What you need to know
A Global Capability Centre is a wholly-owned offshore arm of a multinational — not an outsourcing vendor, but the company itself, operating an engineering site abroad. A bank headquartered in London runs one in Bengaluru. A retailer headquartered in Minneapolis runs one in Hyderabad. The staff are employees of the parent, working on the parent's systems.
The category has a reputation problem rooted in its history, when captives largely did maintenance and support work that headquarters did not want. That description is now roughly fifteen years out of date for the better centres, and the shift has accelerated sharply with AI. Industry reporting through 2026 indicates that roughly two in three new GCC roles now require AI, data science or intelligent automation skills, and that newly-established centres are expected to carry a meaningful share of AI and machine-learning headcount within their first eighteen months.
What has not changed is how badly most candidates search for these roles. They are rarely on the job boards where AI engineers look, they use titles that do not match what the market calls the work, and the strongest ones are filled through internal sourcing before they are advertised at all.
The numbers, and how to read them
Published figures for this sector vary considerably by source and by what each one counts, so it is worth laying them out alongside their caveats rather than picking the largest.
| Figure | Reported value (2026) | How to read it |
|---|---|---|
| GCCs operating in India | Over 2,100 | Counts centres, not their size; ranges from 30 people to 20,000 |
| Total GCC employment in India | More than 2.3 million | All functions, not only engineering |
| Projected employment by 2030 | Around 2.8 million (NASSCOM) | A projection, not a commitment |
| GCC hires, first half of 2026 | Approximately 228,000, up around 11% year on year | Gross hires across all roles and levels |
| New roles requiring AI or data skills | Roughly 64% of roles created in 2026 | "Requiring AI skills" is a broad definition |
| New mid-sized centres planned for 2026 | Around 120, creating close to 40,000 jobs | The highest-opportunity segment for candidates |
Do not add these figures together or treat them as directly comparable. They come from different surveys with different definitions — one counts gross hires across all functions, another projects AI-specific roles, another counts centres. Each is useful as an indication of scale and direction. None should appear in a cover letter as a precise fact, and a hiring manager who works in the sector will notice if you have quoted a headline number without understanding what it measures.
The one row worth acting on specifically is the last. A newly-announced centre is the best entry point in this whole channel: it hires heavily and quickly, it has no internal candidate pool to promote from, its processes are not yet rigid, and early employees get scope that would take years to earn at an established site. It is also the least contested, because most candidates only hear about a centre once it is large enough to be famous.
What the work actually is
AI roles in captives cluster into four fairly distinct shapes, and knowing which one you are interviewing for changes both your preparation and whether you should want it.
Platform and enablement
Building the internal infrastructure other teams use: model gateways, evaluation harnesses, prompt registries, guardrails, cost attribution, observability. This is the most common AI mandate in a large enterprise and the most under-rated by candidates. The scale is real, the constraints are severe, and the skills transfer anywhere. If you want to become genuinely good at production AI infrastructure, this is a better environment than most startups because the failure modes arrive faster and matter more.
Applied AI for a business line
Embedding models into a specific function — claims processing at an insurer, fraud detection at a bank, demand forecasting at a retailer, clinical documentation at a healthcare group. The work is closer to the business than most engineers are used to, the domain knowledge compounds, and the ceiling depends heavily on whether the business line treats you as a partner or a supplier.
Research or advanced engineering
The smallest category, present only in a minority of centres, and the one most worth targeting if you can. Genuine research mandates exist in Indian captives of technology companies and of a few large financial institutions. They are hard to find and hard to get, and the interview loop resembles a lab rather than an enterprise.
Internal productivity
Deploying coding assistants, document tooling and internal copilots across a large employee base. Often dismissed, sometimes the most interesting job in the building, because change management at the scale of thirty thousand employees is a genuinely hard problem that very few engineers have solved. It is also the mandate most likely to be measured on adoption rather than on engineering quality, which suits some people and frustrates others.
