India’s GCCs have the scale; their parent companies now want proof of value

India has 2,117 Global Capability Centres (GCCs). They employ 2.36 million people and booked $98.4 billion in revenue in the year to March 2026, according to a Nasscom-Zinnov report released in May. Two years earlier, revenue was $64.6 billion.

The parent companies that fund these centres are now asking a harder question of what have they changed in the business?

“Enterprises are no longer evaluating AI based on the number of pilots launched, but on the business outcomes delivered,” says Rishi Arora, Managing Partner, IBM Consulting India and South Asia.

The question has teeth. Revenue is already close to the 99billion-105 billion Nasscom had forecast for 2030, Reuters reported in May. Yet the average centre now employs about 1,115 people, down from about 1,130 in 2024, according to Zinnov data cited by Nikkei Asia. Zinnov also puts 55% of GCC work, mostly software testing and routine development, under pressure from AI.

More centres, flat teams, and work that machines can increasingly do. Scale and cost savings, the two arguments that built the industry, are harder to sustain on their own.

What success looks like

Arora says the companies getting the most from AI treat it as a transformation programme, not a technology rollout. They redesign workflows, decision-making and operating models around it, and they name who answers for the results.

He lists the measures that count productivity, faster cycle times, better customer experience, revenue growth and lower risk. Models deployed and proofs of concept completed count for less.

From delivery to ownership

The old GCC delivered functions at scale and cut costs. Arora says AI is speeding its move into what he calls “Global Value Centers”. These centres own product development, AI innovation, platform engineering and enterprise-wide transformation. Some now help build digital products and new revenue lines for their parents.

“Their value is increasingly measured by the business impact they create rather than the functions they support,” he says.

The Nasscom-Zinnov data points the same way. Of the GCCs set up since FY21, 96% launched with product or portfolio mandates from day one, the report found. Some 64% of site leaders also hold a global functional role. More than 1,200 centres have built AI and machine-learning capabilities.

A centre that builds products, shapes technology architecture or runs enterprise platforms sits differently in the parent company from one billed on the volume and price of its services.

Washington’s visa policy adds a complication. The US imposed a $100,000 fee on certain new H-1B petitions in September 2025. A federal court in Massachusetts struck it down in June, and on July 24 the First Circuit refused to let the government enforce it while its appeal proceeds.

Firms did add centres in the meantime. ANSR Research counted more than 60 new GCCs set up in India by US companies in 2025, against about 43 in 2024. Atul Vashistha, chairperson of advisory firm Neo Group, disputes that the visa fee is the cause. Either way, a visa rule can move work. It cannot make the centre that receives it important.

The harder question is talent. The Nasscom-Zinnov report counts about 250,000 AI professionals in the talent base. Whether that pool can grow fast enough to support product ownership is open. Without it, “value centre” becomes a new name on an old cost centre.

Models are no longer the hard part

Choosing a model is the easy bit now, Arora says. Companies need not commit to one model or platform. They can pick the model that fits each job and draw on technology from across the market. The difficulty is making those choices work inside fragmented systems, with data, security and governance requirements attached. IBM sells hybrid cloud and AI governance services, and Arora notes that hybrid architectures that bring enterprise data, governance and model choice together are becoming central to running AI at scale.

Agentic AI raises the stakes. As it moves from producing output to orchestrating tasks and workflows, accountability, security and human oversight matter more. A company that can deploy AI agents but cannot say who answers for what they do is not ready to run them. That means training staff and redesigning workflows so people and agents work together.

The real breakthrough, Arora says, is combining “open innovation, fit-for-purpose models, skilled talent and AI governance.”

Data across borders

An Indian GCC may build AI tools on data from one country, serve users in a second and plug into platforms hosted in a third. Each jurisdiction sets its own rules on regulation, security and governance. The task is to keep oversight consistent without choking the collaboration that makes global operations work.

Arora’s answer is architecture that allows global integration while the company keeps control of its data, models and AI workloads, with controls matched to data sensitivity, regulation and business needs.

“Equally important is embedding governance, security and digital sovereignty principles into the design of AI systems from the outset,” he notes.

In practice, a firm must know how its data is used, where it sits, where AI workloads run, who has access and how decisions are governed in each jurisdiction. Hybrid cloud and sovereign AI set-ups give firms more room to settle those questions, he adds.

Accountability after launch

Governance has to change too. Arora describes a shift from approval-based oversight to what he calls lifecycle accountability. For companies making multi-year AI investments, responsibility does not end at deployment. Reliability, security, explainability and business performance must be tracked for the life of the system. That takes a named owner for each AI system, defined controls over data, models and outputs, and continuous monitoring against business goals and risk.

“Boards need visibility into whether AI systems are operating as intended, whether controls remain effective, and whether the expected outcomes are actually being delivered,” he says.

Responsible AI, he adds, means being able to “demonstrate, with evidence” that systems stay trusted, accountable and effective.

India’s GCCs have proved they can deliver scale. What that scale has changed in the parent business is the evidence they now have to supply.