On February 4, 2026, nothing changed in the global economy. No recession was announced. No interest rates moved. No geopolitical crisis escalated overnight. Anthropic simply released a set of AI tools capable of autonomously handling legal contracts, financial reconciliation, sales pipelines, and customer support workflows.
By the end of that trading session, Indian IT stocks had collectively shed nearly two lakh crore rupees in market capitalisation. The Nifty IT index recorded its steepest single-day decline since the COVID crash of March 2020. Infosys, TCS, Wipro, HCL Technologies, LTIMindtree, all down five to eight percent. Not because their businesses had deteriorated. Because investors suddenly understood, all at once, exactly which part of those businesses was at risk.
Wall Street analysts called it the SaaSpocalypse. The name was dramatic, but the underlying logic was not.
Stock markets are imperfect, but they are fast. And what the market repriced on February 4, and again through February and into March as the Nifty IT index continued its slide toward a 25 percent year-to-date decline, was not the demand for technology services. It was the demand for people to deliver them.
The traditional IT services model is built on a simple and, for three decades, extremely reliable equation. Clients need software built, maintained, tested, and supported. You deploy engineers. The more engineers, the more revenue. Revenue scales with headcount. The business is essentially a staffing arbitrage: Indian engineers at a fraction of Western salaries, delivering work that Western companies need. At its peak, this model powered a $300 billion export industry and employed millions.
The numbers make the structural problem visible. TCS doubled its revenue from $15 billion to $30 billion over the past decade. Its headcount nearly doubled alongside it, from 320,000 to over 600,000. Revenue per employee stayed essentially flat, hovering around $49,000. Growth required people. More people, more revenue. The model was headcount as the unit of value.
That equation is now breaking. Not because the work is disappearing. Because the ratio of people required to deliver it is compressing, fast.
There is a comparison that gained significant traction on LinkedIn in the weeks following the February crash, and it deserves to be taken seriously rather than dismissed as hyperbole.
Kodak invented digital photography. The engineers at Kodak built the first digital camera in 1975, understood what it could do, understood what it would mean for film, and spent the next two decades managing the tension between protecting their existing revenue and embracing the technology that would eventually replace it. When the shift came, it came faster than the organisation could adapt. A company that invented the future failed to survive it.
Indian IT firms are not Kodak. The comparison is imprecise. But the dynamic it points at is worth naming: these organisations understand AI deeply, deploy it for clients every day, have trained hundreds of thousands of employees on generative AI fundamentals, and are simultaneously dependent on a delivery model that AI is designed to compress. The tension between what they sell and what they use to deliver it is real, and it is accelerating.
The question is whether they adapt the model before the market forces the adaptation on them.
To understand why IT stocks are absorbing such a harsh repricing, it helps to understand what the traditional delivery pyramid actually looks like and which layers AI is dismantling first.
The classic IT services pyramid sits roughly like this:
The bottom and middle of that pyramid are where the headcount and therefore the revenue lives. And they are precisely where AI agents are demonstrating the clearest productivity gains.
The non-billable, non-strategic roles, the buffer headcount that large IT firms have historically carried as bench strength and training pipelines, are becoming the liability that investors were pricing. That capacity costs money to maintain. It was defensible when it converted reliably into billable hours. It is harder to defend when AI can cover an increasing share of what it was covering.
On February 25, 2026, TCS CEO K Krithivasan stood at the Nasscom Technology and Leadership Forum in Mumbai and said something that reverberated across LinkedIn for days afterward.
He told employees, publicly and directly, to use AI to deliver work faster and more cost-effectively for clients, even if doing so reduced the company's own billable hours in the short term.
It was an admission that long-term competitiveness now requires the company to actively undermine its own near-term revenue model. That you cannot protect the billable hour and also be genuinely AI-native. That the choice has to be made.
TCS is not alone in making it. Accenture, which trades nearly 50 percent below its 52-week high as of early 2026, launched a unified delivery structure it calls Reinvention Services, breaking down the silos between consulting and technology delivery, removing the layers between strategy and execution. The company trained over 550,000 employees on generative AI fundamentals and is actively separating out the roles that cannot make the transition. At the same time, its AI bookings have reached $11.5 billion cumulatively, proving that the demand is there, if the delivery model can evolve to capture it.
