Why the GCC wave still hasn't crested.

The first frame Xpheno offers, and the one most worth holding onto, is that the GCC narrative has changed without the headline writers catching up. For two decades the country was sold to global CFOs on dollar-per-hour math. That math still works, but it is no longer the reason a Fortune 500 plants its flag in Bangalore. The reason now is that the work that matters (design, architecture, the call that decides what ships) is increasingly being done in India regardless of where it is funded.

The shift is structural. Once a captive starts owning a product end-to-end, the relationship between HQ and the centre stops being delegation and starts being collaboration. That is a different ask of leadership, a different ask of compensation, and a different ask of how India is positioned in the next five-year plan.

Xpheno's read on the count is consistent with Nasscom and EY data: India now hosts roughly 1,700 Global Capability Centres, up from somewhere near 1,200 in 2020, with credible pipeline visibility into another 500-plus over the next five-to-seven years. Total employment in the sector sits around 1.9 million and is climbing on a meaningful slope, not a flat one.

What the count obscures is the composition shift. The new GCCs being set up are not back-office processing units. They are AI labs, cybersecurity centres, fraud-ops hubs, and platform engineering teams reporting directly to a CPO or CTO at HQ. The brochure of a GCC opening in 2026 reads more like a product-engineering job ad than a BPO RFP from 2008.

Every GCC at scale tells the same story in private: hiring a Software Engineer II is mechanical, hiring a Staff Engineer is a six-month project. India produces 1.5 million engineering graduates a year, so there is no shortage of people who can ship features. The bottleneck is the band that can architect a system across three teams, decide what doesn't get built, and survive a HQ design review on equal terms.

That cohort exists, but its depth is shallow. The pyramid in most Indian captives is wide at the base, narrow at the top, and the narrow part is where every senior leader spends two-thirds of their hiring time. It is also why the compensation curve looks the way it does.

Xpheno's data, and most recruiters in Bangalore will say the same thing off the record, is that compensation inflation in the Staff / Principal / Engineering Manager band has run materially ahead of mid-level inflation for three consecutive cycles. A Principal Engineer offer in 2026 is structurally different from the same offer in 2019, and not only because of base salary. RSU components have grown, joining bonuses have re-emerged, and counter-offers from HQ are now standard.

The implication for the GCC business model is that the cost arbitrage at the top of the pyramid is narrowing. It is still there at the base; it is closing fast at the apex. Captives that built their unit economics on a 5-7x cost ratio are quietly running 3-4x at the senior bands.

The senior IC scarcity has a partial solution that almost no GCC head modelled in 2020: the diaspora is coming home. Engineers who left India between 1998 and 2015 are returning in volumes that haven't been seen since the dot-com bust. The driver is not patriotism. It is H-1B insecurity, cost-of-living arithmetic in San Francisco and Seattle, schooling decisions for teenage children, and, quietly, the realisation that the most interesting Indian product roles are now in India.

This is structural supply, not a blip. Once a senior IC repatriates and finds the role meaningful, the next one is easier; they refer their cohort. Xpheno's anecdotal pattern is that successful returns cluster. One ex-Google staff engineer at Walmart Global Tech in Bangalore brings three more inside eighteen months.

The conflation that still happens in policy circles, sometimes deliberately, is between captive Global Capability Centres and Indian IT services firms like TCS, Infosys, Wipro, and HCL. They share infrastructure, talent pool, and tax addresses, but the proposition is structurally different. An IT services firm sells managed engagements to many customers. A captive serves exactly one customer: its parent. The talent does not rotate off after eighteen months. The roadmap is internal. The career path runs through the parent's global hierarchy, not through a service-line P&L.

The consequence is that an engineer in a GCC is being evaluated against the same bar as their HQ counterpart. The work is comparable. The expectation is comparable. The career is comparable, at least until the senior bands, where the geography starts to matter again.

One of the quieter shifts Xpheno tracks is that India's 1,700 GCCs have become a meaningful B2B customer base in their own right: for security tools, observability stacks, HR-tech, talent intelligence, and developer productivity software. These centres run global procurement processes, but they have local budget authority and they have problems specific to a 5,000-engineer Indian site that a global vendor often does not solve well. Indian SaaS founders have noticed.

This is a category that did not exist in any structured way in 2015. By 2025 it is large enough that several Indian SaaS companies report GCC customers as a distinct cohort with its own sales motion. "Selling to GCCs" is now a go-to-market description, not a coincidence.

Bangalore, in Karnataka, remains the gravity well, but the share is no longer monopoly. Hyderabad has spent a decade building infrastructure, single-window clearances, and explicit GCC policy under successive Telangana administrations and is winning real share. Microsoft, Amazon, Google, Apple and Goldman all have Hyderabad anchors. Tamil Nadu has begun an explicit GCC policy framework. Pune holds an automotive-engineering and BFSI tilt. NCR has financial services and consulting density.

