The short answer
An AI-native GCC is designed around a smaller, more senior team that owns outcomes rather than a large team that executes tasks. In practice that means hiring fewer people at a higher band, giving the centre genuine product ownership from the start, budgeting for compute and data infrastructure as a first-class line, and building governance for model risk before the first model ships.
What changed
Centres set up five years ago were sized by headcount: a backlog existed, and people were added until it cleared. That logic weakens when a large share of routine implementation work is assisted or automated. The constraint moves from volume of hands to quality of judgement.
India is well positioned for this — it is currently the largest hiring market globally for AI roles, and the density of people who have deployed models in production rather than only trained them has risen quickly. But the competition for those specific people is intense, and a centre positioned as an execution arm will not win them.
Five design choices
1. Size for ownership, not for throughput
A team of 30 people who own a product area will outperform 70 people who service tickets from elsewhere, and will cost less in total. Resist the instinct to justify the programme with headcount; justify it with scope.
2. Hire a higher band from the start
The first fifteen hires set your ceiling. In an AI-native centre, the ratio of senior to junior should be higher than in a traditional one — closer to 1:2 than 1:5. Budget accordingly rather than discovering it during hiring.
3. Treat data and compute as a real budget line
GPU access, data platform, labelling, evaluation infrastructure and observability are not overhead; they are the production line. Centres that under-budget here end up with expensive engineers waiting on capacity.
4. Build governance before the first model
Model risk, data lineage, evaluation standards, human review paths and incident response. Every one of these is cheap to establish at ten people and expensive at a hundred. Your regulators, your customers and your own risk function will all eventually ask.
5. Give the centre a decision right, not just a mandate
The clearest predictor of whether senior AI talent stays is whether they can ship without three approval layers in another time zone. Write down what the centre decides alone. If the list is empty, expect the hiring to be hard and the retention worse.
What this means for the business case
The savings story weakens and the capability story strengthens. A senior AI engineer in India costs meaningfully more than a mid-level software engineer, so cost-per-head comparisons look less dramatic. The case has to rest on access to a talent pool you cannot hire at home at any price, and on the throughput a small senior team produces.
That is a harder case to write and an easier one to defend two years later.