Read the number twice, because it looks like a typo.
Across the first nine months of India’s 2026 financial year — April through December of last year — the country’s largest IT services companies grew their combined headcount by a net of 17 employees.
Seventeen. Not seventeen thousand. Over the same nine months a year earlier, the top five firms had added 17,764.
These are the companies that turned Bangalore, Hyderabad, Pune and Chennai into a destination. For three decades they ran the largest, most reliable graduate-absorption machine in the world: campus recruitment by the tens of thousands, a training campus, a bench, a badge, a way out. That machine did not shrink last year. It idled.
Where the Seventeen Came From
The aggregate hides a violent split. Three of the big four kept hiring at close to normal volume. One went the other way hard enough to cancel them out.
| Company | Net headcount change, first 9 months of FY26 |
|---|---|
| Infosys | +13,456 — the highest in the sector |
| Wipro | +8,675 |
| HCLTech | +2,595 |
| TCS | −25,816 — the entire drag on the sector |
| Sector net | +17, against +17,764 in the same period a year earlier |
TCS had told the market this was coming. In July of last year, after posting one of its weakest first quarters since the pandemic — constant-currency revenue down 3.1%, dollar revenue down 1.1% — the company said it would cut about 2% of its global workforce, roughly 12,200 roles out of 613,000, over the financial year. It called the programme a “future-ready transformation.” The cuts were aimed mainly at middle and senior management, and the stated reasons were AI, shifting market priorities and tighter client technology budgets.
The reaction in India was not the usual shrug that greets a corporate restructuring. The state of Karnataka formally sought an explanation from the company over the layoffs. The federal IT ministry said it was watching. When the single largest private employer in a $283 billion export sector announces it is getting smaller on purpose, that stops being a company story.
The Thing That Broke Is Older Than AI
The business model under all of this has a name: labor arbitrage. Find work that has to be done, find somewhere it can be done for less, move it there, keep the difference.
It built modern India’s services export economy. It built the Philippine call centre industry. It is the reason your bank’s overnight support desk answers in Manila and your insurance claim is keyed in Noida. And for thirty years every threat to it was really just a competitor: if India got expensive, there was Vietnam; if Manila got expensive, there was Cebu. The work never disappeared. It moved.
Generative AI is the first pressure on that model that does not have a next destination. A ticket that gets closed by a model was not relocated to a cheaper country. It was removed from the pile.
Manila Just Wrote It Down
The Philippines made the same admission in July, in the most institutional way available: it revised its own forecasts.
The IT and Business Process Association of the Philippines (IBPAP) completed a midpoint review of its industry roadmap and cut its 2028 targets. The revenue goal set in 2022 was $59 billion. The best case is now $50.5 billion. The employment goal was 2.5 million workers; the best case is now 2.14 million, and the downside case is 1.85 million. IBPAP described the revision as a more honest assessment of where AI adoption, competitive pressure and changing buyer behaviour were taking the sector.
Trade groups do not lower their own headline numbers casually. That is the number the sector quotes to investors, to Congress, and to every graduate deciding whether a headset is a career.
Then the World Bank put a floor under the diagnosis. Its World Development Report 2026 named the Philippines among the economies most exposed to generative AI on earth, and it was specific about which tasks: customer support, transcription, document processing, data entry. Those are not incidental functions of the Philippine outsourcing sector. They are close to a description of it — a sector generating roughly $38 billion a year and directly employing about 1.82 million people.
The report also carried the single most alarming figure in this whole story, and it is not a forecast. Outsourced work flowing to developing economies fell 39% in 2025. Not projected to fall. Fell.
The Second Front: Washington
AI is not the only thing pressing on Manila, and the other thing is deliberate.
Two bills sit in the US Congress. The HIRE Act — Halting International Relocation of Employment — was introduced in the Senate as S.2976 in October 2025. It proposes a 25% excise tax on outsourcing payments made by American companies to foreign service providers, and would strip the tax deductibility of those payments on top. Alongside it runs the Keep Call Centers in America Act of 2025.
