For thirty years, Indian IT ran on an equation so simple it could fit on a napkin: Revenue equals People times Hours times Rate. You won a deal, you hired two hundred graduates. You won a bigger deal, you built a campus. The entire ecosystem — the campus placement drives, the bench strength, the Whitefield skyscrapers, the bus fleets crossing Bangalore at dawn — all of it was downstream of that multiplication. Add a body, add a rupee. The equation was the industry.
In FY26, the multiplication sign stopped working.
NASSCOM, the industry’s own trade body, now describes providers “moving away from FTE delivery towards outcome-based, risk-sharing constructs.” FTE — full-time equivalent — is billing per person per hour. That is the equation. And the industry body isn’t quietly retiring it. They’re describing the walk away from it in the language of strategic transformation, which is what you do when you don’t want to say the words “we hired too many people for a business model that no longer needs them.”
Here is what that walk looks like in the numbers.
The Headcount That Vanished
TCS closed FY26 with 584,519 employees. That is down 23,460 from 607,979. The announced cut was roughly 12,200. The real reduction was nearly double what they told the press. You can argue about attribution — attrition, performance, restructuring — but 23,460 people who had TCS badges no longer do. That’s a stadium.
Across the top four Indian IT firms — TCS, Infosys, HCLTech, and Wipro — the net headcount change in the June quarter was minus 9,100. The same quarter the previous year, they added 22,622. That is a swing of 31,722 people in twelve months. Add Tech Mahindra to make it the top five, and the combined net reduction for FY26 was 7,389 employees, reversing a net addition of 12,718 in FY25. The sign flipped.
And then there are the experienced ones. Over 7,700 professionals with more than fifteen years of experience lost their jobs across TCS, Wipro, Infosys, HCL, and Cognizant. These are the people who built the delivery machines. They know where the bodies are buried, sometimes literally, in the codebases. They are not freshers who couldn’t clear a technical round. They are the people who ran the floor.
The Silent Ones
The announced numbers are not the real numbers. They never are.
TeamLease estimated that 10,000 to 15,000 people had already lost jobs through “silent layoffs” by May 2026 — and projected 25,000 to 35,000 total job losses for the full year. CIEL HR put the figure at roughly 12,000 eliminated so far, expecting 18,000 to 21,000 total. Combined across 2025 and 2026, the toll could reach 43,000.
These don’t come with press releases. They happen through what the industry calls “silent exits linked to performance reviews and skill relevance.” That phrase deserves unpacking. “Performance review” means your boss put you on a plan that was designed for you to fail. “Skill relevance” means the thing you were hired to do is now done by a model that doesn’t take coffee breaks. And “silent exit” means you sign something, take whatever severance is offered, and leave without making noise, because the industry is small and references matter.
Here is the critical distinction: 2025 was about correcting pandemic overhiring. Firms had stocked up like it was a famine coming. That correction was ugly but comprehensible — they’d overshot, now they were recalibrating. 2026 is different. 2026 is about reshaping workforce structures. The demand hasn’t collapsed. The revenue is still growing. What changed is that firms no longer need the same number of humans to service the same amount of revenue. This is not a cyclical layoff. It is a structural repricing.
Satish Viswanathan, a former MD at Accenture, put it precisely: “The AI era is breaking the old workforce equation in IT and consulting. This does not mean people are irrelevant; it means that the basis of workforce value is being redefined.” Read that again. The basis of workforce value is being redefined. That is a polite way of saying: what you were worth is not what you are worth, and the market is figuring out the new number faster than you can reskill.
The Salary That Never Moved
If you want to understand what an industry actually values, don’t read its annual report. Read its payslips.
Entry-level salaries at TCS and Infosys were Rs 3 to 3.4 lakh per annum in 2007-2008. In 2026, they are Rs 3 to 3.5 lakh per annum. The same. For almost twenty years, the number has not moved. Xpheno’s data traces the arc: in 2013-14, fresher packages were Rs 3 to 3.4 LPA. By 2023-24, they had risen to Rs 3.8 to 4.5 LPA. Over a full decade, the total growth was just over Rs 1 lakh per annum. That is not a career trajectory. That is a rounding error.