Ask this in the first conversation: "Who writes the roadmap for this team, and where do they sit?" The answer separates a centre with real ownership from one executing decisions made elsewhere, and it does so faster than any amount of research. A recruiter who cannot answer it is a signal in itself. Follow it with "what did this team ship in the last quarter that was its own idea?"
Why the hiring channel is different
This is the practical heart of the guide, because the channel is the reason capable engineers miss these roles entirely.
The roles are posted on the parent's global careers site, not on Indian job boards, and often not on the aggregators either. A machine-learning role at a Bengaluru centre of a European insurer appears on that insurer's corporate careers page, filed under a global job family, with the city as the only indication of where it is.
The titles are corporate rather than descriptive. What the market calls an AI engineer may be posted as "Lead Engineer, Data Science", "Specialist — Intelligent Automation", "AVP, Machine Learning Platform" or "Senior Consultant, Advanced Analytics". Searching for "AI engineer" will miss most of them. Search by skill terms and by location instead.
Sourcing dominates advertising. Large enterprises staff internal talent-acquisition teams whose job is to find people, and they work through professional networks and referrals well before a posting goes live. This is the single biggest structural difference from startup hiring, and it means being findable matters more than applying.
The process is longer and more structured. Expect four to six stages over several weeks, including a panel and often a business stakeholder who is not an engineer. This is not indecision; it is how large organisations hire. Plan your search timeline accordingly and do not read slowness as disinterest.
Build a list of thirty multinationals with a centre in your city, and check their corporate careers pages directly every fortnight, filtered by location. It is unglamorous and it works, because you are looking where almost nobody else is. Pair it with a public profile that internal recruiters can find when they search for your skills — the sourcing-first nature of this channel means discoverability is doing more work than applications.
GCC recruiters search for candidates before they post roles. Make sure they find you.
AI Tech Connect lists AI engineers, founders and researchers across India and the UK — and the people hiring browse it to find them. Founding Builder profiles are free while early spots remain.
Become a Verified Builder →What they screen for that startups do not
The technical bar is broadly comparable to a good product company. Three additional filters, however, catch out candidates who prepared only for startup interviews.
Asynchronous working. Your stakeholders and often your manager sit in another time zone. The organisation is testing whether you can write a clear design document, disagree in a comment thread without it becoming an escalation, and make progress on a decision that takes eighteen hours to get an answer. Candidates who describe their work entirely in terms of conversations and whiteboards read as risky here.
Regulated-context judgement. If the parent is a bank, an insurer or a healthcare group, most of the difficulty in an AI project is not the model. It is data residency, model risk governance, auditability, explainability and the approval process. You do not need to be an expert, but you need to show you know these constraints exist and treat them as engineering requirements rather than obstacles. A candidate who says "we would just fine-tune on the customer data" without pausing has failed the question.
Evidence of something shipped. Universal, but weighted heavily here, because enterprises have been burned by hires who could discuss architecture and had never operated anything. One system you built, deployed and kept running, with the specifics of what broke and what you changed, outperforms a long list of technologies. If you do not have that from work, build it in public — the approach in our guide to shipping a public agent and eval harness as proof of work is designed for exactly this gap.
Compensation and the honest trade-offs
Compensation is a genuine strength of this channel and the reason it is worth taking seriously even if the category sounds unexciting. Established captives of large multinationals typically pay at or above the local market for equivalent seniority, with structured bands, defined review cycles and benefits that early-stage startups cannot match. Equity is usually parent-company stock — liquid, real, and worth a fraction of what a startup grant might theoretically be worth and considerably more than what most startup grants actually become.
