Raghuram Rajan discusses how India's economy grew through services exports, why that model may be more resilient to AI than critics assume, and what policymakers need to get right on human capital, universities, and digital access to stay ahead.
You can listen to this podcast on Spotify, Apple Podcasts, or wherever else you get your podcasts. You can also watch this conversation on YouTube, and we have started a Substack for this podcast.
Every morning across India, software debuggers sit down at their desks, log on, and spend their day hunting for bugs that could derail a multi‑million‑dollar project. This is one example of the work that has helped to power India’s services export boom, which has been an engine behind global apps, banks, retailers, and supply chains.
But there’s a catch. As Raghuram Rajan and Rohit Lamba recently put it: Coding sweatshops don’t stand a chance against AI.
So what happens to a growth model built on services, and trading those services with the world, in the age of AI? This is a question all countries pursuing a service-led growth model are being forced to confront.
In this episode of Ideas in Development, Raghuram Rajan joined us to discuss why he feels the standard doom narrative around AI and service-led growth is too simple, why India's model may be more resilient than critics assume, and what policymakers need to do to make sure it stays that way.
Why services are now an engine of economic growth
The traditional development story runs from agriculture to manufacturing to services. Manufacturing, in this telling, acts as an escalator: it offers scale, exports, and productivity gains that pull economies upward. Services, by contrast, were long seen as a residual – haircuts and plumbing, consumed locally and hard to trade.
But that framing of services is increasingly outdated. Raghu argues that the line between manufacturing and services has become deeply blurred, as modern goods are saturated with ideas, software, and design. Consider the iPhone: someone assembles it, but its value lies overwhelmingly in the design, the operating system, the app ecosystem.
"It's hard to know when the services stop and the good begins."
This matters because once services are embedded in traded goods, they inherit many of the same growth properties that made manufacturing attractive: scale, tradability, and exposure to global demand. And other services, e.g. consulting and legal analysis, are now traded directly, without needing a physical product to carry them.
How India built a growth model on services exports
India's path diverged from the East Asian playbook. Countries like Taiwan, South Korea, and China climbed the development ladder through manufacturing exports, starting with textiles and moving up to electronics and machinery.
India "may have missed the manufacturing bus in the 80s and the early 90s."
Raghu explains how India was held back by gaps in education, logistics, and cumbersome bureaucratic regulation.
What filled the gap was services. Around the Y2K scare, Indian software firms, which had grown largely without government support, found increasing demand for their skills. Higher-tech services became a significant source of export revenue, compensating for the manufacturing shortfall. At the same time, India's own urbanising population generated domestic demand for moderately skilled services: construction, carpentry, plumbing, and repair.
This combination, high-tech services for export, moderate-skilled services for the domestic market, has helped to delivered around 6% growth for three decades. Creditable, even if not the spectacular highs of East Asia.
From back offices to global capability centres
More recently, the nature of what multinationals do in India has shifted. Initially, firms moved classic back-office functions like trade matching, medical transcription, and data entry. But improvements in communications technology have pushed the frontier upward. Indian consultants now deliver presentations, rather than simply preparing slides. Chip designers at firms like Qualcomm work on cutting-edge products from Indian offices.
"You can hire the best kid from India's top management school… and you will pay one fourth or one fifth what you would pay an MBA in Europe or the US."
That labour cost differential, combined with strong English-language skills and improving digital infrastructure, has turned what companies once called back offices into what they now call Global Capability Centres, hubs doing semi-front-office or even front-office work.
Does AI threaten India's service-led growth model?
If AI can handle code debugging, legal discovery, or data processing, what happens to the workers and export revenues that depend on those tasks?
Raghu's answer was nuanced. He acknowledges that some tasks will be displaced. Legal discovery done by reading through hundreds of thousands of pages of case law is going to become less important. AI hallucination remains a problem, but that’s a problem AI will fix, and hallucinations do not seem like a permanent state of affairs.
Yet he pushes back hard against the idea that India is uniquely exposed.
"AI is coming for service jobs across the world," Raghu argues "which means all of us have to adapt."
The critical point is that India's cost advantage does not disappear when both sides adopt AI.
"That kid in India is $50,000 while the kid in the US is $250,000. And so when both are equipped with AI, so long as both have access to AI, that differential doesn't change."
And he cautions against zero-sum thinking, i.e. the assumption that if an Indian worker gains, a Western worker must lose.
"As the kid in India becomes more prosperous they will want the Gucci's, the Teslas… the iPhones, and as a result both sides grow richer."
Where AI augments rather than replaces: Plumbers, nurses, and humans in the loop
Not all services face the same level of AI risk. Raghu draws a useful distinction between routine cognitive tasks, which AI can already handle or soon will, and services that require physical adaptation, human judgement, or interpersonal trust.
Plumbing is one example. Nursing is another, where AI could help a nurse practitioner diagnose symptoms and decide when to refer upward, dramatically extending healthcare access in countries with doctor shortages. Agricultural extension workers could consult AI on soil conditions and crop choices, then translate the advice for farmers in a way that an algorithm alone cannot.
"AI plus human in many areas is probably better than AI alone."
He points to the relative disappointment of MOOCs (massive open online courses) where completion rates often languish below 10% without human support. Technology works best when accompanied by a guide who can hold someone's hand through the process.
The policy agenda: Human capital, digital access, and university reform
Raghu identifies two priority areas for policymakers in India and other countries pursuing service-led growth.
The first is human capital. A service-led growth path is far more human capital intensive than a manufacturing one. That means getting nutrition right (India still has high malnutrition rates), improving early schooling, and building genuinely world-class universities. India is creating colleges at a rate of three per day, but the real problem is quality, because finding good teachers is hard. The country needs a tiered system: teaching colleges for broad access, and research universities that train the next generation of faculty and drive innovation in partnership with industry.
The second is bridging the digital divide. Access to AI tools needs to be broad, not confined to elites. Raghu highlights Tamil Nadu's recent programme providing laptops with AI access (via Perplexity) to all students entering college, though he notes that without handholding and integration into coursework, such programmes are unlikely to succeed.
He also makes the case for engaging the Indian diaspora in building research capacity, pointing to China's thousand turtles programme as a model.
"Getting the right people is the first part of any kind of research programme."
Luckily, India has a deep bench of talent working abroad that could seed collaborations at home.
One model to rule them all?
Several important uncertainties remain. On the platform side, it is unclear whether the major AI providers will be able to recover their enormous fixed costs, or whether intense competition will drive prices down to marginal cost, benefiting users but undermining the platforms' business models. Raghu sees differentiation as the key variable for profitability, and hopes that open architectures and domestic competitors will prevent any single platform from dominating.
And he is highly skeptical of ‘singularity’ narratives where a leading AI lab takes off and leaves everyone else in the dust.
“There are no secrets in industry. Just hire a few people from there and get them to implement what they did in your firm and you can keep pace.”
On the macro-financial side, dependence on a small number of AI systems could create systemic risk, but Raghu thinks competition between US platforms, Chinese alternatives, and homegrown systems trained on local data makes a single point of failure unlikely.
India's prospects in the 21st century
"A service-led growth model can be useful, and it's not just high-tech services… it's also a variety of urban services which creates the jobs that people need."
The question is whether India, and other developing countries watching closely, will make the investments now that determine whether AI becomes a complement to their growth models or a substitute for them.
Raghu is cautiously optimistic, with a political caveat.
"India has a wonderful future if it can get its politics right."
At 6% growth, India is doing well. At 8% or even 10%, enabled by investment in people, logistics, and digital infrastructure, it could offer a replicable model for service-led development in a world that is becoming more protectionist towards goods trade.