India's 2,117 offshore centres now run $98.4bn of the world's corporate work, and the maturity data says they own the mandate, not just the cost line. But the wage gap that pays for all of it is the one input artificial intelligence compresses fastest. This report sizes the machine, prices the arbitrage, and names where the exposure is actually tradable.
Gravitywell Research publishes independent research across five pillars: Credit Research, Capital Markets, Industry & Sector Research, Economy & Policy, and Risk & Analytics. Our work is analysis for professional investors across public and private markets, institutions and sophisticated individuals, and for policymakers. It is not regulated ratings or investment advice. We source to primary filings and official statistics, disclose our coverage and confidence, and mark every figure to its point in time.
Currency: USD unless stated; ₹ = INR. FX: ₹96.5/$ (24 Jul 2026) for all conversions unless a source states its own rate; the 12-month average to Jul 2026 is ₹91.8/$. Period basis: FY = Apr–Mar, so FY26 ended 31 Mar 2026; CY where a source reports calendar years. Figures are nominal. Data as of 27 Jul 2026. Estimates are ranges, not points. Headcount is people employed, not full-time equivalents.
Confidence tiers on figures: ● filed/official ◐ modeled/derived ○ Gravitywell Research estimate.
How to read · The executive summary (p03–05) and the investor quick read (p06) stand alone; the one-page tearsheet is the committee version. Every figure carries a superscript to the numbered register (p37–38); the series behind every exhibit is in the data appendix (p41–42) and in the CSV data pack that ships with this report.
India's 2,117 Global Capability Centres employ 2.36 million people and bill $98.4bn a year, and revenue is compounding twice as fast as the centre count.1 That gap is the whole story. The industry is not growing by opening more sites; it is growing because the work inside the existing ones is worth more, with 96% of centres opened since FY21 launching with ownership of a product rather than a queue of tickets. The wage gap that pays for all of it is not closing on any horizon an allocator underwrites: Indian salaries rose 9.1% last year, and the rupee fell enough that the dollar cost of an Indian engineer still went down 2.7%. But the same automation that lets a centre do more per person is what would eventually let it do the same with fewer, and that single variable swings 2030 revenue by $108bn. It is also the one number nobody publishes.
Three consequences follow, and they are what the rest of this report proves. First, the centre itself cannot be owned: a captive is reimbursed its costs plus a markup fixed at 15.5% by statute from Apr 2026, so its accounts will never show the value it creates however good its mandate becomes.3 Second, that value does become a cash flow one layer out, in the campus it occupies, where GCCs took 45% of all Indian office leasing in the first half of 2026 and the listed landlords pay 5.7–6.2% today against guided step-ups.19,5,21 Third, the losers are visible and already priced: the Indian systems integrators trade at 14 to 19 times earnings against a 25.5-times history, having shed headcount in the year their revenue per employee rose.34,35,36
High conviction, Constructive. All three of our 2030 cases clear the $105bn forecast still in general circulation, the regulatory changes of the last two years are net favourable, and two of the three external policy measures now in force push work toward India rather than away. We express it by owning the landlord at a stated entry level, not by owning the centre, and we hold the vendor only as an option on the upside case.
No published series measures the displacement rate, so our 8/20/40% band is a judgement and is marked as an estimate wherever it appears. Named-centre headcounts are press compilations rather than audited disclosures. Three brokerages report first-half 2026 office leasing at 35.7, 43.0 and 48.0 million sq ft on different definitions, so the GCC percentage share carries that spread even though the absolute volume does not.
The asymmetry decides the instrument. Because one unobservable judgement drives both levers behind the forecast range, the correct expression is a cash flow that survives the downside case rather than one that needs the upside, and that rules out most of the obvious candidates.
| Reader | Action | Price or threshold |
|---|---|---|
| Allocator | Buy Indian office trusts as duration on GCC expansion | Forward yield >6.5% or discount to NAV >15%. Brookfield India at ₹343 is 6.2% and −11.3%; Embassy at ₹441 is 6.1–6.5% guided and −10.3%. |
| Hold the systems integrators only as an option on the bull case | 14.0–19.4× against a 25.5× history, with 2.8–4.4% dividend yields paying you to wait | |
| Private capital | Own the platform that builds centres, not the saving inside one portfolio company | Mid-market cohort projected 610+ → 950+ by FY30 on ANSR's base, about 56% |
| Re-mark unlisted enabler holdings against de-rated listed comparables | Listed comparables already 39% below their own history | |
| Corporate operator | Measure and raise the share of the site in owned-product work; give the site leader a global function; elect the safe harbour | National benchmarks: 44% mandate-owning, 64% dual leadership, 15.5% markup with a ₹2,000 crore ceiling |
| Policymaker | Spend on graduate quality and on power, water and commute in the two cities that hold two-thirds of demand; stop bidding cash | 30% of 1.5m graduates fail employability; 69% of GCC leasing is in two cities |
Every argument in this report reduces to whether centres keep adding people while their output per person rises. That is what separates them from the vendors, whose headcount is already falling. The number prints in the annual nasscom–Zinnov census and in half-yearly hiring data, and it is directly observable, unlike the displacement rate it proxies. Growth below 2% in an annual print, or negative in any half, is the signal that the mandate defence has failed and that the campus, not the country, is where this thesis has to be defended.
Five further observables are set out with thresholds and publication frequencies on page 34, together with the three weakest links in this report and the strongest version of the case against it. A reader who disagrees with our conclusion should start there rather than with the scenario table, because that is where the argument is thinnest and where we would expect to be shown wrong first.
The captive books cost plus 15.5% and is unlisted. The landlord collects 45% of national office leasing. Buy at a forward yield above 6.5% or a discount to net asset value above 15%.
The dollar cost of an Indian engineer fell 2.7% in a year wages rose 9.1%, because the rupee fell 10.8%. Even frozen, the saving needs thirteen years to fall below 50%.
The $105bn target for FY2030 was passed in FY26. Our range is $120–228bn, and even the bear case clears the consensus by 14%.
India taxes the centre at 25.17%, ten points above the floor. The exposure is GIFT City and legacy zone units, whose top-up is currently collected by foreign treasuries.
A $100,000 charge on a new H-1B petition raises the price of onshore delivery. The genuine tail is a 25% outsourcing excise, and it has not moved in committee for nine months.
| Allocator | Long Indian office trusts at the stated levels; systems integrators as an option, not a holding; ignore the currency argument. |
| Private capital | Own the build-operate-transfer platform, underwrite an in-house centre's saving as cash rather than value, and re-mark enabler holdings now. |
| Operator and policymaker | Move the site up the mandate ladder and elect the safe harbour; spend public money on graduates and power, not on subsidy. |
India runs more of the world's corporate work than any other country, and the industry passed its own 2030 targets four years early. What the headline count cannot tell you is where the value in that work now sits.
India had 2,117 Global Capability Centres at the end of FY26, running 3,728 physical units, employing 2.36 million people and billing $98.4bn a year.1 A Global Capability Centre is a company-owned offshore site that performs the parent's own work rather than a vendor's, so the two counts measure different things: 2,117 is the number of multinational parents with a centre here, and 3,728 is the number of buildings they run. Both are correct and neither substitutes for the other, and the ratio has held at about 1.8 units per parent since FY24.2
| Measure | FY2019 | FY2024 | FY2026 | FY24–26 CAGR |
|---|---|---|---|---|
| Parents with a GCC in India | 1,430 | 1,700 | 2,117 | 11.6% |
| Physical units operated | n/a | 2,975 | 3,728 | 11.9% |
| Headcount (millions) | n/a | 1.90 | 2.36 | 11.5% |
| Revenue ($bn) | 31.0 | 64.6 | 98.4 | 23.4% |
| Revenue per centre ($m) | 21.7 | 38.0 | 46.5 | 10.6% |
| Revenue per employee ($k) | n/a | 34.0 | 41.7 | 10.7% |
Counts and revenue from source 1 (FY26) and source 2 (FY19, FY24) ●. Per-centre and per-employee lines are Gravitywell Research calculations from those totals ◐; FY2019 headcount is not published on a comparable basis.
Treat every published GCC forecast as out of date until it is re-based. The centre count, the unit count and the headcount all compound at about 11.5% a year, while revenue compounds at 23.4%: the industry is being paid roughly twice as much per person as its own expansion rate would explain. That is either a shift into higher-value mandates or the first reading of the same revenue-versus-headcount decoupling now visible at the listed IT firms, and the difference decides whether the seat count keeps growing.
