TCS Layoffs 2026: 23,460 Jobs Gone, And Why Indian IT’s Business Model Just Broke

Key takeaways

• TCS ended FY26 with 584,519 employees, down 23,460, against an announced cut of about 12,200. The real reduction was roughly double.
• The top four shed a net 9,100 in the June quarter against 22,622 added a year earlier. A swing of 31,722.
• FY26 sector revenue grew 6.1 per cent; headcount grew 2.3 per cent.
• AI agents went from finishing 2.5 per cent of real freelance projects to 16.1 per cent in eight months.
• The equation linking revenue to headcount has broken, and broken equations do not re-link.

For thirty years Indian IT ran on one equation. Revenue equals people multiplied by hours multiplied by rate. You won a deal, you hired. Campus offers, bench strength, the pyramid and the real estate in Whitefield were downstream of that line. In FY26 the multiplication sign stopped working.

How the equation breaks

Every step is in the filings, and step four is where it compounds: the base of the pyramid is not a cost the firms chose to cut, it is the layer AI reaches first. This reverses only if AI gets worse at entry-level work, or clients start paying for headcount again.

What happened at TCS in FY26

TCS closed March 2026 with 584,519 employees against 607,979. Nearly 31,000 left across the middle two quarters, but Q4 turned positive with 2,356 net additions and TCS says the restructuring is finished and it is on track to hire 40,000 freshers. Anyone calling this freefall is ignoring the last bar. Anyone calling it fine is ignoring the two before it.

This is not a TCS story

Demand did not change by 31,722 people worth in a year. The firms no longer need to add a body to add a rupee, so a recovery lifts revenue without putting those seats back. NASSCOM puts FY26 sector revenue near 315 billion dollars, up 6.1 per cent, on direct employment of about 6 million, up 2.3 per cent, and describes providers moving away from FTE delivery towards outcome-based, risk-sharing constructs. FTE delivery means billing per person, per hour. That is the equation, and the industry body is describing the walk away from it.

The natural experiment: HCLTech versus TCS

HCLTech grew revenue more than five times faster than headcount. Note the counter-fact on the left: it also cut margin guidance to 17 to 18 per cent from 18 to 19. Decoupling is a repricing, not free money, and part of the saving goes to the client. That is why the advantage is temporary.

Where the capital went instead

The firms are still spending, just not on people. TCS has 200 acres in Anakapalli plus a Pune site for the first India data centres serving OpenAI, and HCLTech has committed about Rs 14,257 crore in Odisha for roughly 5,000 direct jobs. That is about Rs 2.85 crore of capital per job against roughly Rs 2.5 lakh per job, a ratio of 114 to 1, under what the PLI scheme actually bought, and it is why this decade makes Indian IT a landlord for other people AI. The margin in that stack accrues to whoever owns the compute, so the Nvidia moat and the China chip fight matters more to an Indian engineer salary than any HR memo.

The cost moat is gone for a second reason. A ten-person team with AI subscriptions, no offices and no bench now bids on work that needed fifty people, quotes 40 against a large firm 100, and the large firm cannot match it because it is carrying the building. I ran an agency. When I sold it we had about 180 employees. The same work today needs a fraction of that, and that number has not finished falling.

The machine improves faster than the debate

The Remote Labor Index tests whether agents finish real freelance projects to a standard a paying client accepts: 240 real briefs across 23 domains. It does not say 84 per cent of work is safe, and two readings are not a curve. But it was 2.5, it is 16.1, and that direction is the input every CFO is planning against. It stalls only if the remaining work is qualitatively harder rather than merely next in line.

The freshers take the hit first

Wipro cut fresher guidance to 7,500 to 8,000 and about 200 recruits flagged onboarding deferred beyond seven months. A layoff removes someone who already has skills. A frozen entry rung removes the ladder, because the simple tickets, basic tests and first-pass code that trained a fresher are exactly what AI does most cheaply. A 3.5 lakh package compounding at 10 to 12 per cent for twenty years was the deal. Remove the reliable increment and it is a different deal.

What I think happens next

• AI margins are temporary. Once everyone uses AI well the saving is competed away to clients. The HCLTech guidance cut is the first data point.
• Revenue per employee is the only metric worth watching. Not headcount, not deal wins.
• The squeeze lands hardest in the middle. Large firms have balance sheets, AI-native shops have no cost base.
• Entry-level hiring does not return to FY22 levels, and not in the old shape. Nobody needs the base of the pyramid.

If you have just been let go, or your offer has gone quiet

Demanding the jobs back is the expensive mistake. But telling people to treat AI as leverage is not advice either, so here is the specific version, written for one reader: someone holding a separation letter from an Indian IT firm, or an offer letter with no joining date on it.

• Collect the paperwork before your badge stops working. Relieving letter, experience letter, the last three months of payslips and Form 16. Indian background verification runs on those four documents, and every one of them is harder to extract from outside the building than from inside it.
• Transfer the provident fund, do not withdraw it. Move the balance to the new employer through the EPFO portal. Withdrawal before five years of continuous service is taxable, and a transfer keeps the five-year clock running.
• If the offer is deferred, get a date in writing. Wipro guided freshers down to 7,500 to 8,000 and about 200 recruits reported waits beyond seven months. An email naming a month is the difference between a delay and a quiet cancellation, and it is what lets you accept something else without guessing.
• Aim at the part of the sector that is still hiring. The top four cut 9,100 in the June quarter, but HCLTech grew revenue 11.2 per cent, and NASSCOM still counts about six million people in the industry with AI revenue at 10 to 12 billion dollars. The hiring is in AI delivery, data engineering, cloud migration and the data centre buildout. It is not in the ticket queue you came from.
• Publish evidence, not a skills list. Three shipped things with your name on them, in public, beat a certification on an Indian resume screen in 2026. A reviewer can verify a repository or a working tool in about ninety seconds. Nobody verifies a course badge.
• Price yourself on output, not years served. Revenue per employee is the metric that survives this, so walk in with your own version of it: what you shipped, how long it took, and what it replaced.

The uncomfortable part is that the entry rung is not returning in its old shape, so no plan that starts with waiting works. The same repricing is running through capital as well as labour, which is the argument in why the Indian funding drought is a diversion, not a winter.

Read next

Follow this with what Indian IT is building instead of software, then how China is closing the Nvidia moat, the layer actually collecting the margin.

Sources

• TCS quarterly and annual filings, FY26
• NASSCOM Strategic Review 2026
• Quarterly filings: TCS, Infosys, HCLTech, Wipro
• Scale AI and Center for AI Safety, Remote Labor Index

Figures are as reported by the sources named above at the time of writing.