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.
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; BW). By our own read, GCCs already steer $55B of India’s IT exports (The India AI Stack). 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, the entire offshore model “requires the labour-arbitrage gap to persist.” Generative AI is compressing that spread from both ends.
A Western client pays $150–200 an hour for work a Bengaluru developer does at $25–35 (The Demand Paradox). 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). 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?).
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; our Oct 24 digest). 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). That cuts both ways: the cost savings vendors promised haven’t arrived, and neither has the full margin hit. Both are coming.
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). 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). 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).
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). New revenue lines are appearing: LTIMindtree booked its first $150M of standalone AI revenue (India AI Stack), TCS’s AI revenue crossed 10% of its quarterly total (Oct 13 digest), and Genpact now earns 50%+ of revenue on outcomes rather than hours (Oct 6 digest). 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).
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). A cost centre optimises Save. A capability centre is built for Serve and Strengthen.
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). But that rebound depends on demand holding for human-centered, judgment-heavy work — exactly the work a capability centre is built to do.
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.
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.
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.
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.
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.
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.