The decoupling in §01 has a simple explanation once you look at what the centres now do. Nearly half of India's GCCs operate at what nasscom classifies as high maturity, meaning they own a product line or a global process rather than executing tickets, and 96% of the centres opened since FY21 launched with that ownership from day one.1 Two-thirds of site leaders now carry a dual mandate, running both the site and a global function.1 So the work is worth more per head. The awkward part is that the entity doing the work cannot show it, because a captive is paid its costs plus a fixed markup, and from Apr 2026 that markup is 15.5% by safe-harbour rule.3
| Layer | How it is paid | Margin it books | Owned by an outside investor? |
|---|---|---|---|
| The parent multinational | Keeps the entire wage differential as avoided cost | All of it | Only diluted inside a global P&L |
| The GCC entity | Reimbursed costs plus a fixed markup | 15.5% | No wholly owned, unlisted |
| The landlord | Rent on the campus, ₹102 psf/month on FY26 re-leases | 82% | Yes three listed office REITs |
| The enabler | Build-operate-transfer and managed-services fees | n/d | Private only |
| The systems integrator | Loses the billing the centre insources | Negative | Yes and priced for it, §17 |
| The state government | Pays capital, rent and power subsidy to attract the centre | Negative | Not applicable |
Markup per CBDT safe-harbour rules effective 1 Apr 2026 ◐.3 Landlord margin is Embassy REIT FY26 net operating income over revenue, ₹3,760cr / ₹4,582cr ◐, and the re-leasing rent is its FY26 weighted average agreed rent ●.5
Stop looking for a way to own the centre. The captive is a cost-reimbursed entity whose accounting profit is set by a tax rule, so its books will never reflect the value it creates however good the mandate gets. The two places that value becomes a cash flow an outsider can hold are the campus it sits in and the vendor it displaces, and those are the two exposures this report ends on.
The mandate ladder in §02 moved because the scarce input moved. A decade ago it was cheap process capacity; today it is machine-learning and platform engineering that Western labour markets cannot supply at any price. India supplies both: 1,200 of the 2,117 centres have embedded artificial intelligence in their work, 250 run a dedicated centre of excellence for it, and roughly 250,000 of the 2.36 million people employed are AI specialists.1
| What the parent buys | The evidence | Still the binding reason? |
|---|---|---|
| Scarce engineering and AI skill | 250,000 AI specialists on site; 250 dedicated centres of excellence | Primary |
| Ownership and speed | 96% of post-FY21 centres launched owning a product; 64% of site leaders hold a global function | Primary |
| Cost avoidance | 70–80% saving on a fully loaded senior engineer, §06 | Enabling |
| Operational resilience | Regulated parents must show a controlled, owned site rather than a vendor's floor, §20 | Rising |
| Labour arbitrage on routine process | Outpost-tier centres, which do staffing only, are down to 13% of the base | Receding |
The demand case has flipped from supply of hours to supply of scarce skill, and that is a better business but a narrower one. A centre selling process capacity competes with every other low-cost location and with automation, while a centre selling 250,000 AI engineers competes with nobody because no other market has them at this depth. So the growth question is not whether parents want more Indian centres. It is whether India can produce the engineers, and §05 shows that is where the constraint now sits.
The skills case in §03 only holds if rivals cannot match it, and on price several already beat India. Poland runs 2,081 centres for 1,258 companies employing 488,700 people, and the sector is 5.7% of Polish GDP.6 The Philippines ended 2025 with about 1.9 million workers and $40bn of export revenue, targeting 2.5 million and $59bn by 2028.7 Malaysia has more than 600 firms and 250,000 staff.8 Against that field India is neither the largest employer per dollar nor the cheapest seat, and the revenue-per-employee ladder below is the clearest way to see where each country actually sits.
| Market | Staff | Revenue | Wins on | Takes from India |
|---|---|---|---|---|
| India | 2.36m | $98.4bn | Depth: 2,000-engineer campuses, 250k AI specialists | — |
| Poland | 489k | $42.3bn | EU data residency, European languages, same time zone | EU-regulated finance and R&D for European parents |
| Philippines | 1.90m | $40.0bn | Price and neutral-accent voice work | Customer operations and voice-led process |
| Malaysia | 250k | $8.1bn | Regional treasury, Islamic finance, ASEAN coverage | APAC regional hubs for multinationals |
| Mexico | n/d | n/d | US time zone and nearshore travel | Nearshore analytics and support for US parents |
Poland revenue is CY2024 exports ◐; India FY26 ●; Philippines CY2025 export revenue ◐; Malaysia annual sector revenue ◐. Mexico has no comparable published sector census, which is itself a caution against sizing the nearshore threat precisely.
India's position is defensible where scale is the requirement and vulnerable everywhere else. Nobody else can staff a 2,000-engineer product campus, so platform and product mandates are safe. But voice-led customer work already flows to Manila at half the cost per head, European regulated work sits in Warsaw for data-residency reasons India cannot legislate away, and US analytics increasingly goes nearshore. Read India's rising revenue per head with that in mind: some of it is the mandate upgrade of §02, and some of it is simply wage inflation on a cost-plus book, which §07 separates.
India graduates roughly 1.5 million engineers a year, of whom about 70% are assessed as employable, so the genuinely hireable pool is near 1.05 million.9,10 GCCs added about 460,000 people across FY25 and FY26, an average of 230,000 a year, which is close to a fifth of that pool before the IT services firms, the domestic product companies and the startups take their share.1 The squeeze is sharper in artificial intelligence: GCCs employ around 250,000 AI specialists against a national pool that nasscom and Deloitte put on a path from roughly 650,000 in 2022 to 1.25 million in 2027, while projected openings reach 2.3 million.11
| Constraint | The number | Binds? |
|---|---|---|
| Volume of graduates | 1.5m engineers a year, output rising | No |
| Quality at the entry level | Roughly 30% of graduates fail employability assessment | Partly |
| Depth at the senior and specialist level | Openings of 2.3m against a 1.25m AI pool in 2027 | Yes |
The shortage shows up as price, not as empty desks. India will not run out of engineers, and 9.1% increments against 17.1% attrition are what a market short of senior specialists looks like when it clears through pay. That matters for the thesis because the arbitrage in Part II is measured against onshore salaries rising at 3–4%: every year the Indian premium runs six points hotter, a little of the gap closes on its own. The question §07 answers is how little.
Every mandate in Part I is paid for out of one number: the gap between what an engineer costs in Bengaluru and what the same engineer costs in Seattle. Here is that gap, its cost stack, the clock running against it, and the only part of the chain a public investor can hold.
The price of a seat can be derived rather than guessed. A captive is reimbursed its costs plus a fixed markup, so its revenue per head divided by that markup gives its cost per head: $41.7k of FY26 revenue at a 15.5% markup implies about $36.1k of fully loaded annual cost for the average GCC employee.1,3 A comparable Western seat runs three to five times higher: a senior engineer of seven to ten years costs a US employer $220–320k fully loaded against $48–84k in India, a saving of 70–80% rather than the 50% often quoted.12
| Cost line | $ per seat / yr | Share | Can it protect the arbitrage? |
|---|---|---|---|
| Salary, benefits and statutory contributions | ~28,000 | ~78% | It is the arbitrage. Rises 9.1% a year in rupees. |
| Technology, cloud, security and tooling | ~4,500 | ~12% | No. Priced in dollars, so no saving at all. |
| Real estate and facilities | 1,000–2,500 | 3–7% | No. Too small to matter even at half the rent. |
| Compliance, legal and entity overhead | ~1,500 | ~4% | No, and it rises with transfer-pricing scrutiny. |
| Total fully loaded cost per seat | ~36,100 | 100% | One line carries four-fifths of the case. |
The $84–144m scales the blended seat, not the senior engineer of FIG 6.1. At a $36.1k Indian blended cost and a 70–80% saving, the implied onshore blended seat is $120–180k; the chart's senior role is $270k against $66k, so an all-senior centre would save nearer $204m per 1,000. Total derived from FY26 revenue per employee at the 15.5% statutory markup ◐. Real estate from published per-seat benchmarks ◐13 and cross-checked against Embassy REIT's ₹95 psf/month in-place rent at 80–100 sq ft per seat ●.5 The remaining lines are Gravitywell Research allocations ○ and are indicative, not audited.
Underwrite the arbitrage as a pure wage trade, because that is what it is. Scaled to a benchmark 1,000-seat centre at the blended grade the parent avoids $84–144m a year, and roughly four-fifths of that comes from one line. Cheaper floor space cannot save it, nor can a tax incentive, and a better mandate raises what the centre is worth rather than what it costs. So the only two things that can break this economics are the rupee price of an Indian engineer and the number of engineers the work actually needs. §07 takes the first; §21 takes the second.
The standard bear case says salary inflation eventually eats the saving in §06, and on rupee numbers it looks plausible: Indian pay rose 9.1% in 2026 against a US salary-increase budget of 3.5%, a gap of 5.6 points a year.14,15 But the parent pays in dollars, and the rupee went from an average of ₹86.07 to the dollar in Jul 2025 to ₹96.48 on 24 Jul 2026, a 10.8% fall in the dollar value of any rupee salary.16 Put the two together and the dollar cost of the average Indian engineer fell 2.7% over the year while the American one rose 3.5%. The gap widened by more than six points.
| Rupee path | Net closure per year | Years to a 50% saving | Reached in |
|---|---|---|---|
| Rupee fixed at ₹96.5 | 5.6pp, all of it wage | 13 | 2039 |
| Rupee drifts −3.0% a year | 2.3pp, compounded | 31 | 2057 |
| Rupee repeats the last year, −10.8% | Negative: the gap widens | Never | — |
| Memo: gap today | $270k against $66k | 4.1× | 76% |
Gravitywell Research model ○. Assumes the 2026 increment differential of 5.6pp persists, that mix is constant, and that a 50% saving is the level at which offshoring stops clearing a parent's hurdle. Each is an assumption, not a finding; §23 sizes what happens if they are wrong.
