India's Services Escalator: How AI Reprices the Engine of India's Economy

Two forces, one engine

India runs an unusual external economy. For most large economies the export story is goods, and services are a supporting act. In India the ratio is close to one to one, and services are the surplus leg while goods are the deficit leg. Services exports reached a record $387.55 billion in FY2024-25 and a provisional $421 billion in FY2025-26, against merchandise exports of roughly $442 billion. The net services surplus, about $189 billion in FY2024-25 and a provisional $214 billion the year after, offsets close to two thirds of the goods trade deficit, and together with the world's largest remittance inflow it is what holds India's current-account deficit near 0.6 percent of output. This engine is not a sideline. It is the stabilizer of the whole external account.

That engine is now meeting the most disruptive general-purpose technology since the internet, and the debate about what happens next has hardened into two stories told past each other. The first is disruption: artificial intelligence automates the routine, English-language, rules-governed knowledge work that India built its offshore industry on, and the labor-arbitrage model that created a formal middle class begins to unwind. The second is opportunity: the same technology is India's chance to climb from selling cheap hours to selling AI-augmented, intellectual-property-led work, with a talent base already the deepest in the world outside the United States.

Both stories are true, and read together they describe a single mechanism rather than a contradiction. This study argues that AI does not simply grow or shrink India's services engine. It reprices it, tier by tier. It compresses the value of the lowest-skill work fastest, rewards a climb into higher-value work, and in doing so threatens to pull two of the engine's outputs apart: its dollar value, which can keep rising as machines do more per worker, and its capacity to create mass formal employment, which strains as the bottom of the pyramid automates. The early data already shows this split. Industry revenue kept rising through 2025 and 2026 while the largest firms cut headcount and entry-level hiring collapsed.

The method is stated at the outset. This is a synthesis of dated, attributed public data from the RBI, the Commerce Ministry, NASSCOM, Zinnov, and the international research bodies, not a forecast model. Every figure carries an evidence tier: established where independent sources converge, emerging where a figure is recent or single-source, and contested where credible parties disagree on magnitude. The projections at the end are transparent scenarios built on stated assumptions, not predictions. The direction of travel is well corroborated; the exact magnitude is genuinely disputed, and the study says so wherever it is.

AI does not simply grow or shrink India's services engine. It reprices it, tier by tier, pulling the engine's dollar value apart from its capacity to create jobs.

The engine, measured

The scale is easy to understate. India's share of world services exports rose from 2.0 percent in 2005 to 4.3 percent in 2024, and in telecommunications, computer and information services specifically India holds around a tenth of the global market, second only to the United States. Software and information services are about half of the country's services exports, near $206 billion in FY2025-26. The fastest-growing block beneath the software headline is "other business services", the consulting, engineering, research, legal and analytics work that rose from about 19 percent of services exports in FY2013-14 to roughly a quarter a decade later, and now stands near $124 billion. That second block is the statistical shadow of the capability-centre boom this study returns to below.

The technology industry at the core of this engine is measured separately by NASSCOM, and the two annual reviews disagree slightly on the base, a revision worth naming rather than hiding. The Strategic Review published in early 2025 put the industry at $282.6 billion for FY2024-25; the review a year later restated that base near $297 billion and projected $315 billion for FY2025-26, a growth of 6.1 percent, of which exports are about $246 billion and roughly 83 percent of the total. The sector contributes on the order of 7 percent of India's GDP. Within it, IT services are the largest slice at about $149 billion, but the signal in the segment mix is that engineering research and development, near $63 billion, has now overtaken business process management at about $59 billion. The higher-value work is growing faster than the back office.

The workforce behind these numbers is about 5.8 million people, rising toward 6 million, with women now about 36 percent of it. And it rests on a single durable advantage: cost. An Indian engineer typically costs a quarter to a third of a comparable hire in the United States, a gross wage gap on the order of 60 to 70 percent, and that arbitrage is what the entire offshore model has monetized for three decades. The question this study asks is what happens to an engine built on that arbitrage when the machine starts doing the arbitraged work.

The Services Escalator

To read the impact without collapsing it into a single verdict, the study uses a proprietary lens, the Services Escalator. It ranks service work in four tiers by how exposed it is to automation by current AI, and the organizing claim is that AI acts on each tier differently. It automates the bottom, augments the middle, and rewards a climb to the top. The escalator is descriptive and testable, not a claimed law, and each tier below is measured against dated evidence.

The lens reframes the headline dispute. Asking whether AI is "good or bad" for Indian services is the wrong question, because the answer is different on every rung. The tier where most of the jobs sit is the tier where automation is most complete, and the tier where India can defend and grow value is the tier that employs the fewest people. That asymmetry is the whole argument.

The table below sets out the four tiers, what AI does to each, and where India's exposure sits.