| Dimension | Established GCC | Funded AI startup |
|---|---|---|
| Cash compensation | Strong, banded, predictable | Variable; often below market at seed stage |
| Equity | Listed stock, liquid, modest upside | Illiquid options, wide outcome distribution |
| Scope per engineer | Narrower, deeper | Broader, shallower |
| Speed of shipping | Slower; more approval surface | Fast |
| Data and infrastructure access | Excellent — real scale, real users | Often limited early on |
| Job security | Higher, but centre mandates can be withdrawn | Lower, and visibly so |
| External visibility of your work | Usually low; publishing needs approval | Usually high |
That last row is the trade-off engineers underestimate. Work done inside an enterprise is often invisible outside it. You can spend three excellent years building an AI platform serving forty million customers and have nothing public to show for it, which makes your next move harder than it should be. The countermeasure is deliberate: maintain a public profile that describes the shape and scale of what you did without disclosing anything confidential, and keep one small thing of your own that you can point to. This is exactly the problem a Verified Builder profile is designed to solve, and it is worth setting up while you are employed rather than when you need it.
The version of this that applies from the UK
Two things make this directly relevant to a reader in London, Manchester or Edinburgh rather than Bengaluru.
First, a large share of Indian GCCs belong to UK-headquartered organisations — banks, insurers, retailers, media groups. The AI platform teams in those companies are frequently split, with architecture and product decisions in the UK and a substantial engineering contingent in India. If you are applying to the UK end, understanding how the split works, and being credible about collaborating across it, is a differentiator that most candidates do not bring. If you are the person who makes that split work well, you are unusually valuable to both sides.
Second, the model itself has generalised. The same captive structure now operates in Poland, Portugal, Spain and elsewhere, for the same reasons. The structural questions in this guide — who owns the roadmap, where the decisions are made, whether the mandate is durable — transfer without modification. For readers weighing a cross-border move in either direction, our guides on remote global roles for India and UK engineers and on visas and relocation cover the mechanics.
How to tell a good centre from a bad one
The variance within this category is enormous, and the diligence is the whole game. Signals that a centre is worth joining:
- Engineering leadership for your area is based locally, not visiting quarterly.
- The centre owns a product or platform end to end, with a name you can be told.
- Engineers there speak at conferences or contribute to open source, with approval.
- The centre has existed long enough to have promoted people internally to senior roles.
- The interview includes someone who will be your peer, not only managers and recruiters.
Signals to be cautious about:
- Nobody can name a system the centre owns outright.
- The role description is a list of technologies with no problem attached.
- Senior engineering titles all sit in the headquarters country.
- Headcount is described in terms of a target number rather than a mandate.
- The AI work is described entirely as "supporting" or "enabling" another team.
A centre where the AI mandate arrived in the last six months, has no named owner locally, and is being staffed against a headcount target. That is a cost-arbitrage decision wearing the language of capability, and mandates like it get withdrawn in the next planning cycle. The tell is that nobody can describe what the team will have shipped in a year.
A 30-day plan
If you decide this channel is worth your attention, the sequence below is what actually produces interviews.
Week one. Build the target list. Thirty multinationals with a presence in your city, with the corporate careers URL for each. Add the parent's sector, because sector determines the constraints you will be interviewed on. Set up alerts for new centre announcements in the business press.
Week two. Fix your discoverability. This channel sources more than it advertises, so a searchable public profile listing your actual skills and one or two things you have built is worth more than fifty applications. Make sure your professional network profile uses the vocabulary these recruiters search for rather than startup shorthand, and get a Verified Builder profile up so that people looking for AI engineers in India and the UK can find you without knowing your name. Our playbook for recruiter search covers the keyword mechanics.
Week three. Prepare the two things startup preparation misses. Write one page on a system you built, structured for an asynchronous reader: problem, constraints, what you chose, what broke, what you would change. And read enough about the regulatory context of two target sectors to ask an intelligent question about it.
Week four. Apply narrowly and reach out directly. Ten well-matched applications through the corporate sites beat a hundred scattergun ones. In parallel, find engineers already at your target centres and ask them the roadmap question. Most will answer, because almost nobody asks.
None of this is fast, and the process at the other end will not be fast either. What it is, is uncrowded. The startup channel in Indian and UK AI is intensely competitive because everyone can see it. This one is larger, pays comparably, and most of your competition does not know it is there. More career guides are collected in our tips section.