What this means in practice is that a Fortune 500 setting up its first Indian centre in 2026 is not making a city decision, it is making a between-state-incentives decision: land, power, water, transit, the size of the dual-income talent pool, and the speed of the state's clearances. Cities now market themselves the way industrial parks once did.

A second-order trend visible in any of the GCC cluster cities: hyperscaler and colocation data centre capacity is being built where the captive engineering teams already are. In Bangalore, Hyderabad, Chennai and Mumbai the data-centre buildout is following talent density rather than only cable landing stations and power availability. It is no longer a coincidence that AWS, Azure, and Google Cloud are adding region capacity in the same cities the GCCs already populate.

The compounding effect is real. A captive that builds workloads in-region against in-region data-centre capacity has lower latency, simpler data-residency posture under DPDP, and a recruiting story for infra engineers. The trio of captive, hyperscaler region and talent pool reinforces each city as a destination.

The newest GCC archetypes Xpheno is staffing into are dedicated cybersecurity operations centres and fraud-ops hubs for global banks and payment networks. The reasons are unromantic: these are 24/7 functions where the time-zone overlap with US and EU operations is a feature, not a bug; the talent supply has caught up; and the work is high-judgment but well-defined, a good fit for senior Indian engineers who don't want to do six-month research projects.

A new cybersecurity GCC in 2026 looks different from a classic engineering captive. It is more operations-heavy, more compliance-heavy, runs 24x7 shift patterns, and rarely scales past 800-1200 people. But there are dozens being announced, and they are pulling a distinct talent profile (security analysts, threat intel, IR) into the GCC pipeline for the first time at this volume.

The most visible signal that the GCC category has crossed an inflection is the wave of explicit AI labs being set up in Bangalore and Hyderabad since 2023: Microsoft AI, Google's India research presence, NVIDIA's expanded footprint, dedicated ML engineering teams from every large US enterprise. These are small teams. They are also the most expensive teams the GCCs run, and they are not back-office anything.

An AI lab GCC of fifty senior ML engineers is, in compensation terms, equivalent to a 1,500-person traditional captive. It is also a different recruiting problem: the bench Xpheno is hiring from is global, the offers compete with OpenAI and Anthropic in San Francisco, and the work has to be visibly real or the engineers leave.

Xpheno's read on the boardroom decision matters: a CFO will still run the cost-ratio analysis before approving a new centre, and the answer still works. But the conversation that gets a captive opened in 2026 is no longer led by cost. It is led by talent availability, time-zone coverage, product ownership ambition and, increasingly, the parent company's view that the future of the function will be Indian-led regardless.

The narrative shift is consequential. When cost is the headline, the captive lives on a margin treadmill. When strategy is the headline, the captive earns budget like any other business unit and starts to act like one.

Attrition in Indian tech peaked uglily in 2021-22, when 25-30% wasn't unusual at the mid-band, and the war for talent that produced was global. It has cooled significantly. Xpheno's view is that retention is now more about the quality of the work and the legibility of the career path than about counter-offers. A senior IC who has come back from Seattle to lead a real product team in Bangalore does not leave for a 15% raise. They leave when the work stops being real.

That is a different management problem. The HR playbooks from 2021 (retention bonuses, accelerated vesting) don't move the needle the same way. What moves it is whether the captive's leadership has actual authority over the roadmap or is still a glorified delegation node.

The senior leaders running India captives today, the country GMs and site heads at Walmart Global Tech, JP Morgan, Goldman Bangalore, Microsoft IDC and Google India, are a small population. Most are first-generation in the role. Most do not have an obvious successor inside the GCC; the bench beneath them is strong on engineering but rarely on the political and commercial fluency required to manage a parent company across a twelve-hour time zone gap.

This is the succession problem Xpheno watches at the top of the senior bench. It is not solved by hiring more Staff Engineers. It is solved by deliberately growing the GM-track inside captives: rotations into HQ, P&L responsibility earlier, deliberate exposure to enterprise sales and finance. Almost no captive does this systematically yet.

A quieter point that doesn't make most GCC-cheerleading articles: the original tax and infrastructure incentives (STPI in the 1990s, SEZ from 2005, the various state-level industrial parks) are still load-bearing for the unit economics of mid-tier captives. The headline tax holidays have largely sunset, but the cluster effects they produced (concentrated talent, shared infrastructure, single-window clearance) are what make a new GCC viable in 2026.

The implication is that policy continuity matters more than policy excitement. The DPDP regime, the GIFT City framework, state GCC policies: what matters is whether they are stable for a decade, not whether they are flashy in year one. A captive's economic model is built on five-year-rolling assumptions; volatility in any of them kills the bet.