Wipro’s fresher range is Rs 1.5 to 5.1 LPA, with the “traditional” package sitting at Rs 3 to 4 LPA — a band the company itself describes as “stagnant for a decade.” These are the words of the employer.
Now turn the paper over and look at the top.
TCS CEO compensation went from Rs 1.31 crore in FY14 to roughly Rs 25 crore. Infosys: from Rs 16 lakh under Shibulal to Rs 66 crore under Parekh. HCL: from Rs 4.22 crore to Rs 84.16 crore. Wipro: from Rs 6.57 crore to approximately Rs 50 crore. Deloitte India’s survey found average CEO compensation at Rs 13.8 crore in 2024 — up 40 per cent from pre-COVID levels. Every second CEO had target compensation above Rs 10 crore in 2024, compared to every third CEO in 2020.
You can draw the graph yourself. A flat line for the bottom. A rocket for the top. The gap between the person writing the code and the person signing the offer letter has gone from a hill to a canyon, and the canyon is still widening.
And here is where it gets structural rather than merely unfair: the industry says it cannot find “deployable” talent. India produces more than 1.5 million engineering graduates annually, but only a small proportion are considered ready for work. The complaint is real — engineering education in India is uneven, and many graduates need substantial training before they can bill. But notice what the industry has done with that complaint. It has used it to justify keeping salaries frozen while simultaneously cutting the training programmes that were supposed to bridge the gap. The problem of unemployable graduates becomes the justification for not employing anyone.
Meanwhile, the same firms complain about talent shortages in AI and ML. Neeti Sharma, CEO of TeamLease Digital, lays out the actual market: “Specialised digital skills command 30 to 60 per cent higher pay than standard entry-level roles. Entry-level AI/ML roles typically range between Rs 6 to 12 lakh, while niche generative AI roles can exceed Rs 22 lakh.” So the money exists. The industry can pay Rs 22 lakh when it wants to. It simply does not want to — not for the person maintaining the Java monolith, not for the tester writing manual cases, not for the developer whose work AI can now draft in seconds.
The Ladder Without a Bottom Rung
There is a difference between a layoff and a frozen entry rung. A layoff removes someone who already has skills. A frozen entry rung removes the ladder.
Wipro cut its fresher hiring guidance to 7,500-8,000. Roughly 200 recruits reported onboarding deferred beyond seven months. These are people who were offered jobs, who stopped looking, who told their families they were set — and then waited. And waited. Some are still waiting.
Think about what a fresher used to do in Indian IT. They picked up simple support tickets. They ran basic test cases. They wrote first-pass code that a senior would review. They monitored dashboards. They sat in rooms with fifty other freshers and learned how an enterprise codebase worked, how a client call ran, how a production deployment happened at 2 AM. None of this was glamorous. All of it was the on-ramp.
That is the work AI handles first.
The simple tickets, the basic tests, the boilerplate code — this is precisely the territory where large language models are most capable and most cost-effective. You don’t need a Rs 3.5 lakh fresher to write a unit test that a model can generate in four seconds. You don’t need a junior developer to produce a first draft of a CRUD API when the model can scaffold it before your standup ends. And so the bottom rung doesn’t just shrink. It gets sawed off.
Which means the mid-level engineer of 2028 — the person who would have entered as a fresher in 2026 — never enters. The pipeline is broken not at the output end but at the input end. The industry is not producing fewer engineers. It is producing fewer engineers who know how to be engineers.
The 10-Person Bid
Here is where the abstraction becomes concrete.
