India's Automation Trap

Shreshth Rajan. Originally written Summer 2025, updated April 2026.

Vietnam has one-fourteenth of India's population. In 2024 it exported $405 billion in goods, nearly matching India's $428 billion. Its factory workers earn roughly double what India's do, and yet Vietnam keeps landing the manufacturing supply chains India wants.

The usual explanations (infrastructure, regulation, ease of doing business) capture part of it but miss the deeper point. India is caught in a trap where cheap labor, its supposed advantage, is what keeps it poor. When you can hire a worker for $150 a month, there is no reason to buy a machine, and without buyers, nobody builds the machines in the first place. Every firm acts rationally, and the country ends up stuck at $8 of output per working hour, 133rd in the world.

I discussed an earlier version of these ideas with Gita Gopinath at the IMF last summer.

The Trap

Think of it as a prisoner's dilemma played by a million firms. Any single company, faced with abundant labor at $150 a month, will hire people instead of buying machines. That is the rational move. But when every firm makes that move, the economy stays low-productivity, no domestic automation industry emerges, and India remains poor.

The structure of the economy reinforces this. India's development path diverged early toward services rather than scale manufacturing, and 58% of workers today are self-employed, with only 22% holding regular salaried jobs. Labor regulations push firms to stay small: a firm with 99 employees has every incentive to stay at 99, because crossing 100 triggers compliance requirements that make the math worse than just not growing. The same rules that keep firms from scaling also kill any chance of technology adoption.

Robot adoption makes this visible. India installed 8,510 industrial robots in 2023, a 59% jump that ranked it 7th globally. But density is what matters: 5 robots per 10,000 manufacturing workers, compared to 392 in China and over 1,000 in South Korea. The same pattern shows up across China and Europe: when wages rise, robots follow. When labor is cheap, nobody bothers.

The productivity gap within India looks similar. Hsieh and Klenow found that the most productive Indian firms produce roughly five times the output per worker of the least productive, a spread far wider than in the US or China. Small firms cannot justify machines when they can hire people for almost nothing, so the gap compounds.

AI Makes This Existential

Everything above describes a slow problem. AI has made it a fast one. The IMF puts India's high-AI-exposure employment at roughly 26%, concentrated in the exact services India exports. Tech is a $254 billion industry, and its revenue comes from billing clients for developer-hours. What happens when the hours collapse?

Two years ago, AI coding meant Copilot autocompleting a line. Today 95% of developers use AI tools weekly. More than half do the majority of their work with AI. Claude Code went from zero to the most-used coding tool in eight months. Devin writes, tests, and ships pull requests autonomously. AI has moved from assisting engineers to doing their work.

The question is not whether AI replaces Indian engineers. It is whether five engineers with these tools do the work of fifty. If the answer is even half-yes, the labor arbitrage model that Indian IT runs on breaks. The evidence is already here: TCS cut 12,000 jobs by March 2026, and net IT hiring across the top firms dropped to 1-1.5% of the existing workforce. A business model built on selling developer-hours does not survive a world where the hours shrink.

China, meanwhile, is building the tools. DeepSeek's R1 hit frontier performance at under $6 million in training costs. By early 2026, V4 was coming: a trillion parameters, optimized for coding, running on Huawei chips. Chinese firms hold six of the top ten open-weight positions. India produces 1.5 million engineering graduates a year. It has nothing on this leaderboard.

Vietnam

Since 2005, Vietnam's exports have grown 983% against India's 339%, while Vietnam's factory workers earn $294-340 a month compared to India's $150-180. Higher wages, more factories. Not despite the wages. Because of them.

Vietnam simplified regulation and invested heavily in infrastructure, and crucially never built size-based regulatory thresholds into its labor code. Electronics are now over a third of its exports, up from 14% in 2010, and its manufacturing share of GDP is nearly double India's.

Vietnam got richer by making it easy to build factories. India stayed poor by making it easy to hire cheap labor.

Tamil Nadu

One Indian state broke out. Tamil Nadu now ships 41% of India's electronics exports: $14.65 billion in FY24-25. Just a decade ago, this was negligible.

