# India's GCCs were built on arbitrage. AI is taking the arbitrage away.

> India's Global Capability Centres were built on a wage gap. Generative AI is compressing that arbitrage from both ends — and handing the GCCs that become capability centres, not cost centres, a bigger prize than the one they were built for. A grounded read.

Source: https://theaidaily.in/analysis/gcc-india.html
Published: 2026-11-17

---

The signal behind the AI headlines — **ranked, not recapped**[Get the daily brief →](/subscribe.html)

[********The AI Daily](/)[Ask Aida](/ask.html)[Subscribe →](/subscribe.html)☰

🇮🇳 Deep dive · India & GCCs

India’s GCCs were built on arbitrage. AI is taking the arbitrage away.

For two decades, Global Capability Centres turned India’s wage gap into a global advantage. Generative AI is quietly closing that gap — and handing the GCCs that move up the value chain a bigger prize than the one they were built for.

The AI Daily · Analysis · **November 17, 2026**

## The one-scroll version

- **A GCC is a spread, not a moat.** More than **1,700 GCCs** employ ~1.9M people in India and steer **$55B** of its IT exports — and the whole model “requires the labour-arbitrage gap to persist.”

- **AI compresses that spread from both ends.** Persistent’s CEO puts the effort deflation at **25–30%**; the same support bot returns ~42× in the US but only ~**5×** in India, where the labour it replaces was cheap to begin with.

- **The market has already repriced it.** Accenture fell **18% in one session**; Indian peers are down **22–44%** over a year; Infosys has guided FY27 growth to just **1–2%**.

- **The hiring mix inverts.** Fewer generalist engineers, far more domain experts — and the entry-level rung is the casualty.

- **The winners become capability centres.** Outcomes over hours: Infosys’s AI-first GCC for Truist, LTIMindtree’s first **$150M** of AI revenue, TCS’s **$7.4B** pivot to compute landlord.

## The machine that arbitrage built

Global Capability Centres — the in-house offshore arms multinationals run out of Bengaluru, Hyderabad and Pune — are one of the most successful structural bets in modern business. India now hosts **more than 1,700 of them, employing roughly 1.9 million people and generating about $64.6B in revenue**, on track for **$99–105B by 2030** ([Zinnov–Nasscom India GCC Landscape Report, FY2024](https://community.nasscom.in/communities/nasscom-insights/india-gcc-landscape-report-5-year-journey); [BW](https://www.businessworld.in/article/india%E2%80%99s-gcc-market-revenues-to-touch-99-105-bn-by-2030-532640)). By our own read, GCCs already steer **$55B** of India’s IT exports ([The India AI Stack](/analysis/india-ai-stack.html)). The logic never changed: same work, same quality, a fraction of the wage.

But that logic rests on one assumption — that the wage gap is a moat. It isn’t a moat. It’s a *spread*. And as we argued in [The Demand Paradox](/analysis/it-services-paradox.html), the entire offshore model “requires the labour-arbitrage gap to persist.” Generative AI is compressing that spread from both ends.

## The squeeze: when the spread stops paying

A Western client pays **$150–200 an hour** for work a Bengaluru developer does at **$25–35** ([The Demand Paradox](/analysis/it-services-paradox.html)). Generative AI attacks that from the demand side: a Seattle team with Claude Code turns a ten-person offshore engagement’s 10,000 hours into 2,000 — same billing rate, far fewer hours. Persistent’s CEO puts the deflation at **25–30%** ([India AI Stack](/analysis/india-ai-stack.html)). And the automation math that was supposed to rescue margins doesn’t rescue them here: the same support bot that returns ~**42×** in the US returns only ~**5×** in India, because the labour it replaces was cheap to begin with — even as Indian BPO wages climb ~**9.5% a year** ([What Would It Take for India Inc to Scale AI Adoption?](/analysis/india-ai-adoption.html)).

The market has already repriced this. In June 2026, Accenture fell **18% in a single session** to a nine-year low after cutting growth guidance to 3–4%; Indian peers — HCL, Infosys, LTIMindtree, Wipro — followed, down **22–44%** over the trailing year. Infosys has since guided FY27 growth to just **1–2%** ([The Demand Paradox](/analysis/it-services-paradox.html); [our Oct 24 digest](/issues/2026-10-24.html)). As we put it: *“The market is not picking winners and losers within the sector. It is discounting the model entirely.”*

The GCC twist: captive centres run on the same economics as the sector being repriced around them. The squeeze is still ahead of them, not behind.

And the squeeze is still landing. Our own reporting notes that so far *“AI gains [have] failed to cool rising staff costs at Indian IT firms”* — the labour-substitution thesis isn’t showing up in the P&L yet ([Nov 6 digest](/issues/2026-11-06.html)). That cuts both ways: the cost savings vendors promised haven’t arrived, and neither has the full margin hit. Both are coming.

## The hiring shift: fewer generalists, more judgment

AI needs *fewer* developers for the same work — but total software work is exploding. Global git pushes rose **46% year on year**; India’s own went from 19.5M to 31.9M, on the world’s largest developer base (**24.6M** and growing) ([The Developer J-curve](/analysis/developer-j-curve.html)). The casualty is the entry-level rung: Stanford finds a **13% relative employment decline** for 22–25-year-olds in the most AI-exposed roles. The work that remains moves *up* — from typing code to domain judgment: what to build, which risks matter, how a regulated industry actually works.

