70% of India’s GCCs are stuck in AI pilot mode, says report

India’s Global Capability Centres (GCCs) are moving into a new phase where their success will be measured less by the number of AI pilots they run and more by whether they can turn AI into measurable business outcomes, according to a new report by Dell Technologies and consulting firm Zinnov.

The report, titled “India GCCs 2030: From Capability Centers to Agentic Transformation Engines”, is based on surveys and interviews with more than 50 senior GCC leaders across BFSI, retail, manufacturing and software.

India currently has more than 2,100 GCCs employing 2.36 million people and generating $98.4 billion in revenue in FY26, the report said. About 70% of GCCs have a defined AI roadmap or charter, while more than 1,200 centres have built AI and machine learning capabilities.


But the report found that nearly 70% of GCCs remain stuck at the AI pilot stage, with many struggling to take successful experiments into sustained enterprise use.

The problem, according to the report, is less about a lack of AI ambition and more about the foundations needed to scale it. Fragmented data, legacy systems, unclear governance, immature security controls and talent models designed for a pre-AI workplace are holding back adoption.

“The most influential GCCs of 2030 will not be measured by the number of AI initiatives they launch, but by their ability to industrialise AI responsibly and at scale,” Manish Gupta, president and managing director, Dell Technologies India, said.

Also Read: 60% of organisations could fail at AI governance if they ignore data culture: Gartner

The report also found that the GCC maturity curve is shortening. About 27% of new GCCs now reach “Portfolio Hub” maturity within five years, compared with nearly a decade historically. AI mandates are also arriving earlier, pushing GCCs to build capabilities that were previously expected only at more advanced stages.

The shift is also changing what companies expect from their India centres. About 64% of GCC leaders now hold dual global mandates, running the India centre while also owning a global function. Meanwhile, 66% of GCC leaders rank top-line business impact as a high priority for their enterprise AI strategy.

“The next decade will be about ownership,” Sidhant Rastogi, president at Zinnov, said. As AI and agentic systems become embedded into enterprise workflows, GCCs will increasingly be expected to own products, platforms, markets and measurable business outcomes, he added.

The problem with AI pilots

The report said AI projects often struggle when they move from controlled experiments into production because real-world enterprise data is more complex, governance is added too late and standalone use cases can be difficult to integrate with existing platforms.

The economics can also change significantly at scale. Agentic workflows can consume between 10,000 and 500,000 tokens per workflow, compared with 1,000 to 2,000 tokens for a standard chat interaction, according to the report.

This means GCCs need to make decisions around compute, data, security, governance and AI economics before workloads reach production, rather than treating infrastructure as a later-stage consideration.

The report identifies four areas that will shape the next phase of GCCs: embedding AI into core business functions, planning AI architecture before production, taking ownership of products and markets, and redesigning the workforce around engineering, product and business problem-solving.

It estimates that 55% of routine GCC work is already exposed to AI-driven automation, while 60% of the workforce will require reskilling by 2030.

Also Read: Deliveries over deployment: IT firms must reassess AI integration as pricing pressures rise, says BCG

The report argues that this will require more than incremental AI training, with GCCs having to rethink how work itself is organised as AI takes over more repetitive tasks.

It also proposes a framework for deciding which AI workloads companies should own and which can be run through leased or managed infrastructure. Sensitive data, regulatory exposure, business-critical processes and predictable high usage may require greater infrastructure control, while lower-risk or experimental workloads could be better suited to flexible models.

For GCCs working with regulated or proprietary data, the report proposes a “Sovereign Sandbox” model that allows teams to experiment in a controlled environment before moving workloads into production.

The larger shift, the report argues, is from GCCs as centres that provide capabilities to centres that own technology, products and business outcomes.