Wage convergence is not the risk to this trade on any horizon an allocator underwrites. Even holding the rupee still, the saving takes thirteen years to fall below 50%, and the rupee has not held still in any five-year window this century. Anyone modelling the GCC thesis breaking on salary inflation is modelling the wrong variable, and should look instead at the number of seats the work requires. That is §21.
A captive earns what the rulebook allows. From 1 Apr 2026 India folded software development, IT-enabled services, knowledge process outsourcing and contract research into one Information Technology Services category with a single safe-harbour margin of 15.5% on operating cost, and lifted the eligibility ceiling from ₹300 crore of covered transactions to ₹2,000 crore.3 A centre that adopts it is not audited on its pricing for five years, which is why §02's mandate upgrade never reaches the entity's accounts: ticket-handler or product owner, the declared margin is the same 15.5%.
| Feature | Until FY26 | From FY27 | Effect on a captive |
|---|---|---|---|
| Service categories | Separate margins for software, ITeS, KPO, contract R&D | One category, Information Technology Services | Simpler |
| Safe-harbour margin | 17–24% depending on category | 15.5% flat | Lower |
| Transaction ceiling | ₹300 crore | ₹2,000 crore | Far wider |
| Advance pricing agreements signed | 815 cumulative to FY25 | 1,034 cumulative; 219 in FY26 alone | Record |
| Bilateral agreements | 200 cumulative | 284, across 13 treaty partners | Double-tax cover |
APA counts from CBDT's FY2025-26 release ◐;18 FY25 figures are the FY26 totals less that year's signings ◐. Pre-2026 margins are the superseded Rule 10TD bands ◐.
Read the 2026 rules as India competing for centres rather than taxing them harder. A lower flat margin, a ceiling nearly seven times higher and a record 219 advance agreements in one year all point one way: New Delhi has decided the prize is the payroll, not the profit. For a parent that removes the largest administrative reason to prefer Warsaw or Manila, and it leaves the tax line too small to break the thesis. §19 tests whether the global minimum tax changes that, and finds it mostly does not.
Rent is a rounding error to the parent and the entire business to the landlord, which is what makes it investable. GCCs leased 19.2 million sq ft in the first half of 2026, 45% of gross office take-up across India's top seven cities, up from 15.78 million sq ft and 41% a year earlier.19 National average rents reached ₹110 per sq ft per month, 9% higher than a year before, while vacancy fell to 13.7% on one house's series and 15.0% on another.20 The denominator is contested: half-year gross leasing is reported at 35.7, 43 and 48 million sq ft by three houses depending on city coverage and whether renewals count, so we anchor to the mid.
| REIT | Price | FY26 DPU | Yield | NAV/unit | Disc. to NAV | Occupancy |
|---|---|---|---|---|---|---|
| Embassy Office Parks | ₹441 | ₹25.28 | 5.7% | ₹491.62 | −10.3% | 94% |
| Brookfield India | ₹343 | ₹21.40 | 6.2% | ₹386.66 | −11.3% | 96% |
| Mindspace Business Parks | ₹493 | ₹24.09 | 4.9% | n/d | n/d | 95.7% |
| Embassy on FY27 guidance | ₹441 | ₹27.00–28.60 | 6.1–6.5% | — | — | 95–96% |
Embassy figures are from its FY26 earnings materials ●;5 Brookfield and Mindspace are from published summaries of their FY26 results ◐;21,22 prices as at 27 Jul 2026 ◐. Embassy occupancy is by asset value; the other two are committed occupancy, so the three are not exactly comparable. Embassy's in-place cost of debt is 7.25%.
Own the floor, not the tenant. Embassy re-let space in FY26 at ₹102 per sq ft against ₹82 expiring, a 24% spread, and still guides to double-digit distribution growth for a second year, which tells you the landlord captures the tenant's expansion without carrying the tenant's automation risk. The catch is honest and worth stating: at a 5.7% trailing yield against a 7.25% in-place cost of debt, the equity is not yet cheap on cash terms, and the case rests on the FY27 guided step-up and the 10% discount to stated asset value closing. §24 sets the entry level that makes that work.
Floor space is where the money lands, and it lands unevenly. Two cities take two-thirds of it, four more specialise, and the tier-2 story that dominates policy speeches accounts for a small share of the people.
The rent in §09 is not spread evenly, and concentration is increasing rather than easing. Bengaluru and Hyderabad together took 69% of all GCC office leasing in the first half of 2026, Bengaluru on 44% and Hyderabad on 25%.19 Bengaluru leased 10.5 million sq ft in total across all occupier types, of which GCCs were about 70%; Hyderabad leased 7.2 million sq ft, a 47% jump on the year.20 Bengaluru houses more than 880 GCC units and roughly a third of the national GCC workforce.23 The two are not interchangeable: one is a depth market, the other a growth market.
| Dimension | Bengaluru | Hyderabad | Edge |
|---|---|---|---|
| Share of GCC leasing, H1 2026 | 44% A | 25% B | Bengaluru |
| Growth in leasing, year on year | Broadly flat B | +47% A | Hyderabad |
| Installed units | 880+ A | 355+ B | Bengaluru |
| Senior and specialist talent depth | Deepest pool in Asia A | Strong but thinner B | Bengaluru |
| Cost and infrastructure headroom | Water, commute and rent pressure C | Planned corridors, cheaper land A | Hyderabad |
| Concentration risk to an occupier | Everyone is already there C | Room to be a large fish B | Hyderabad |
Underwrite Bengaluru for the mandate and Hyderabad for the marginal square foot. A parent placing a first product-owning centre still has one rational choice, because depth of senior engineering talent cannot be manufactured and Bengaluru has a fifteen-year head start. A landlord allocating capital should read it the other way: Bengaluru's constraint is now physical, so the growth in rent per square foot is more likely to come from Hyderabad's corridors than from a Bengaluru market already at 70% GCC penetration. The other four cities are not junior versions of these two, and §11 explains what each of them actually sells.
The 31% of GCC leasing that Bengaluru and Hyderabad do not take is not a consolation market. Each of the four remaining metros sells a different thing, and the rent each commands tracks that specialism closely: Chennai clears at ₹78 per sq ft a month and Mumbai at ₹180, with the Bandra-Kurla Complex at ₹380–400.20 A parent that treats them as interchangeable low-cost options mis-sites the centre, because the binding requirement is usually adjacency, to a regulator in Mumbai, to a manufacturing base in Pune and Chennai, or to a policy ministry and an international airport in the National Capital Region.
| City | GCC leasing share | Units | What it actually sells | Verdict |
|---|---|---|---|---|
| Delhi NCR | 13% | 465+ | Policy access, airport connectivity, shared-services and consulting scale | Broad |
| Pune | 9% | n/d | Automotive and industrial engineering next to the manufacturing belt | Focused |
| Chennai | 7% | 305+ | Hardware, semiconductors, and the cheapest metro floor space | Focused |
| Mumbai | 3% | n/d | Regulated finance that must sit near the regulator and the exchanges | Narrow |
Leasing shares are of GCC take-up in Q1–H1 2026 ◐;19 unit counts from published city directories ◐.23 Mumbai's low leasing share coexists with the highest rent, which is the point: it buys adjacency, not space.
Site the centre against the constraint, not the rent card. Mumbai costs 2.3 times Chennai and is still correct for a bank that must put its risk function inside a regulator's reach, while Chennai is correct for silicon and Pune for powertrains. For a landlord the read is different again: the four specialist markets have narrower tenant pools than Bengaluru, so their rent is less resilient if the specialism moves, and Mumbai's 3% share of GCC take-up is a reminder that a high headline rent is not the same as a deep occupier bid. Which raises the obvious question about the cities further down the ladder, and §12 answers it.
