advertisement, and a worker to understand what an employer’s AI productivity monitoring system is measuring. A democracy that cannot govern AI is not governing itself. That governance begins in the classroom.

Redesign the engineering curriculum from first principles. India’s more than six thousand technical institutions are producing graduates calibrated for a services economy of the 1990s. The AICTE must mandate AI, machine learning, data science, and AI ethics as core requirements across all engineering programmes — not electives that motivated students can add, but foundational subjects that every graduate carries into professional life. An agricultural engineer who understands AI-driven crop advisory and precision irrigation is immeasurably more valuable to India than one who can implement a binary search tree.

The liberal arts deficit must be addressed. The engineers who will design India’s AI systems — systems that will govern credit allocation, healthcare triage, criminal risk assessment, and employment decisions — need to understand power, discrimination, institutional failure, and ethical reasoning as well as they understand gradient descent. Ambedkar’s analysis of structural caste discrimination is directly relevant to the design of AI systems that will operate in Indian social contexts. The IITs and NITs must integrate these perspectives into their curricula as substantive intellectual requirements, not ornamental additions.

A mid-career upskilling mission at scale. The four to five million people currently employed in India’s IT services sector cannot wait for a redesigned school curriculum. They need intensive, outcome-linked mid-career upskilling programmes that are financially accessible and genuinely transformative. Singapore’s SkillsFuture programme — which provides every citizen above twenty-five with a training credit redeemable at accredited institutions — offers a model worth adapting. The key design principle is that the credit follows the worker, not the employer, so that individuals can make genuine choices about their own professional futures.

AI-literate faculty as an urgent bottleneck. We cannot produce AI-literate graduates without AI-literate faculty. The shortage of qualified AI instructors in India’s universities and colleges — particularly outside the IITs and IIMs — is severe and worsening, because the private sector bids away the talent that academic institutions need. A National AI Faculty Development Programme — combining international fellowships, structured industry sabbaticals, and competitive salary supplements — is an investment in the multiplier, not just in the immediate output.

Women in AI: from the bottom of the pipeline to its architects. The FT documentary captures, with uncomfortable clarity, where women currently sit in India’s AI economy: at the base of the data collection pyramid, their labour indispensable and their compensation negligible. India’s aspiration must be that the women of small-town Tamil Nadu and rural Punjab are not merely training AI systems but designing, deploying, and governing them. Targeted doctoral fellowships, women-led startup incubators with dedicated capital, and industry-level gender parity commitments with genuine accountability mechanisms are minimum requirements for a country that cannot afford to leave half its talent on the table.

Conclusion: The Obligation of Scale

India has one thing that almost nobody else has at comparable scale: human talent across a vast and diverse geography, in sufficient volume to staff every rung of the AI value chain simultaneously if deployed with strategic intelligence. The FT documentary names this plainly: talent at very large scale is India’s unique asset. It then asks the question that should haunt India’s policymakers: if you do not have enough good jobs for these people, what is the scale worth? Nothing.

The prescriptions I have set out are not radical innovations. Most have working precedents in countries that have navigated technological transitions with more strategic discipline than India has shown so far. What they require is not cleverness. It is will — the political will to make choices instead of announcements; the institutional will to coordinate across the ministries, regulators, and industry bodies that currently operate as isolated fiefdoms; and the intellectual honesty to acknowledge that being the world’s most sophisticated data annotator is not the same as being the world’s AI power.

India will be a developed country by 2047, or it will not. If it is, AI will have been central to that journey. If it is not, the failure to master AI on sovereign terms will be a significant part of the explanation. The woman in Tamil Nadu wearing a camera on her forehead is building something. The question that her government, her industry, and her educational institutions owe her an answer to is: does she own any part of what she is building? Or does she merely make it possible for others to own it?

That is not a technical question. It is a political one. And the answer depends entirely on choices that have not yet been made.