How can a GCC build specialised AI, data, cloud and software engineering capability?

A capability is not headcount with new job titles. A centre has a real AI, data, cloud or software capability when it can take a problem from idea to a system that runs reliably in production, and then keep improving that system. That depends on a few experienced people, the skills of the wider team, and the practices that hold the work together.

What is the starting position for most GCCs?

Many centres have already started. The Nasscom and Zinnov GCC Value Orbit report counts more than 1,200 GCCs in India with embedded AI and machine learning capabilities as of FY26, and says nearly half of the GCCs set up since FY2021 were built with AI as a core focus. It puts India’s AI talent base at about 250,000 professionals [1].

Demand is likely to stay ahead of supply. A Bain & Company study reported in March 2025 projected that AI job openings in India could reach 2.3 million by 2027, against a talent pool of about 1.2 million [2]. Projections like this are uncertain, but the direction is clear enough: hiring alone is unlikely to close the gap.

Adoption is also not the same as results. McKinsey’s 2026 State of AI survey found that 44% of respondents report scaling AI across their organisation, while 37% attribute any EBIT impact to it [3]. The distance between those two numbers is mostly about how work is built, run and owned, not about which models are used.

Where should you start?

Start with the mandate, not the technology. What will the centre own? A model that informs a business decision, a data platform used across regions, the cloud infrastructure behind a global product, or engineering for a specific product line? Each needs a different mix of people. Two GCCs in the same industry can have very different answers, and the right plan follows from yours.

Then pick one or two use cases that matter to the business and are small enough to reach production within a few months. Capability grows from shipped work, not from training plans alone.

What are the four levers?

1. Hire a few experienced people who have shipped this work

A small number of anchor hires sets the standard for everyone else. Look for people who have taken systems to production and run them: an ML engineer who has handled model monitoring and rollback, a data engineer who has built pipelines other teams depend on, a platform engineer who has managed cloud cost and security at scale. Their main job in the first year is as much reviewing, mentoring and setting practice as it is writing code.

2. Upskill the team you already have

Your existing engineers know your domain, your systems and your stakeholders, which new hires take months to learn. Many can move into adjacent roles with structured support: backend engineers into MLOps, data analysts into data engineering, testers into AI evaluation. Bain’s own commentary on its projection points the same way, calling for reskilling and upskilling of existing talent and a move beyond traditional hiring approaches [2]. Pair people with anchor hires on real projects and give them protected time.

3. Bring in senior leadership part-time if you need direction first

If the centre is still proving the case for AI or data work to headquarters, a permanent senior hire may be premature. A part-time (fractional) leader can choose the first use cases, set the architecture and governance, define the roles and the hiring bar, and hand over to a permanent leader once the work has shape.

4. Adopt practices that make the work hold up in production

• Define how you will evaluate a model or feature before you build it, with a test set the business agrees with.
• Version data, models and prompts, and make every release reversible.
• Monitor quality, drift, latency and cost in production, with a named owner for each system.
• Put guardrails and human review where decisions carry real risk.
• Treat security, privacy and access control as part of the design, not a final check.

Which roles anchor each capability?

How do you show progress to global leadership?

Report outcomes rather than activity. Useful measures include the number of use cases running in production (not in pilot), the time from idea to production, incidents and reliability, cost per unit of work, the share of work the centre owns end to end, and how many existing team members have moved into new roles. These numbers tell headquarters whether the centre is building capability it owns, which is usually what leadership wants to know.

How Viamagus select helps

We keep a bench of 10+ vetted people for each of ten roles, including AI infrastructure, data science, AI/ML, MLOps, full-stack and DevOps, trained in a 6–8 week AI-native programme and refreshed weekly. We also offer fractional AI leadership and AI services covering modernisation, guardrails and AI CoE setup. For one enterprise client with a stalled AI initiative, we hired eight engineers in four weeks; the proof of concept is now in production and the client is taking the team onto its payroll at month 13. Read the proof or see how it works.