Every state policy in §13 is written around tier-2, so it is worth being precise about the size of the thing. Ahmedabad, Coimbatore, Kolkata, Thiruvananthapuram and Jaipur together hosted more than 250 GCC units and about 85,000 professionals in FY25, which is 3.6% of a national headcount of 2.36 million.24 The share of centres located outside the six established metros moved from 5% in FY19 to 7% in FY24.24 Growth is genuinely faster off the small base, with tier-2 GCC hiring up 21% in 2025 against 11% in the metros, but a fast-growing 3.6% is still 3.6%.
| Requirement | Tier-2 today | Workable? |
|---|---|---|
| Entry-level engineering volume | Local colleges supply steadily; salaries run below metro levels | Yes |
| Floor space and cost | Cheaper land, state rent reimbursement of up to 50% in Karnataka | Yes |
| Senior and specialist hiring | Thin; senior engineers relocate to metros, not away from them | No |
| Global travel and client proximity | Limited long-haul connectivity outside Ahmedabad and Kochi | No |
| Replacement hiring after attrition | A single large employer can exhaust the local pool | Risky |
Treat tier-2 as a delivery annexe, not as the next Bengaluru. It works for volume engineering and back-office scale where the pyramid is wide and junior, and it fails for the product-owning mandates of §02 because the senior layer will not move there. For an allocator that means tier-2 real estate is a yield play on a single tenant rather than a market, and it should be priced as such. For a state government it means the subsidies in §13 are buying the low-value end of a business whose value is migrating in the opposite direction.
States are bidding against each other for the centres §12 says will mostly not move. Karnataka's policy, notified in Nov 2024, targets 500 new centres to reach 1,000 by 2029, with 350,000 jobs and $50bn of output.25 Uttar Pradesh followed in May 2025 targeting more than 1,000 new centres and 500,000 jobs.26 India added 417 centres in the two years to FY26, about 208 a year.1 The arithmetic does not reconcile.
| State · policy | Target | Headline incentive | Status |
|---|---|---|---|
| Karnataka · Nov 2024 | 1,000 GCCs and 350,000 jobs by 2029 | 75% capex for labs; 50% rent reimbursement outside Bengaluru; electricity-duty relief | Live |
| Uttar Pradesh · May 2025 | 1,000+ new GCCs; 500,000 jobs | Land subsidy, full stamp-duty reimbursement, payroll support | Live |
| Madhya Pradesh · 2025 | 50+ GCCs by 2030 | Capital subsidy, payroll support, research grants | Live |
| Andhra Pradesh, Maharashtra, Odisha, Gujarat, Tamil Nadu, Rajasthan, Bihar | Various | Capital and payroll subsidy; Andhra Pradesh reimburses up to ₹3 lakh of salary per employee | Announced |
| Union · national framework, Budget 2025-26 | Guidance to states on tier-2 centres | No direct fiscal incentive | Draft |
Policy terms as published by each state ◐.25,26,27,28 The Union framework was announced in the Feb 2025 budget and remained in consultation at Jul 2026 ◐.28
State subsidy is a transfer to parents, not a driver of location. The centres in §10 cluster where senior engineers already live, and no rent reimbursement changes that: 69% of leasing went to two cities in a year when a dozen states were bidding. So the auction's practical effect is that a multinational which had already chosen Bengaluru or Hyderabad now collects a cheque for going there. The useful policy lever is the one §05 identified, graduate quality and specialist depth, and none of these policies pulls it hard. Part IV turns from where the centres sit to what they do, because that decides which survive automation.
Not every centre carries the same risk. The work that codifies best is the work that automates first, which puts the deepest sector in the industry at the front of the queue and the most engineering-heavy sector at the back.
What the centres actually do starts with banking. Banking, financial services and insurance is the largest single sector, at roughly 28% of India's GCCs, and together with technology it accounts for 56% of all GCC hiring.29,30 It hires hardest relative to its base too, at 1.17 times.30 Concentration inside it is extreme: JPMorgan Chase employs about 55,000 people in India, HSBC 42,000, Wells Fargo 37,000 and Goldman Sachs 9,000, so four institutions carry roughly 143,000 seats. That depth was built on process work, and process work is what §21's automation reaches first.31
| Work type | Why it moved offshore | Automation exposure |
|---|---|---|
| Reconciliation, settlement, reporting | High volume, fully codified rules | Very high |
| Know-your-customer and onboarding review | Labour-intensive document handling | Very high |
| Application testing and maintenance | Repeatable, specification-driven | High |
| Model validation and quantitative research | Scarce quantitative talent at a fraction of the cost | Moderate |
| Regulatory change and controls design | Judgement, accountable to a named officer | Low |
| Platform and product engineering | Ownership mandate, §02 | Low |
Exposure ranking is a Gravitywell Research assessment ○ based on how codified each activity is and whether a regulator requires a named human owner. It is a judgement, not a measurement, and §21 gives the observable that would confirm or refute it.
Depth and exposure are the same trait in this sector. A bank that put 40,000 people into India did so because its work was standardised enough to write down, and anything standardised enough to write down is standardised enough to automate. The offset is real but partial, because regulated finance must keep an accountable human owner for controls and regulatory change, which floors a slice of the headcount. So expect BFSI centres to keep growing in mandate while flattening in seats, and read any bank announcing a large Indian hiring programme in 2027 as evidence against this call rather than for it.
Against BFSI's exposure, engineering-led centres hold the defensible ground. Manufacturing accounts for about 12% of India's GCCs and industrial plus automotive mandates take roughly 15% of GCC hiring, so this is a minority of the industry by headcount.29,30 It is not a minority by strategic weight. India employs roughly 125,000 semiconductor design engineers, about a fifth of the world's total, and Indian teams tape out around 3,000 integrated circuits a year.32 Silicon, powertrain and airframe work carries three properties the codified processes of §14 lack: physical test benches, certification signed by a named engineer, and knowledge that is not written down anywhere.
| Property | What it means in practice | Transfers to a model? |
|---|---|---|
| Physical validation | Silicon tapes out, engines run on a bench, airframes are flown | No |
| Certification liability | A named engineer signs; regulators hold that person accountable | No |
| Tacit process knowledge | Yield tuning and failure analysis live in teams, not documents | No |
| Toolchain lock-in | Design suites are licensed, versioned and audited per project | Partly |
| Specification-driven coding | The one layer that does automate, and it is the junior layer | Yes |
Weight engineering mandates higher than their headcount suggests. A fifth of the world's chip designers already sit in India, and that work cannot be re-hosted to a model or moved to Manila, which makes Pune and Chennai structurally stickier than any BFSI back office of comparable size. The trade-off is that this cohort hires below its footprint, so it will not carry the national seat count. For an allocator, a landlord with engineering tenants owns a longer effective lease than the stated term implies; one leaning on process-heavy financial tenants owns the opposite.
The engineering cohort in §15 will not carry the seat count, and neither will the large-cap cohort, because that one is close to fully penetrated. 506 of the Forbes Global 2000 already run a centre in India and only 105 Fortune 500 companies have yet to open one.1,33 What is left is smaller and, for the first time, financially sponsored: 583 of India's centres belong to mid-market parents and 504 are backed by private equity.1 Healthcare and life sciences shows the shift most clearly, at about 10% of centres but 11% of 2026 hiring, the only large sector other than technology hiring above its footprint.29,30
| Cohort | Size today | What it needs | Who serves it |
|---|---|---|---|
| Mid-market parents | 583 centres, projected 950+ by FY30 | 50–300 seats, no in-house India capability, fast setup | Build-operate-transfer |
| Private-equity portfolio companies | 504 centres | Margin expansion inside a hold period, exit-ready structure | Sponsor-led platforms |
| Healthcare and life sciences | ~10% of centres, 11% of hiring | Clinical data, pharmacovigilance, regulatory submissions | Own centres |
| Retail and consumer | ~8% of centres, 7% of hiring | Supply chain, merchandising analytics, digital storefront | Own centres |
| Large-cap multinationals | 506 of the Forbes G2000 | Expansion of existing sites, not new ones | Already in-house |
The next thousand centres are smaller, sponsored and served by someone else. A mid-market parent opening 150 seats cannot build its own entity, payroll, compliance and facility stack, so it rents them, which is why the enabler layer of §02 is where the private-capital money in this chain actually is. That also changes the risk, because 504 centres now sit inside private-equity hold periods and answer to exit timetables rather than to a parent's strategy. §25 takes that up; §17 first settles what the centres have taken from the vendors.
The work in §14 and §16 used to be bought from a vendor. A global systems integrator, the third-party IT-services firm a parent hires instead of building its own centre, now competes with the captive it often helped set up, and the evidence sits in the vendors' own FY26 accounts. Tata Consultancy Services cut net headcount by 23,460 to 584,519, a fall of 3.9%, while lifting revenue per employee 3.4%.34 It trades at 15.6 times earnings against its 2011–2025 average of 25.5 times, a 39% discount to its own history.36
| What moved | Evidence | Who gained |
|---|---|---|
| Application maintenance and testing | Captives now run it in-house at cost plus 15.5%, §08 | The parent |
| Senior engineering talent | GCCs pay above vendor scales; sector increments 9.1% against 6.9% at the integrators | The GCC |
| Headcount growth | TCS net headcount −23,460 in FY26 while revenue rose | Nobody |
| Setup and transition work | Build-operate-transfer mandates for mid-market parents, §16 | The integrator |
| Multi-year managed services | Still the integrator's, where the parent lacks scale to insource | The integrator |
| The valuation | 15.6×, 14.0× and 19.4× against a 25.5× history | The buyer, later |
Increment differential from the 2026 salary surveys ◐;14 headcount and revenue per employee from FY26 results ◐.34,35 The ledger's attribution of cause is a Gravitywell Research judgement ○: the vendors face artificial intelligence and captive insourcing at the same time, and the published data does not separate them.