What they did was counterintuitive. They raised effective labor costs by enforcing minimum wages, then offered capital subsidies of up to 30% for manufacturing in less-developed districts, plus training subsidies and matching central incentives. Foxconn, Pegatron, and Samsung came in. Not because labor was cheapest. Because the state made machines cheaper than people. The Chennai-Hosur corridor is now India's most automation-intensive manufacturing region.

Change the price ratio between labor and capital and firms choose machines. Tamil Nadu proved it.

The Counterargument

There is a real objection here. Dani Rodrik has shown that developing countries are hitting premature deindustrialization: manufacturing employment peaks earlier and at lower incomes than it did for Britain or Germany. India's manufacturing employment share peaked around 13%. If the manufacturing window has closed, trying to follow East Asia is pointless.

Rodrik himself evolved on this in 2024, arguing that services-led development can work, but only if those services are high-productivity. Which is exactly what AI threatens. If India's high-value services get compressed by AI tools and manufacturing was never built up, both paths close. The trap is not about manufacturing alone. It's the risk of having neither path left.

What Exists and Why It Is Not Enough

India is not doing nothing. The PLI scheme committed $23 billion across 14 sectors, attracted $20 billion in realized investment, and grew electronics production from $26 billion to $63 billion in four years.

But as Raghuram Rajan has noted, much of this is assembly, not manufacturing. Net phone and component trade worsened from -$12.9 billion in FY17 to -$21.9 billion in FY23. PLI rewards production volume. It leaves labor thresholds intact, does nothing about the regulations keeping firms small, and leaves the coordination failure exactly where it found it. It can bring Foxconn to Chennai. It cannot make a hundred thousand small firms across India choose machines over another hire.

What Would Actually Work

Every previous attempt at Indian labor reform failed for the same reason: the people who would lose protection fought hard today, while the people who would benefit were diffuse and years away. AI changes this equation. A quarter-trillion-dollar industry is now watching its margins compress in real time, which creates a concentrated, powerful constituency for reform that did not exist before.

Tamil Nadu showed the playbook: crisis plus coalition. Electronics assemblers wanted simpler rules, the state wanted investment, together they overwhelmed small-firm resistance. The national version starts the same way. New firms get a simplified single labor code with no size thresholds. Existing firms keep current rules. Nobody loses today. As new firms outcompete old ones, the business community demands extension. China's dual-track reforms worked this way. You do not threaten incumbents. You let the new system win.

Beyond new firms, you scale what already works at the state level. Fund Tamil Nadu, Karnataka, and Telangana to expand their models, and if Maharashtra will not reform, let firms reincorporate somewhere that will. Revise the GST formula to reward formal employment so that states compete on governance rather than subsidies. Give the IT industry simplified zones in exchange for backing labor reform, turning 1,500+ Global Capability Centers into the constituency that pushes reform into national policy.

Transition costs have to be handled honestly, and in the right order. Portable benefits through JAM, earned-income rebates for low-wage formal workers, retraining targeted by PLFS data. Denmark's flexicurity works because the net came before the freedom. Reverse that sequence and the politics collapse.

I think the window is closer to three years than five. Most people I say this to think I'm exaggerating. But even ordinary engineering teams I know have stopped hiring for seats and started buying AI tools instead. Fewer people, more code. When that logic reaches the middle of the talent pool, not just the frontier, the game is already over. India's policymakers are confusing comfort for safety. By the time the pain is obvious, the window will be closed.

If this is not done by 2030, it probably will not happen. India needs eight million new jobs a year. The current path delivers informal work at $8/hour productivity. The automation path delivers fewer, better jobs at multiples of that, with growth absorbing the displaced. The transition is painful. The alternative is permanent: stuck near $3,500 per capita while China pushes past $25,000 and Vietnam targets $10,000 by 2035.

Every firm in India is making the rational choice. Hire the worker. Skip the machine. It's the right call for each of them and a disaster for all of them. China broke out of this through state power. Vietnam opened radically to foreign capital. Taiwan spent decades on industrial policy. Each found its own escape from the same trap. India has to find one that works inside a federal democracy, which is harder but not optional. You cannot underbid on wages forever. You can out-invest in tools today.