Neither has moved. As of early this year the HIRE Act was stalled without bipartisan support, and its practical effect so far has been psychological rather than fiscal: buyers hedging toward staff-leasing arrangements instead of transactional outsourcing contracts, to avoid a compliance cost that does not exist yet. Manila is not waiting to find out. Ambassador Jose Manuel Romualdez has said the Philippines is “working with lawmakers and industry stakeholders to seek exemptions,” and IBPAP has been lobbying US congressional offices directly alongside the trade and foreign affairs departments.
The two pressures point the same way but work differently. A tax makes offshore delivery more expensive than onshore delivery. AI makes the delivery unnecessary at either address. Only one of those can be lobbied.
What This Is Not
It is not a collapse, and anyone selling it as one is ahead of the evidence.
IBPAP still expects the Philippine sector to close 2026 at $42.3 billion in revenue and 1.96 million employees, both up from $40 billion the year before. Both India and the Philippines added outsourcing jobs in 2025. Philippine unemployment fell from 9% in 2021 to roughly 4% in March of this year. The World Bank’s own global framing is genuinely mixed: it puts 4.5% of jobs in low- and middle-income countries at risk of automation, while 16.2% stand to see meaningful productivity gains.
What changed is not the level. It is the ceiling, and which end of the ladder is load-bearing.
The exposure is concentrated in exactly the work a new hire is given on day one: password resets, order-status checks, appointment scheduling, first-line troubleshooting, keying documents. The work that has held up is the work that requires judgement — escalations, complex resolution, quality assurance, regulated healthcare and finance processes. That is a defensible position for an experienced agent and a closed door for a graduate, because the entry-level tier was never really a job description. It was an apprenticeship that happened to be billable.
Remove it, and in five years there is no one qualified for the tier above.
IBPAP’s answer is to try to skip the problem: it has set a target of two million AI-enabled digital Filipino workers by 2028. Whether a call centre workforce can be retrained into the roles that survive automation, at that scale, on that timeline, is the entire bet the Philippine economy is now making. Nobody has done it before.
What To Watch
- Whether TCS’s number is the outlier or the leading indicator. Infosys added 13,456 people in the same window TCS shed 25,816. One of those two companies is reading the next three years correctly. Their FY27 hiring guidance will say which.
- Campus recruitment volumes, not total headcount. Net headcount hides everything. The number that describes what actually happened to a generation is how many fresh graduates were hired — and it is the number the industry reports least prominently.
- Whether the 39% decline repeats. A single year of outsourced work falling off could be a demand cycle. Two consecutive years is a structural break, and it would arrive before any of the 2028 forecasts are due.
- The HIRE Act’s co-sponsor list. It is stalled, not dead. In an American election year, a bill that taxes sending jobs abroad is the kind of thing that finds bipartisan support quickly.
- Whether “AI-enabled worker” gets a definition. Two million by 2028 is a headline until someone specifies what the training is, who pays for it, and what the retrained worker is actually hired to do.
The Takeaway
Most of the AI jobs debate is a forecast. This one is a filing. India’s largest IT services firms added a net 17 employees over nine months against 17,764 the year before, TCS shed 25,816 roles while calling it a transformation, the Philippine industry association cut its own 2028 revenue target by $8.5 billion and its jobs target by 360,000, and the World Bank recorded outsourced work to developing economies falling 39% in a single year. None of that is a projection about 2030.
The story is not that these countries are losing to a cheaper competitor. There isn’t one. For thirty years the answer to expensive labor was always cheaper labor somewhere else, and the work kept arriving somewhere. AI is the first answer that does not require the work to arrive anywhere at all.
The sector is still growing. It will probably still be growing in 2028. But it is growing into a smaller ceiling than the one it planned for, and it is doing it by holding on to the people who already know the job while quietly closing the door that let them in. That is a stable arrangement for about a decade, and then it isn’t.