The integrators are cheap for a reason, and the reason is not temporary. Two forces hit the same revenue line: captives take the work in-house, and models take the labour intensity out of what remains. The vendors are the worse exposure because they carry the pricing risk without owning the mandate. Even so, at 14 to 19 times earnings with dividend yields above 4%, a buyer is now paid to be wrong, which is why §24 treats this as a valuation call rather than a structural short.
The rules decide how much of the saving the parent keeps. India's own tax case for a centre has thinned to almost nothing, the global minimum tax lands more narrowly than the commentary claims, and the sharpest external risk arrives through immigration policy rather than trade policy.
How much of that saving the parent keeps depends on the rulebook, and India's has thinned to almost nothing. A parent choosing India in 2015 could put its centre in a special economic zone and pay almost nothing on the profit for five years. That option closed: the section 10AA deduction is available only to units that began operating before 1 Apr 2021, and from 1 Apr 2026 the provision itself moved into section 144 of the new Income-tax Act, 2025 without being reopened.37 A centre incorporated today pays the standard 25.17% domestic corporate rate on the 15.5% markup of §08, which works out at ₹3.90 per ₹100 of cost. The 2% equalisation levy went in Aug 2024.38 What remains as an incentive is the state auction of §13.
| Instrument | A centre opened in 2015 | A centre opened in 2026 | Worth |
|---|---|---|---|
| SEZ profit deduction, s.10AA | 100% for 5 years, 50% for 5 more | Not available | Gone |
| Headline corporate rate | ~34% before the 2019 cut | 25.17% under s.115BAA | Better |
| Equalisation levy exposure | 2% on specified digital supplies | Abolished Aug 2024 | Better |
| Transfer-pricing certainty | Category-specific safe harbours, ₹300cr ceiling | One 15.5% margin, ₹2,000cr ceiling, 5 years | Better |
| IFSC deduction, s.80LA | Available in GIFT City | Available, but built for financial-services units | Narrow |
| State cash | Generic IT-park incentives | GCC-specific capital, rent and payroll subsidy | The main lever |
Do not underwrite an Indian centre on a tax argument, because there is not one left at the Union level. The rate is ordinary, the export holiday is closed, and the genuine improvements of the last two years are all about certainty rather than cost. That is the right trade for India, since §08 showed the corporate tax take is trivial next to the payroll, but it means a parent comparing India with Poland or Malaysia is comparing wages, talent depth and state cash, not tax rates. It also sets up the question in §19: if India already taxes above the global floor, how much of the Pillar Two commentary actually applies here?
Much of the commentary treats the OECD global minimum tax as a threat to India's offshore economics, and for the ordinary centre it is not. Pillar Two tops up a group's effective rate wherever it falls below 15%, and an Indian captive taxed at 25.17% on its markup is nowhere near that line.17,39 India had not enacted a qualified domestic minimum top-up tax as at mid-2026, so any top-up on low-taxed Indian profit is collected by the parent's jurisdiction rather than by New Delhi.39 India's first response was an accounting one, with amended accounting-standard rules notified in Mar 2026.40
| Entity type | Effective rate | Exposed to top-up? | Collected by |
|---|---|---|---|
| Captive on cost-plus, s.115BAA | 25.17% | No, well above the floor | — |
| Legacy SEZ unit still in holiday | 0–12.6% | Yes, while the deduction runs | Parent jurisdiction |
| GIFT City IFSC unit, s.80LA | ~0% during the holiday | Yes, and the incentive loses its value | Parent jurisdiction |
| Group as a whole | Blended | Depends on the whole group's map | Parent jurisdiction |
Because India has no qualified domestic top-up tax as at Jul 2026, any top-up on Indian low-taxed profit is collected abroad ◐.39 That is a fiscal loss to India rather than a cost to the parent, and it is the strongest argument for New Delhi legislating one.
Price Pillar Two as a narrow risk to India's remaining tax incentives, not to the offshore model. The ordinary captive is unaffected because India taxes it above the floor, which is §18's point restated: India is not a low-tax jurisdiction for this activity. Where it bites is GIFT City and the legacy special-economic-zone units, whose holidays now transfer value to a foreign treasury rather than to the investor. Watch for India enacting a domestic top-up tax, which would capture that money at home and signal the incentive islands are being wound down.
Two American measures point in opposite directions for India's centres, and the market treats them as one. A presidential proclamation of 19 Sep 2025 imposed a $100,000 supplemental fee on certain new H-1B petitions, effective from 21 Sep 2025 and running through at least Sep 2026.41 That raises the cost of bringing an Indian engineer to America, which argues for leaving the work in India. The measure that would genuinely hurt is the HIRE Act, introduced in the Senate in Oct 2025, proposing a 25% excise on outsourcing payments plus loss of the deduction, a combined burden near 46%.42 It has been stuck in committee since.
| Measure | Mechanism | Direction | Status |
|---|---|---|---|
| H-1B supplemental fee, $100,000 | Raises the cost of onshore delivery by an Indian national | Tailwind | In force |
| EU Digital Operational Resilience Act | Critical third-party providers face direct supervision, so parents prefer an owned site to a vendor's floor | Tailwind | In force |
| India DPDP Rules | Permissive by default; cross-border provisions from May 2027 | Neutral | Phasing in |
| HIRE Act, S.2976 | 25% excise plus loss of deduction, roughly 46% combined | Severe | Stalled |
The HIRE Act targets payments to foreign persons for services benefiting US consumers ◐;42 a third-party estimate puts the margin impact on Indian IT services at 500–1,000 basis points ◐. It is a proposal, and pricing it as anything else would be wrong.
Buy the confusion between the two measures. Restricting the movement of Indian engineers into America raises the value of the work staying in India, and Europe's resilience rules push regulated parents toward centres they own rather than vendors they hire. The genuine tail is an outsourcing excise, and it is a proposal with no committee progress in nine months, so it belongs in the risk register at low probability and high severity rather than in the base case.
One variable decides the answer, and it is not policy. If artificial intelligence takes a large enough share of the codified work, the seat count stops compounding and the whole chain reprices. Here is that arithmetic, the range it produces, and what each reader should do about it at today's prices.
Every other variable in this report moves slowly. This one is already moving. Tata Consultancy Services cut net headcount 3.9% in FY26, to 584,519, while lifting revenue per employee 3.4%, and Infosys reduced headcount by 8,440 in a single quarter to 328,594.34,35 Over the same period India's GCCs grew headcount 11.5% a year and revenue per head 10.7% a year, and hired 227,991 people in the first half of 2026 alone, 11% more than a year earlier.1,45 Both groups are getting more output per person. Only one is still adding people, and §02 explains why: a centre that owns a mandate expands scope when it automates, while a vendor billing by the hour simply bills less.
| Layer | Share of headcount | Displacement by 2030 | Seats at risk |
|---|---|---|---|
| Outpost and satellite tiers, execution work | 56% ≈ 1.32m people | Bear 40% · Base 20% · Bull 8% | 106k – 529k |
| Portfolio and transformation tiers, owned mandates | 44% ≈ 1.04m people | Scope expands rather than shrinks | — |
| Net effect on the FY26 base, before growth | 2.36m | Bear −22.4% · Base −11.2% · Bull −4.5% | −529k to −106k |
Gravitywell Research model ○. Tier shares are nasscom's FY26 maturity distribution ●;1 displacement rates are assumptions, not observations, and are the single largest source of uncertainty in this report. §22 applies them; §23 sizes what happens if they are wrong.
Watch net GCC headcount, not commentary. The mechanism is visible in the vendors' accounts today and absent from the centres' numbers, and that difference is structural rather than a matter of timing. A centre with a product mandate that automates a workflow redeploys the people onto the next workflow, because its budget is set by the parent's ambition rather than by billable hours. The moment that stops being true it will show up as a single number, national GCC headcount growth turning negative, and nothing in the FY26 data suggests it yet.
The displacement rates in §21 are the only real disagreement, so the scenarios are built on them rather than on a growth assumption. FY26 revenue of $98.4bn across 2.36 million people is the starting point.1 The bull case extends the FY24–FY26 rate of 23.4% for four more years to $228bn, on continued mid-market formation rather than any one cohort arriving. The base case slows headcount growth to 6.2% a year with revenue per head rising 8%, giving $170bn. The bear case has headcount effectively flat, edging from 2.36m only to 2.40m as execution work is displaced almost as fast as mandates expand, leaving revenue at $120bn. All three sit above the $105bn forecast still in general circulation.2
| Case | Headcount 2030 | Rev/head | Revenue | What has to be true |
|---|---|---|---|---|
| Bear | 2.40m | $50k | $120bn | Execution work displaces at 40%; mandate expansion barely offsets it, so the base of 2.36m edges only to 2.40m. Rupee stable. |
| Base | 3.00m | $56.7k | $170bn | Displacement at 20%; headcount compounds 6.2% as mid-market and sponsor-backed centres of §16 arrive; revenue per head +8% a year. |
| Bull | 3.60m | $63.3k | $228bn | Displacement at 8%; the FY24–FY26 pace continues on mid-market and sponsor-backed formation plus expansion of existing sites. The 105 absent Fortune 500 names add centres, not the bulk of the seats. |
| Memo · consensus | n/d | n/d | $105bn | Set in Sep 2024, already exceeded in FY26. |
Gravitywell Research model ○. Revenue is the product of the two columns and may differ by rounding. The model assumes the 15.5% markup holds, so a markup change would shift all three cases proportionally.