The lower tiers: where AI bites first

The clearest evidence of repricing is at the bottom of the escalator, and it is already in the employment data rather than in a forecast. Entry-level hiring, the mechanism that turned millions of graduates into a formal middle class, has contracted sharply. Annual fresher intake across the industry fell from a peak near 600,000 in FY2021-22 to about 120,000 in FY2024-25, a decline on the order of 80 percent, and entry-level technology openings dropped about 44 percent year over year into 2026. The work that absorbed those graduates, manual testing, tier-1 support, documentation and basic coding, is precisely tier-1 and tier-2 work, and one large advisory estimates AI can already perform 20 to 40 percent of common technology tasks and that entry-level roles have declined 20 to 25 percent as a result.

The revenue model built on those hires is breaking in public. The industry's decades-old equation, more engineers means more projects means more revenue, is being abandoned by the majors themselves. In FY2025-26 the four largest firms shed headcount into flat or falling revenue: the largest, TCS, cut net headcount by more than 23,000 while lifting its operating margin to a four-year high and reporting revenue down 2.4 percent in constant currency. Chief executives now name the force directly. Multiple leaders describe "AI deflation", productivity gains passed to clients as lower prices, with an industry base case near 3.5 percent annual price erosion on traditional services and one chief executive putting it plainly: a deal that once billed $100 million "would be much less today, maybe $80 million", even as it requires more effort. Analysts estimate 12 to 15 percent of sector revenue faces direct AI-driven displacement risk and that headline growth will stay range-bound near 3 percent.

The voice back office, tier 1, faces the most direct substitution. India's roughly 2.8 million business-process workers are about 60 percent in voice roles, and AI can now deflect 40 to 60 percent of routine contact-centre inquiries. The counter-signal is that seat counts have not yet collapsed, because deflection reduces volume per agent faster than it removes total agents, and contact-centre employment was still growing in 2025. But the direction is not in dispute. The tiers that hold the most jobs are the tiers the machine handles best.

The tiers that hold the most jobs are the tiers the machine handles best. That is the whole difficulty in a sentence.

The upper tiers: where India can climb

The escalator has an up direction, and India is genuinely climbing it. The strongest evidence sits in the capability centres. Global firms now run more than 2,100 captive Global Capability Centres in India, employing about 2.36 million people and generating $98.4 billion in FY2026, up from roughly $64.6 billion two years earlier. That figure is striking for a second reason: it already matches the $99 to $105 billion the sector had projected for 2030, reached about four years early, which tells you the older target will be revised sharply upward rather than that the boom is near its ceiling.

These are not the cost centres of the last decade. In FY2026 more than 1,200 of these centres held AI and machine-learning capability and over 250 ran dedicated AI centres of excellence, employing on the order of 250,000 AI professionals. About 96 percent of the centres set up since FY2021 launched with product or portfolio ownership rather than pure delivery, 64 percent of India site leaders now hold global mandates, and more than 6,500 global leadership roles are run from India. This is tier-3 and tier-4 work migrating into the country, engineering research and development growing faster than the industry as a whole, the same capability pool that shows up in the balance of payments as the surge in "other business services".

The talent base supports the climb. India recorded the highest year-over-year AI hiring growth in the world in 2024, about 33 percent, and ranks second globally on AI-skill penetration at roughly two and a half times the world average. The AI talent pool, about 420,000 in 2024, is projected to reach 1.25 million by 2027. The applied and vernacular layer is real and accelerating: a sovereign large-language-model effort reached unicorn status, and a national translation platform now spans 22 official languages, a genuinely differentiated position given India's linguistic scale.

The climb is real but it must be stated with its limits, because this is the least-proven leg. AI revenue is still only 3 to 4 percent of the sector, the pricing shift from time-and-materials to outcome-based delivery is confirmed by management but hard to scale, and the depth of talent is thinner than its breadth: more than 90 percent of early-career professionals use AI tools, but only about a quarter qualify as AI-native engineers able to build and deploy systems. India also depends on imported compute, which analysts name the single biggest long-term hurdle, and the flagship IndiaAI Mission has released only about 4 percent of its five-year budget in two years even as it put tens of thousands of subsidized GPUs online. The capability is installed. Whether it scales fast enough is the open question.

The reshoring question

Underneath the tier-by-tier story sits a sharper, contested claim: that AI removes the escalator's reason to exist rather than merely moving Indian work up it. The argument, advanced by analysts at firms such as HFS Research, is that AI enables a "services-as-software" model that reduces headcount "regardless of location", letting Western firms automate work onshore rather than offshore it, and so dismantling the labor arbitrage that built the industry. If AI voice agents become cheaper than a human agent in every geography, the cost advantage that sent the work to India disappears. Early signals are cited in support: headcount falling at the majors, and at least one American firm closing its India operations less than two years after opening them, a case some called a watershed for AI-driven operations.