Our own reporting already shows the pattern inside the centres: *“GCCs ramp up hiring for AI, data and cybersecurity… face a talent shortage that could cap the country’s AI ambitions”* ([Nov 6 digest](/issues/2026-11-06.html)). The firms are retooling at scale — TCS has put Claude in front of **50,000** employees, Wipro retrained **225,000** on GenAI, Cognizant committed **25,000+** to Claude delivery. The durable advantage isn’t generalist capacity; it’s vertical depth, because *“domain knowledge compounds; AI accelerates but does not replicate years of it”* ([The Demand Paradox](/analysis/it-services-paradox.html)).

## The reframe: from cost centre to capability centre

The GCCs that win stop selling saved hours and start owning outcomes. It’s already visible. Infosys is building an **AI-first GCC for Truist** in Hyderabad — an estimated $500M build adding 1,000 staff, described as *“a live, large-scale AI-first enterprise deployment”* ([Oct 20 digest](/issues/2026-10-20.html)). New revenue lines are appearing: LTIMindtree booked its first **$150M** of standalone AI revenue ([India AI Stack](/analysis/india-ai-stack.html)), TCS’s AI revenue crossed **10%** of its quarterly total ([Oct 13 digest](/issues/2026-10-13.html)), and Genpact now earns **50%+** of revenue on outcomes rather than hours ([Oct 6 digest](/issues/2026-10-06.html)). At the extreme, TCS is committing **$7.4B** to a one-gigawatt AI data-centre campus in Hyderabad — an explicit hedge as the core services business shrinks ([India AI Stack](/analysis/india-ai-stack.html)).

The frame worth carrying is the one from our India adoption work: **Save, Serve, Strengthen**. Cost-out (Save) is the *smallest* prize; the compounding value is reaching the previously unservable (Serve) and building institutional capability — fraud, compliance, decisions (Strengthen) ([India AI Adoption](/analysis/india-ai-adoption.html)). A cost centre optimises Save. A capability centre is built for Serve and Strengthen.

## The stakes

This isn’t a corporate reshuffle — it’s a national-stakes story. GCCs sit on top of the world’s largest developer base and a $55B export engine. If they stay cost centres defending headcount, AI deflation grinds them down. If they become genuine capability centres — product, AI R&D, decisions — India captures a bigger slice of a *larger* market: the WEF projects a net **+78M jobs by 2030**, with software developers among the fastest-growing roles ([The Developer J-curve](/analysis/developer-j-curve.html)). But that rebound depends on demand holding for human-centered, judgment-heavy work — exactly the work a capability centre is built to do.

## The honest counter-case

#### Three ways this read could be too neat

First, **“move up the value chain” is easier to write than to do.** Thousands of captive centres were set up precisely to execute, not to decide; rebuilding them around domain judgment is a decade-long cultural change, not a reorg. Second, **the arbitrage isn’t gone yet.** Staff costs are still rising faster than AI is cutting them, so for now the model limps on — which can breed complacency right up to the cliff. Third, **our scale data is thinner than the thesis.** Beyond the GCC count and the $55B export figure, India’s archive of hard GCC-specific outcomes is still shallow; treat the direction as well-evidenced and the magnitude as a live question.

## So what — for leaders running or building GCCs

#### Stress-test the arbitrage, not the headcount
Announced seats are a vanity metric. The number that decides your centre’s future is the **spread between what a client pays and what the work now costs with AI in the loop** — and that spread is narrowing. Model it explicitly.

#### Rotate the book toward outcomes
The highest-risk revenue is time-and-materials and staff augmentation, where “one AI-equipped developer replaces three augmented seats.” The lowest-risk is **outcome delivery and domain-specific consulting**. Shift the mix before the market forces it.

#### Buy domain depth, not generalist capacity
Vertical knowledge — BFSI, manufacturing, healthcare, regulation — is the one moat AI accelerates but can’t replicate. Hire and retain for it; it is where the pricing power moves next.

#### Run it as a capability centre
Reframe the mandate from cost-out to **product, AI R&D and decision-making**. The centres booking standalone AI revenue and building AI-first deployments are writing the template — the rest are defending a spread that won’t hold.

## The bottom line

India’s GCCs were an arbitrage machine: take the wage gap, bank the difference. Generative AI is **closing the gap from both ends** — deflating the work from the demand side and shrinking the ROI of automating cheap labour from the supply side — and the market has already started discounting the model that depends on it.

The arbitrage that built the GCC is ending. What can replace it — capability, domain depth, outcomes, AI R&D — is a **better business, for the firms and for India**. But only for the ones that move before the spread closes on them.

Related: [The Demand Paradox](/analysis/it-services-paradox.html) · [The India AI Stack](/analysis/india-ai-stack.html) · [India AI Adoption](/analysis/india-ai-adoption.html) · [The Developer J-curve](/analysis/developer-j-curve.html)