We weight the base case, because the bear case requires something the data does not yet show. To hold headcount flat at 2.4 million, displacement has to run at twice the rate we can observe anywhere, while the mid-market and private-equity cohorts of §16 stop arriving. Both could happen; neither is happening. The variable that moves us between cases is the displacement rate itself, and §23 shows exactly how much of the range each input owns.
A range is only useful if you know which assumption produced it. The displacement rate of §21 is not one input among several. It is the driver that sets both levers the scenarios move, because the share of execution work that survives determines how many seats the industry adds and what the average seat is worth. That is why the §22 cases differ on headcount and revenue per head together, and why the $108bn spread belongs to that one judgement rather than to four independent ones. The tornado below asks a different question: what each lever is worth alone, holding the rest at base.
| Stress | India cost | US cost | Saving | Change |
|---|---|---|---|---|
| Today, ₹96.5 to the dollar | $66k | $270k | 75.6% | — |
| Rupee 10% stronger, ₹86.9 | $73k | $270k | 72.9% | −2.7pp |
| Rupee 10% weaker, ₹106.2 | $60k | $270k | 77.8% | +2.2pp |
| Wage gap closes 5.6pp a year for 5 years | $86k | $270k | 68.1% | −7.5pp |
| Wage gap closes 5.6pp a year for 13 years | $134k | $270k | 50.4% | −25.2pp |
Gravitywell Research model ○ at the midpoints of §06. Wage stress holds the rupee at ₹96.5 and applies the 2026 increment differential of 5.6pp, which is the assumption §07 flagged as the least reliable.
Stop hedging the rupee and start monitoring the mandate. Currency and wages together move the saving by single-digit percentage points over a five-year horizon, and no allocator changes a position for that. The displacement rate moves the answer by more than the entire current revenue of the industry, it drives the other two levers rather than sitting beside them, and it is not directly observable, which is why page 34 lists the proxies you can actually watch. That asymmetry also decides how to express the view: own instruments whose cash flow survives the bear case rather than instruments that need the bull case, and §24 prices exactly that.
The asymmetry in §23 sets the rule: prefer the cash flow that survives the bear case. On that test the listed office landlord is cleanest, because its rent is contracted, occupancy runs 94–96% and its tenant's automation risk does not reach the lease. Embassy REIT at ₹441 yields 5.7% on FY26 distributions and 6.1–6.5% on FY27 guidance of ₹27.00–28.60, against net asset value of ₹491.62 and a 7.25% in-place cost of debt.5 Brookfield India at ₹343 yields 6.2% on ₹21.40 and trades 11.3% below its ₹386.66 net asset value.21
| Expression | Instrument | Price today | What you are paid for | Survives bear? |
|---|---|---|---|---|
| The landlord | Embassy REIT · Brookfield India REIT | 6.1–6.5% fwd · 5.7% trailing −10.3% and −11.3% to NAV | Contracted rent, 94–96% occupancy, 24% FY26 re-leasing spread | Yes |
| The vendor | TCS · Infosys · HCLTech | 15.6× · 14.0× · 19.4× vs a 25.5× history | A 39% discount to history and 2.8–4.4% dividend yields | No |
| The enabler | Build-operate-transfer and managed-office platforms | Private | Fees on the 583 mid-market and 504 sponsor-backed centres of §16 | Partly |
| The centre itself | None | Not investable | Cost plus 15.5%, wholly owned, unlisted | n/a |
Own the landlord, at a price, and treat the vendor as an option rather than a holding. The office trusts are not cheap on running yield, since 6.1–6.5% sits below a 7.25% cost of debt, so the case rests on the guided distribution step-up and on the 10% discount to stated assets closing, and it fails if either slips. Our entry discipline is a forward yield above 6.5% or a discount to net asset value wider than 15%, and Brookfield is closer to both than Embassy. The integrators at 14 to 19 times are a different proposition: they need the bull case in §22 to be right, so size them as a position you can be wrong about, not as the way you express this view.
Public and private capital reach this theme through different doors and should not use the same instrument. The public-market investor buys contracted rent from a tenant whose lease outlives its automation risk, priced in §24. The private-market investor is already exposed whether it intended to be or not: 504 of India's 2,117 centres are backed by private equity, so one centre in four answers to a hold-period timetable.1
Buy the office trusts on a forward yield above 6.5% or a discount to net asset value wider than 15%. Brookfield India is nearer both tests than Embassy.
At 14–19× against a 25.5× history the integrators are an option on §22's bull case, not a way to own the structural story.
A 10% rupee move shifts the saving under three points. It does not change a position; the displacement rate does.
Assets let to semiconductor, automotive and aerospace occupiers carry a longer effective lease than the stated term.
| The exposure | The mechanism | What to do |
|---|---|---|
| A GCC inside a portfolio company | Creates real EBITDA, but a buyer can replicate the saving, so it may not earn a multiple at exit | Underwrite as cash, not as value |
| The platform that builds centres | Monetises the same capability across many parents, including the 950+ mid-market cohort of §16 | Own this |
| Sponsor-backed centres facing an exit | Hold-period timetables, not strategy, decide whether the centre scales or is sold | Diligence the mandate tier |
| Marks on unlisted enabler platforms | Carried against listed comparables that have de-rated 39%, so the markdown arrives late | Re-mark before you are asked |
Sponsor-backed and mid-market counts from the FY26 census ●.1 Exit-multiple judgements are Gravitywell Research views ○.
Buy the toll on centre formation, not the saving inside any one centre. A cost reduction a buyer could reproduce in eighteen months does not earn a multiple, which makes building a captive a fine operating decision and a poor value-creation story to sell. The platform layer converts that one-off saving into a recurring fee across hundreds of parents, and ANSR puts that cohort up about 56% by 2030 on its own base. For limited partners the caution is stale marks: the listed comparables in §17 have de-rated 39% and unlisted enabler valuations carried against them have not.
The displacement asymmetry in §23 reads differently from inside the building. For an operator the question is not whether artificial intelligence arrives but whether the site owns enough of a product to redeploy people when it does, measurable as the share of headcount in the portfolio and transformation tiers, 44% nationally.1 For a policymaker the finding in §13 is uncomfortable: a dozen states are paying to attract centres that cluster where senior engineers already live, while the constraint identified in §05 is that specialist depth, not floor space, is what runs short.
Track the share of your site in owned-product work against the national 44%. A centre below that is on the wrong side of §21's displacement.
64% of Indian site leaders now hold a dual mandate. A site whose leader only runs the site is an execution outpost by definition.
The 15.5% margin with a ₹2,000 crore ceiling buys five years without a pricing audit, and the compliance saving exceeds the margin difference for most centres.
Tier-2 works for volume engineering, not for the senior layer. Mumbai buys regulator adjacency at 2.3× Chennai's rent. Match the city to the mandate.
| Lever | What the evidence says | Priority |
|---|---|---|
| Graduate quality | About 30% of 1.5m engineering graduates fail employability assessment, and the shortage is at the specialist end | First |
| Power, water and commute | Bengaluru's constraint is now physical, and it holds 44% of GCC leasing | First |
| A domestic top-up tax | Without one, top-up tax on low-taxed Indian profit is collected by foreign treasuries, §19 | Second |
| The national framework | Announced in Feb 2025, still in consultation seventeen months later | Second |
| Capital and rent subsidy | A dozen states bid for centres that cluster in two cities regardless; the cheque follows a decision already made | Last |
Employability from the 2026 AICTE and industry assessment ●;10 leasing concentration from H1 2026 data ◐;19 policy status as at Jul 2026 ●.28 Priority ranking is a Gravitywell Research judgement ○.
Both readers are being asked to spend on the wrong thing. An operator that measures success by seats filled is optimising the metric §21 says will stop growing, while the one that measures owned mandates is buying the only insurance available. A state that competes on cash is bidding for a decision already taken on talent depth, and the honest lever, harder and slower, is the quality of the engineering graduate and the reliability of the power supply in the two cities that already hold two-thirds of the demand. India's advantage in this industry was never price, as §04 showed, so defending it with subsidy defends the wrong thing.