The counter-thesis is that this is a shift from cost arbitrage to capability arbitrage, not a collapse. On this reading India stops competing on cheap hours and competes instead on owning the higher-value AI and engineering work, exactly the migration the capability-centre data shows. One think-tank's own displacement math is deliberately modest, on the order of 24,000 to 90,000 net capability-centre job losses under its scenarios, small against a base heading toward millions. The disagreement is genuine and unresolved, and it is presented here as a debate rather than settled: the pessimists say the arbitrage that made a middle class is ending regardless of who climbs; the optimists say India climbs faster than the routine work is automated away.

The comparative picture favors India relative to its nearest competitor. The Philippines, the world's call-centre capital with about 1.3 million business-process workers, is judged more exposed precisely because it is concentrated in the voice work of tier 1, with projections of a net loss on the order of 200,000 jobs this decade. India's diversification into higher-value IT and capability-centre work is a real hedge. Whether that hedge is a durable shield or merely a slower path to the same automation is, again, contested. A second front is political rather than technological: proposed United States legislation to penalize offshoring would pressure the model independent of AI.

The macro stake: growth without jobs

The reason this matters beyond any single firm is that India's services engine carries two national burdens at once, and AI threatens them unequally. The first burden is external stability, and here the engine is strong: the services surplus and remittances convert a large goods deficit into a near-balanced current account, and a sustained shock to IT export earnings would transmit directly to the rupee and the reserves. On the dollar value, the escalator's up direction and the sheer momentum of the capability-centre boom make continued growth plausible.

The second burden is employment, and here the arithmetic is unforgiving. The technology sector is India's premier maker of formal, middle-class jobs, but it is numerically tiny against the labor force. Capability centres employ under half a percent of India's workforce; the whole tech sector is under 6 million against a labor force above 600 million, roughly 90 percent of it informal. Meanwhile India must absorb 7 to 8 million young people into the workforce every year, the working-age share peaks around 2030 and then declines, and the unemployed are disproportionately young and educated. This is the collision at the center of the study. The formal escalator into the middle class narrows exactly as the demographic window that needed it most begins to close, and the manufacturing sector that absorbed surplus labor in East Asia has stayed stuck near 15 to 17 percent of Indian output and did not step up.

This is why the decoupling matters more than any single automation figure. It is entirely possible, on the current evidence, for India's services exports to keep setting records in dollars while the sector's capacity to hire the next cohort of graduates weakens. Revenue rose in 2025 and 2026; fresher hiring did not. Growth without jobs is not a paradox here. It is the most likely reading of the data, and it is a macro problem, not merely a labor-market one, because the same engine underwrites both the external account and the social contract of upward mobility.

The formal escalator into the middle class narrows exactly as the demographic window that needed it most begins to close.

Three scenarios to 2030

Projection here is deliberately built as scenarios rather than a single forecast, because the decisive variables, the pace of automation, the speed of the climb, and policy, are choices as much as trends. Each scenario below states its assumption. The employment range is anchored to the published band from NITI Aayog, NASSCOM and BCG, whose own 2025 roadmap put net tech-sector jobs by 2031 anywhere from a loss of about 1.5 million to a gain of about 2.5 million, with up to 4 million new roles possible under sustained action, and explicitly called the outcome conditional. The export-value figures are illustrative extrapolations from the stated growth rates, not measured, and are marked contested. They are included to show the shape of each path, not to predict a number.

The three paths are not equally likely, and the study's reading is that the middle one, the split, best fits what the data already shows.

The limits of this reading

Several cautions bound the argument. The macro exposure literature genuinely disagrees on India: the IMF and Goldman Sachs frameworks rank India among the less-exposed major economies at the national level because so much of its economy is manual and informal, even as the IT-BPM sector is among the most exposed slices precisely because it is high-value clerical work delivered in English. Both readings are true because they use different denominators, and this study is about the sector, not the whole labor force.

The strongest near-term figures are also the least settled. The productivity uplift from coding assistants is well measured in the laboratory, around 55 percent faster on a controlled task, but enterprise results are more mixed once review and debugging are counted. The sharpest fresher-decline percentages originate partly in trade and career press rather than primary datasets, and are treated as directional. Several NASSCOM and Zinnov headline numbers were reachable only through secondary coverage during this research and are tiered accordingly. And the consensus of the international bodies, the ILO in particular, is that transformation and augmentation, not wholesale replacement, is the most likely path for most roles, a corrective against the loudest disruption claims.

What survives all of that is narrow and well-founded. India's services engine is large, strategically load-bearing, and climbing into higher-value AI work with real installed capability. The same technology that lets it climb is repricing the lower tiers where its mass employment sits, and the early data already shows revenue and headcount pulling apart. The dollar value of the engine and its capacity to make jobs are no longer the same measurement, and for a country with India's demographic clock, the gap between them is the thing to watch.