The Remote Labor Index — a tracking project run by Scale AI and the Center for AI Safety — measures what AI agents can actually do against real freelance briefs. Not benchmarks. Not demos. Two hundred and forty real projects across twenty-three domains. In eight months, AI agents went from completing 2.5 per cent of these projects to client standard, to 16.1 per cent. That is not a curve. That is a wall.
And “client standard” does not mean a lab test or a toy benchmark. It means the work was delivered to a real paying client who reviewed it, accepted it, and paid for it — the same standard a human freelancer is held to. So when you read 16.1 per cent, read it correctly: nearly one in six real freelance jobs can now be completed by an AI agent to a paying customer’s satisfaction. That is not a theoretical projection about future capability. It is a measurement of what already happened, on real briefs, with real money, in domains ranging from software development to data analysis to content writing. The terrifying part is not the trajectory. It is that the number is already here.
Now translate that into competitive dynamics. A ten-person team with AI subscriptions, no offices, no bench, no HR department processing two thousand payslips a month, can now bid on work that previously needed fifty people. They quote 40 against a large firm’s 100. Same deliverable. Half the cost. No campus recruitment drive required.
The large firm cannot cut its price to 40 without admitting that its cost structure — the campuses, the bench, the middle management, the bus fleets — is no longer a competitive advantage. It is a liability. So the large firm loses the bid, or wins it at a margin that makes the win feel like a loss, and then goes looking for the next line item to cut. The line item is usually a person.
The Decoupling
India’s IT services industry is worth more than $315 billion and employs roughly six million people directly. In FY26, sector revenue grew 6.1 per cent. Headcount grew 2.3 per cent. Those two numbers used to move together — they were the same equation, after all. Add revenue, add people. Now they are pulling apart. Revenue is going one way. Headcount is going another. The equation has decoupled.
HCLTech grew revenue five times faster than headcount — and also cut its margin guidance from 18-19 per cent to 17-18 per cent. Read that carefully. Decoupling does not mean free money. It means the work is being done differently, the pricing is under pressure, and the margin profile of the new model is thinner than the old one. The industry is repricing itself. It is repricing its work, its margins, and its people.
The Landlord Problem
There is a counter-narrative, and it deserves to be taken seriously. Global Capability Centres are growing. New AI-focused roles exist. The industry is not contracting in revenue terms — it is growing, and some of that growth is in genuinely new categories of work.
But look at where the capital is going.
TCS is investing in 200 acres in Anakapalli and Pune for India data centres serving OpenAI. HCLTech committed Rs 14,257 crore in Odisha for approximately 5,000 direct jobs. That is Rs 2.85 crore of capital per job. The historical ratio — the cost of creating one IT job in the old model — was roughly Rs 2.5 lakh per job. The new ratio is 114 times higher.
This is not the same industry. The old industry deployed capital to hire people who wrote code. The new industry deploys capital to build infrastructure for other people’s AI. Indian IT is becoming a landlord for other people’s intelligence. And in a landlord business, the margin accrues to whoever owns the building — the compute, the models, the platform. Not to the person who manages the building. Not to the person who sweeps the server halls. The structural position has changed, and with it, the bargaining power of everyone who works inside it.
What Survival Costs
The headline says the industry is fine. Revenue is growing. Firms are profitable. The stock market is sanguine. NASSCOM talks about transformation and value-added services. By every metric that matters to an analyst, Indian IT is surviving.
But survival is not free, and the cost is not distributed evenly.
The cost is paid by the 23,460 people who left TCS, only half of whom were in the announcement. It is paid by the 7,700 senior professionals who thought fifteen years of institutional knowledge would be a moat, and discovered the moat was six inches deep. It is paid by the fresher whose onboarding has been deferred for seven months and counting, who watches batchmates in other industries start their careers while they wait for a letter that may not come. It is paid by the engineer who hasn’t had a meaningful raise in three years and hears the CEO talk about AI investment on an earnings call where executive compensation is up 40 per cent since COVID. It is paid by the mid-level manager who is told to do more with less, which is always a euphemism for doing the same amount with fewer people and no additional compensation.