The central claim is that the value in India's GCCs is migrating from headcount to mandate, and that the tradable expression is the campus rather than the centre. Here is what would falsify it. Each item is a published number with a stated threshold and a stated date, so a reader can hold this report to it rather than take its word.
| Claim | What we say | What would break it | Prints |
|---|---|---|---|
| Headcount keeps growing | GCC headcount compounds 6.2% a year to 3.0m by 2030 | National GCC headcount growth below 2% in any annual census, or negative in any half | Annual |
| Mandates keep rising | 44% of centres own a product or global process, and rising | The portfolio and transformation tiers falling below 40% of the base | Annual |
| Rent is defensible | GCCs take 45% of national office leasing; re-leasing spreads are positive | GCC share of gross leasing below 35%, or a listed trust reporting negative re-leasing spreads | Quarterly |
| The landlord case | Forward distribution yield of 6.1–6.5% with guided growth | Any listed office trust guiding to flat or lower distributions per unit | Quarterly |
| Automation is not yet displacing | Vendors shed headcount; centres do not | Two consecutive quarters of falling aggregate GCC hiring while vendor headcount stabilises | Half-yearly |
| Weakness | Why it matters | Tier |
|---|---|---|
| The displacement rate is an assumption, not a measurement | It drives both scenario levers and so carries the $108bn range in §22. No published dataset measures it, and our 8/20/40% band is a judgement about how much execution work is codifiable. | ○ estimate |
| Named-centre headcounts are not audited disclosures | The four banks in §14 are press and compilation figures. The order of magnitude is safe; the individual numbers may be a year stale. | ◐ modeled |
| The office-leasing denominator is contested | Three brokerages report H1 2026 gross leasing at 35.7, 43 and 48 msf, and vacancy at 13.7% or 15.0%. GCC absolute volumes agree; the shares do not. | ◐ modeled |
The strongest bear argument is not that automation displaces work, which we accept, but that mandate ownership is a weaker defence than we claim. On that reading the 96% of post-FY21 centres that "launched with a product" is a labelling exercise, the maturity tiers are self-reported by the centres being surveyed, and a parent under cost pressure will repatriate or shrink a mandate as easily as it granted one. We take that seriously. Our answer is that the physical evidence, 19.2 million sq ft leased in six months and 227,991 people hired in the same period, is harder to relabel than a maturity tier. If both the leasing and the hiring roll over while the tier mix holds, the bear is right and we are reading a survey instead of a market.
India's Global Capability Centres are a $98.4bn industry compounding revenue twice as fast as its own centre count, and the maturity data says the reason is mandate ownership rather than headcount.1 The wage gap that funds it is not closing on any horizon an allocator underwrites, and the rupee widened it 6.3 points last year alone. The one genuine risk is that automation displaces execution work faster than mandates expand, which swings 2030 revenue by $108bn and is not directly observable. That asymmetry, not the growth rate, should decide how the view is expressed.
Indian office trusts convert GCC expansion into contracted rent that survives the bear case. Price: forward yield above 6.5% or a discount to net asset value wider than 15%; Brookfield India at ₹343 (6.2%, −11.3%) is nearer both than Embassy at ₹441.5,21 Exit trigger: any trust guiding to flat distributions.
A captive is reimbursed cost plus a statutory 15.5% and is wholly owned, so no listed instrument exists. Anything marketed as GCC exposure is exposure to an adjacent layer; identify which row of Table 2.1 before buying.
TCS, Infosys and HCLTech at 15.6×, 14.0× and 19.4× against a 25.5× history are priced for permanent disintermediation. Price: the 4%+ dividend yields pay you to wait. Trigger: stabilising headcount alongside revenue growth would confirm the de-rating overshot.
A cost saving inside one portfolio company is replicable and earns no multiple; a build-operate-transfer platform monetises it across a mid-market cohort ANSR projects up about 56% by 2030. Trigger: mid-market centre formation falling below 60 a year would break the fee pool.
Semiconductor, automotive and aerospace mandates resist substitution, and Hyderabad grew leasing 47% against a physically constrained Bengaluru. Trigger: Bengaluru below 35% of GCC leasing would confirm the shift is structural.
The base rests on a definitive industry census and on landlord and company results, with source variance disclosed and tiers set conservatively: only seven register entries carry ●. The one input carrying the range is marked as an estimate throughout.
All three 2030 cases clear the forecast still in circulation, the regulatory changes of the last two years are net favourable, and the external measures now in force push work toward India rather than away.
Prices as at 27 Jul 2026 ●. This is independent research and not investment advice; the entry levels are the desk's own discipline, not a recommendation calibrated to any reader's mandate or constraints.
This report rests on one definitive industry census, one set of landlord earnings materials read in full, statutory instruments, and a wider body of professional-firm and market reporting, plus nine figures Gravitywell Research constructed. Tiers are set conservatively: only seven of the 45 register entries carry ●, and figures that originate with an official body but were reached through an alert or a press report are marked ◐. The constructed figures are what a due-diligence reader should test first, so the derivation of each is set out here rather than described.
| Figure | Derivation | Tier | Used in |
|---|---|---|---|
| Revenue per employee, $41.7k | FY26 revenue $98.4bn divided by FY26 headcount 2.36m | ◐ | §01, §04 |
| Revenue per centre, $46.5m | FY26 revenue divided by 2,117 parents | ◐ | §01 |
| Cost per seat, $36.1k | Revenue per employee divided by 1.155, the statutory safe-harbour markup | ◐ | §06 |
| Cost-stack shares | Real estate from published per-seat benchmarks; the remaining lines allocated by us to sum to the derived total | ○ | §06 |
| Dollar cost change, −2.7% | 1.091 × (86.07 ÷ 96.48), the 2026 increment applied to the year's currency move | ◐ | §07 |
| Convergence clock, 13 years | ln(4.09 ÷ 2.00) ÷ ln(1.056), the ratio of midpoint costs closing at the 5.6pp increment differential | ○ | §07 |
| Corporate tax of ₹3.90 per ₹100 | 15.5% markup taxed at the 25.17% effective rate | ◐ | §08 |
| Displacement rates, 8 / 20 / 40% | Judgement about how much of the two execution tiers is codifiable. No published series measures this. | ○ | §21, §22 |
| 2030 revenue of $120 / $170 / $228bn | Headcount and revenue-per-head paths compounded from the FY26 base under each displacement case | ○ | §22, §23 |
In scope: company-owned offshore centres in India, their cost structure, geography, sector mix, tax treatment and the listed instruments that carry exposure to them. Out of scope: third-party business-process outsourcing except where it competes directly with a captive; India's domestic technology sector; and the parent-side accounting of avoided cost, which is not disclosed by any parent in a form we could verify. Where data is thin it is said so: Mexico has no comparable published census, Pune has no published unit count, and Mindspace does not disclose net asset value per unit on the same basis as its peers.
Derived measures follow OECD and Joint Research Centre practice for composite indicators: the components are published, the arithmetic is shown, and no weighting is applied that is not stated. Every figure is marked to the date given and will not be silently restated. Material errors are corrected in a dated erratum appended to this report and noted in the next edition, and the permalink always serves the current version. Where sources disagree, this report states the spread and names the anchor rather than selecting a single value.
Conviction is set at High because the base rests on a primary industry census, filed landlord earnings materials and quoted market prices, and because the report's central claim about mandate migration is corroborated by two independent physical measures, office absorption and gross hiring, rather than by survey response alone. It is not set higher because the single input that owns the 2030 range, the displacement rate, is a judgement no published dataset currently measures.
Every in-text n resolves here. ● filed, official or the definitive published dataset · ◐ modeled, derived or a credible secondary estimate · ○ Gravitywell Research estimate. A figure reached only through a secondary outlet quoting a primary is marked ◐.
Where a class of source is described rather than named (city directories, cost benchmarks, brokerage absorption series), several publishers report the same series on different definitions and none is authoritative. In those cases the report states the spread and anchors to the mid, and the tier is ◐ rather than ●. Only seven entries carry ●: 1 and 4 (the industry census and its publisher's own report page), 5 (Embassy REIT's filed earnings materials, parsed directly), 17, 37 and 38 (statutory provisions) and 43 (an EU regulation). Everything else is ◐, including figures that originate with an official body but which we reached through a professional-firm alert or press report rather than the primary text. That is deliberate under-claiming, and the register says which is which. Every entry in the register is cited at least once in the body.
Terms are glossed where they sit outside the home domain of at least one part of this readership, which spans public and private markets, corporate strategy and policy. Terms the whole audience prices cold are not defined.
Twenty-four figures and twenty-eight tables. The series behind each is in the data appendix and in the CSV data pack shipped with this report.