The market is not punishing IT because there is less work. It is repricing the firms based on how convincingly they can demonstrate that their cost structure and delivery architecture can adapt to a world where AI handles an increasing share of execution.
The delivery model that works in an AI-native environment looks materially different from the one the industry built over thirty years. This is not a theoretical exercise. It is a design problem with specific answers.
The natural instinct when restructuring is to cut from the bottom. That instinct is precisely wrong in an AI transition. The junior execution layer is where AI delivers the clearest productivity gains, but the roles disappearing fastest are not junior engineers. They are the roles in between: the project coordinators whose job was to translate client requirements into task queues, the delivery managers whose value was knowing who on the bench was available and allocating them accordingly, the QA leads whose entire function was manually reviewing work that automated testing now covers faster and more thoroughly.
Industry utilisation benchmarks consistently show that management roles in professional services bill between 30 and 50 percent of their available time, compared to 85 to 90 percent for junior delivery staff. The remainder is coordination: status updates, bench allocation, internal reporting, meeting overhead. In an AI-native model, that coordination work moves to software.
For decades, large IT firms sold their scale as a feature. Thousands of engineers on the bench meant they could ramp a project from zero to a hundred people in weeks. Clients paid a premium for that optionality. In an AI-native delivery world, that argument inverts. Bench strength is now carrying cost for a capability that AI can increasingly substitute.
The test is simple: if a bench role could be filled by an AI agent plus a smaller number of senior people in the time it takes to onboard a bench resource, the bench role is overhead. If it requires institutional knowledge, client relationship continuity, or regulatory accountability that cannot be delegated to an agent, it is strategic reserve. Most organisations, if they do this exercise honestly, find the majority of what they carry as buffer capacity falls into the first category.
The billable hour creates a perverse incentive in an AI-native delivery environment: the more efficient your delivery, the less you earn. A firm that uses AI to complete a six-month project in ten weeks has just destroyed a third of its expected revenue on that contract, while delivering exceptional client value. The economics do not work, and no rational firm will continue optimising for efficiency under that structure.
Outcome-based pricing resolves the incentive problem. The contract is priced on the result: the system delivered, the process automated, the error rate reduced to a defined threshold, the migration completed. How the firm achieves it, how many people it deploys, how much AI does the heavy lifting. Those are delivery decisions, not billing variables. When delivery cost drops because AI has compressed the execution layer, the margin improvement flows to the firm, not back to the client through reduced hours.
There is a framing that treats human roles in an AI transition as the ones that are too complex for AI to handle yet. That framing is too defensive. The better framing is the roles that are made dramatically more capable by AI: the senior architect who can now evaluate and prototype three solution approaches in the time it used to take to document one; the client partner who can run a discovery workshop and return with a fully modelled delivery estimate the same day; the domain expert who can translate institutional knowledge into a working AI-assisted workflow without needing a development team to build it.
The bearish read on IT stocks treats this as a story about contraction: fewer billable hours, smaller teams, compressing margins. That read is accurate for the firms that cannot adapt.
Indian IT firms collectively train and deploy some of the most technically capable engineering talent in the world. They have deep relationships with the largest global enterprises. They understand legacy systems at a level that no AI can replicate from outside: the thirty-year-old COBOL mainframes, the custom ERP implementations, the interdependent systems where changing one component without institutional knowledge of the whole creates catastrophic risk. That knowledge, held by experienced practitioners, is not being replaced. It is being amplified.
The ones that are still managing the tension rather than resolving it are the ones the market is discounting hardest.
The February 2026 crash was a forcing function. It made explicit what had been implicit for two years: the headcount model is not the safe option anymore. Carrying large bench strength, billing on hours, and managing delivery through hierarchical coordination layers was the defensible position when AI was hypothetical. It is a liability now that AI is operational.
The structural question for every IT services leader is not whether to change the delivery model. The market has already answered that. The question is the sequence and the speed.
These are not technology questions. They are organisational design questions. And the firms that answer them well over the next eighteen months will look very different from the ones that absorb the full impact of the repricing without changing anything.
The model is breaking. That is not the same thing as the industry breaking. The distinction matters, and the next few years will be defined by how clearly each firm understands the difference.
We help enterprises design the transition — not just the technology. Map exceptions, redesign roles, and build systems that hold at scale.
Book a Scoping Call