And it is paid — quietly, structurally, across years — by an entire generation of engineers who were told that the ladder existed, that the entry-level salary was a floor and not a ceiling, that the company would invest in them, and that if they worked hard, the equation would work. The equation was supposed to work for everyone. That was the promise of the industry. You join, you learn, you grow, the company grows, everyone rises.
The equation still works. It just doesn’t need you anymore.
That is what survival costs when the headline says things are fine. It costs the assumption that fine applies to you. It costs the belief that your labour is the thing being valued, when the thing being valued is increasingly the model, the compute, and the outcome — and the people are the variable that the equation is learning to do without.
What Comes Next
So what would an actual response look like? Not the generic “reskilling” hand-waving that every panel discussion produces — real, specific policy interventions. Wage subsidies for workers displaced by AI-driven restructuring, modeled on what European countries did during deindustrialisation, not the trickle of VET programmes that pay lip service to transition. Mandatory AI-impact disclosures: if a firm is shedding headcount while deploying AI systems that replace that headcount, it should say so publicly, and the disclosure should include numbers, not narratives. Genuine reskilling programmes — meaning programmes with income support during retraining, not the expectation that a 35-year-old engineer with a family will somehow self-fund a pivot to machine learning operations on weekends.
The GCC growth is real, and it matters. But honesty requires acknowledging the math. For every traditional IT services worker displaced, the new AI economy creates perhaps one AI infrastructure or engineering role — and that role frequently requires skills the displaced worker doesn’t have and can’t acquire in a three-month bootcamp. The capital-per-job ratio has gone from Rs 2.5 lakh to Rs 2.85 crore. You don’t close that gap with good intentions. The GCC boom absorbs a slice of the top of the pyramid while the bottom falls through. The numbers do not balance, and pretending they do is the new version of the old promise that the ladder existed for everyone.
Then there is the legal vacuum. India’s labour laws were written for factories — physical workplaces, shift bells, assembly lines. The framework for “silent layoffs via performance reviews” does not exist because the category did not exist when the laws were written. A company can shed thousands of knowledge workers through rigged performance plans, deferred onboarding, and “voluntary” separations, and none of it triggers the protections that a factory closure would. NITES and other worker organisations have been pushing this fight, but they are operating inside a legal architecture designed for a different century. The law needs to catch up to the layoff. Specifically: performance-managed exits at scale should trigger the same disclosure and consultation requirements as retrenchment. If it walks like a layoff and talks like a layoff, it should be regulated like one.
None of this is false optimism. The numbers in this essay are real, and they are grim, and no policy response will make them un-happen. But India’s IT workforce is not a passive population waiting to be priced out. They organised. They unionised. NITES filed petitions. Workers shared their stories even when the industry preferred silence. The question was never whether this workforce can adapt — they have been adapting every three years as the technology stack changed beneath them. The question is whether anyone with power will make the adaptation survivable instead of just survivable-sounding. Whether the government will treat this as the labour crisis it is, rather than an innovation story with awkward footnotes. Whether the firms will be honest about what they are doing. Whether the six million people inside this industry — and the million and a half about to enter it — will get more than a platitude and a link to a Coursera course.
The cost of surviving is the people left behind. The cost of surviving is the ladder with no bottom rung. The cost of surviving is an industry that still needs engineers but no longer needs all of them, and is still figuring out how to say so.
It has not figured it out yet. The silent layoffs are how it avoids figuring it out. The frozen salaries are how it avoids figuring it out. The deflated guidance and the optimistic earnings calls and the transformation narratives — all of it is a way of not saying the thing.
The thing is this: the old model is gone. The new model works, but it works for fewer people. And nobody has a plan for the ones it doesn’t work for. Not the government, not NASSCOM, not the firms, and certainly not the 1.5 million engineering graduates who will walk out of college next year into an industry that used to hire them by the trainload and now wonders, quietly, whether it needs them at all.