The series behind every exhibit, so the model is reproducible. Tier: ● filed · ◐ modeled · ○ estimate. Source numbers refer to the register on pages 37–38. A machine-readable CSV of these series ships with this report.
| Series | FY19 | FY24 | FY26 | Tier | Src |
|---|---|---|---|---|---|
| Parents | 1,430 | 1,700 | 2,117 | ◐ | 1,2 |
| Units | n/a | 2,975 | 3,728 | ◐ | 1,2 |
| Headcount (m) | n/a | 1.90 | 2.36 | ◐ | 1,2 |
| Revenue ($bn) | 31.0 | 64.6 | 98.4 | ◐ | 1,2 |
| Rev/centre ($m) | 21.7 | 38.0 | 46.5 | ◐ | 1,2 |
| Rev/employee ($k) | n/a | 34.0 | 41.7 | ◐ | 1,2 |
| FY2030 forecast ($bn) | — | 105.0 | — | ◐ | 2 |
| Series | FY26 | Tier | Src |
|---|---|---|---|
| Outpost tier | 13% | ● | 4 |
| Satellite tier | 43% | ● | 4 |
| Portfolio hub | 39% | ● | 4 |
| Transformation hub | 5% | ● | 4 |
| Post-FY21 centres with a product mandate | 96% | ● | 1 |
| Site leaders with a dual mandate | 64% | ● | 1 |
| Centres with embedded AI | 1,200 | ● | 1 |
| Dedicated AI centres of excellence | 250 | ● | 1 |
| AI specialists employed | 250,000 | ● | 1 |
| Post-FY21 cohort (modelled) | 513 | ○ | 1 |
| Market | Staff | Revenue | $k/head | Tier | Src |
|---|---|---|---|---|---|
| India | 2.36m | $98.4bn | 41.7 | ● | 1 |
| Poland | 488,700 | $42.3bn | 86.6 | ◐ | 6 |
| Philippines | 1.90m | $40.0bn | 21.1 | ◐ | 7 |
| Malaysia | 250,000 | $8.1bn | 32.4 | ◐ | 8 |
| Series | Value | Tier | Src |
|---|---|---|---|
| Engineering graduates per year | 1,500,000 | ◐ | 9 |
| Assessed employable | 70.15% | ◐ | 10 |
| Hireable pool | 1,050,000 | ◐ | 9,10 |
| GCC annual absorption | 230,000 | ◐ | 1 |
| Share of hireable pool | 21.9% | ◐ | 1,9,10 |
| AI talent pool 2027 | 1,250,000 | ◐ | 11 |
| AI openings 2027 | 2,300,000 | ◐ | 11 |
| 2026 salary increment | 9.1% | ◐ | 14 |
| Attrition | 17.1% | ◐ | 14 |
| Series | Value | Tier | Src |
|---|---|---|---|
| US senior engineer, fully loaded | $220–320k | ◐ | 12 |
| India senior engineer, fully loaded | $48–84k | ◐ | 12 |
| Cost per seat, all grades | $36.1k | ◐ | 1,3 |
| Real estate per seat | $1.0–2.5k | ◐ | 13 |
| Salary share of the stack | ~78% | ○ | — |
| Saving per 1,000 seats | $84–144m | ○ | 12 |
| Series | Value | Tier | Src |
|---|---|---|---|
| India increment 2026 | +9.1% | ◐ | 14 |
| US salary budget 2026 | +3.5% | ◐ | 15 |
| USD/INR, Jul 2025 average | 86.07 | ◐ | 16 |
| USD/INR, 24 Jul 2026 | 96.48 | ◐ | 16 |
| Rupee value change | −10.8% | ◐ | 16 |
| India cost change in USD | −2.7% | ◐ | 14,16 |
| Gap widening | +6.3% | ◐ | 14,15,16 |
| Years to a 50% saving, rupee flat | 13 | ○ | — |
| Series | Value | Tier | Src |
|---|---|---|---|
| GCC leasing H1 2025 | 15.78 msf | ◐ | 19 |
| GCC leasing H1 2026 | 19.20 msf | ◐ | 19 |
| GCC share of gross leasing | 45% | ◐ | 19 |
| National average rent | ₹110 psf/mo | ◐ | 20 |
| Vacancy, Q2 2026 | 13.7% | ◐ | 20 |
| Embassy price / DPU / NAV | ₹441 / 25.28 / 491.62 | ● | 5 |
| Embassy FY27 guidance | ₹27.00–28.60 | ● | 5 |
| Embassy cost of debt | 7.25% | ● | 5 |
| Brookfield price / DPU / NAV | ₹343 / 21.40 / 386.66 | ◐ | 21 |
| Mindspace price / DPU | ₹493 / 24.09 | ◐ | 22 |
| City | GCC leasing | Units | Rent | Tier |
|---|---|---|---|---|
| Bengaluru | 44% | 880+ | n/d | ◐ |
| Hyderabad | 25% | 355+ | n/d | ◐ |
| Delhi NCR | 13% | 465+ | ₹110 | ◐ |
| Pune | 9% | n/d | ₹95 | ◐ |
| Chennai | 7% | 305+ | ₹78 | ◐ |
| Mumbai | 3% | n/d | ₹180 | ◐ |
| Tier-2, five cities | n/d | 250+ | n/d | ◐ |
Tier-2 headcount ~85,000, 3.6% of the national total; tier-2 hiring +21% in 2025 against +11% in the metros ◐.24
| Sector | Centres | Hiring | Tier |
|---|---|---|---|
| BFSI | 28% | 28% | ◐ |
| Technology | 25% | 28% | ◐ |
| Manufacturing | 12% | 9% | ◐ |
| Healthcare, life sciences | 10% | 11% | ◐ |
| Retail, consumer | 8% | 7% | ◐ |
| Named BFSI seats, four banks | ~143,000 | ◐ | |
| Semiconductor design engineers | ~125,000 | ◐ | |
| Series | Value | Tier | Src |
|---|---|---|---|
| TCS P/E · dividend yield | 15.6× · 2.76% | ◐ | 34 |
| Infosys P/E · dividend yield | 14.0× · 4.43% | ◐ | 35 |
| HCLTech P/E · dividend yield | 19.4× · 4.15% | ◐ | 35 |
| TCS 2011–2025 average P/E | 25.5× | ◐ | 36 |
| TCS FY26 net headcount | −23,460 | ◐ | 34 |
| 2030 revenue · bear | $120bn | ○ | — |
| 2030 revenue · base | $170bn | ○ | — |
| 2030 revenue · bull | $228bn | ○ | — |
| Displacement band | 8 / 20 / 40% | ○ | — |
Machine-readable copy: gcc-india-data.csv ships alongside this report, with columns exhibit, series, period, value, unit, tier, source_ids, note. Rounding may cause derived rows to differ from the product of their inputs in the last digit.
The Gravitywell Research Global Technology Desk, which is responsible for this report, certifies that the views expressed accurately reflect its independent judgement about the subjects and securities discussed, and that no part of its compensation was, is, or will be directly or indirectly tied to the specific recommendations or views expressed herein.
As of the publication date, Gravitywell Research and its analysts do not hold positions in the securities, trusts or assets discussed in this report. Gravitywell Research has no advisory, banking or commercial relationship with any entity named. This report was not commissioned, sponsored, or reviewed prior to publication by any issuer, trust, industry body, government or other entity named in it, and no such party received an advance copy.
Conviction — High, Medium or Low, set by coverage depth, source tier and how far the thesis has been stress-tested. Stance — Constructive, Neutral or Cautious, the direction of the house view on the opportunity. The two are independent: a Cautious stance can be held with High conviction. This report: High conviction · Constructive. The rationale is on the methodology page and the falsification conditions are on page 34.
Every figure carries a confidence tier: ● filed, official or the definitive published dataset; ◐ modeled, derived, or a credible secondary estimate; ○ a Gravitywell Research estimate. Nine constructed figures are listed with their arithmetic in Table 28.1. Every figure is marked to the date shown and will not be silently restated in a later edition. Where sources disagree, the report states the spread and names its anchor rather than selecting a single value; the office-leasing denominator and the mid-market centre count are the two live examples.
For professional and institutional investors across public and private markets, sophisticated individual investors, and policymakers. Not for general retail distribution, nor for readers who lack the expertise to assess the assumptions set out here. Intended recipients may not redistribute without attribution. Availability of this research in some jurisdictions may be restricted; recipients are responsible for their local rules. This is independent research. It is not regulated ratings, not investment advice, and not an offer or solicitation.
Material errors are corrected in a dated erratum appended to this report and noted in the next edition; the permalink always serves the current version. Permalink: www.gravitywellresearch.xyz/research/gcc-india. Correspondence is directed to the Global Technology Desk through the firm's site; this report carries no named individual author by policy. Version: GWR-2026-IN-002 · v1.0 · Jul 2026 · as of 27 Jul 2026.
First edition. Gravitywell Research has published no prior call on India's Global Capability Centres, so there is nothing to score. From the second edition onward this section will carry a dated scorecard of what we said, what happened, and whether we were right, wrong or early.
Derived measures follow OECD and Joint Research Centre practice for composite indicators: components published, arithmetic shown, no undisclosed weighting. Coverage and its gaps are disclosed on the methodology page rather than left implicit. No index or benchmark is constructed here, so IOSCO benchmark principles are not engaged; where a future edition constructs one, it will be run through the desk's index review and validation process before publication.
2,117 centres · 2.36 million people · $98.4bn · and one variable that decides the next $108bn.