The financial press loves a neat comeback story. Every quarter, analysts squint at earnings reports from Bengaluru and Mumbai, spot a slight uptick in discretionary spending, and declare that Indian IT services groups have finally turned the corner thanks to booming AI revenue.
It is a comfortable narrative. It keeps institutional investors calm and justifies bloated valuations. If you liked this article, you should look at: this related article.
It is also completely wrong.
What the market celebrates as an "AI-led recovery" is actually a slow-motion margin compression disguised as innovation. The major outsourcers—Tata Consultancy Services (TCS), Infosys, Wipro, and HCLTech—are not riding a fresh wave of exponential growth. They are frantically running up a down escalator, trading high-margin, multi-year human labor contracts for lower-margin, volatile software integration work that inherently destroys their own billing models. For another look on this event, check out the recent update from Financial Times.
The Lazy Consensus of AI Revenue
When an IT services giant announces a multi-billion-dollar "AI deal pipeline," financial journalists treat it like net-new money falling from the sky. They assume Fortune 500 CIOs suddenly found extra cash under the sofa cushions to fund generative AI experiments.
That is not how enterprise IT budgets work. Budgets are zero-sum.
When a global bank signs a $500 million contract with an Indian outsourcer for AI-driven cloud modernization, that money is almost always cannibalized from existing application maintenance and business process outsourcing (BPO) budgets. The vendor isn't expanding the pie; they are slicing their own piece thinner and calling it a feast.
I have sat across the table from Chief Procurement Officers at Fortune 100 firms during these exact renewals. They do not view AI as a reason to pay Indian vendors more. They view AI as a cudgel to demand a 30% to 40% reduction in baseline run-rate costs. If the vendor refuses to pass those AI productivity gains straight back to the client in the form of lower fees, the client simply takes the work to a hungry mid-tier competitor or builds an internal Global Capability Center (GCC) right down the street in Hyderabad.
The Linear Trap vs. Exponential Code
To understand why the current optimism is misplaced, you have to look at the fundamental unit economics of traditional Indian IT. For three decades, the model was beautifully simple: linear headcount growth. You won a contract, you hired 5,000 fresh engineering graduates from tier-two colleges, you trained them for three months, and you billed them out at a healthy arbitrage spread. More revenue required more bodies.
Generative AI breaks that equation permanently.
If an AI coding assistant allows a mid-level developer in Pune to write, test, and debug legacy Java migrations twice as fast, the client does not pay double for the developer. The client demands the project in half the time with half the team.
Right now, IT executives are masking this deflationary pressure by bundling small, experimental proof-of-concept (PoC) AI projects into massive, traditional infrastructure renewals. They count the entire Total Contract Value (TCV) as an "AI-enabled win." It is an accounting sleight of hand. Once those PoCs move into production, the steady-state maintenance revenue—the golden goose that funded Indian IT's 20% operating margins for twenty years—evaporates.
Why You Are Asking the Wrong Question
If you are an investor, a tech buyer, or an industry analyst, asking "How much AI revenue are Indian IT firms generating?" is the wrong question entirely.
The real question is: "What is their net revenue retention per employee after accounting for AI-driven deflation?"
Nobody wants to highlight that metric because it exposes the structural vulnerability of the labor-arbitrage model. You cannot hype headcount utilization when your core technological offering is explicitly designed to eliminate billable hours.
Let’s dismantle the three biggest myths currently propping up the sector's narrative.
Myth 1: "AI will create more jobs for IT services than it destroys."
This is the standard corporate talking point, usually accompanied by vague promises of "upskilling" hundreds of thousands of engineers. Let's look at the mechanics. Yes, migrating messy enterprise data lakes to make them ready for Large Language Models (LLMs) requires significant human effort today. Data engineering is booming.
However, data cleanup is a finite, project-based spike, not a permanent annuity. Once an enterprise's data architecture is unified and automated agents handle Tier-1 and Tier-2 IT support, manual QA testing, and routine code refactoring, the tail of human labor drops off a cliff. Upskilling 300,000 manual testers into AI prompt engineers doesn't work when the market only needs 30,000 of them.
Myth 2: "Proprietary AI platforms will turn service firms into high-margin software companies."
Every major Indian outsourcer now has a branded AI platform—TCS has WisdomNext, Infosys has Topaz, Wipro has ai360. Executives love pitching these platforms as intellectual property that will unlock SaaS-like margins.
In reality, enterprise buyers rarely treat these platforms as standalone software products. They treat them as accelerators—free wrappers that the vendor is expected to throw in to win the services bid. Nobody is paying Salesforce or ServiceNow licensing multiples for an outsourcer's orchestration layer built on top of OpenAI, Anthropic, or Microsoft Azure APIs. The underlying foundation model providers and hyperscalers capture the software margin; the Indian IT firms remain stuck doing the messy, low-margin plumbing.
Myth 3: "Global Capability Centers (GCCs) are just a temporary trend."
While analysts celebrate minor sequential revenue ticks at traditional outsourcers, the real boom in India is happening inside GCCs—captive tech centers owned directly by Western corporations like JPMorgan Chase, Walmart, and Goldman Sachs.
Why are Western firms aggressively expanding their own captive centers in Bengaluru and Chennai instead of handing those AI budgets to Infosys or Wipro? Because no sane CIO wants to outsource their core AI intelligence. When you outsource AI development, you hand your proprietary operational workflows and domain data to a third party who will inevitably use those learnings to serve your competitors. Enterprises are keeping high-value AI engineering in-house and leaving the commoditized, low-margin legacy maintenance to the traditional outsourcers.
A Thought Experiment on the Margin Squeeze
Imagine a scenario where a Tier-1 Indian IT firm manages a $100 million annual application support contract for a European telecom giant. Historically, this required 1,200 engineers working in shifts, yielding a comfortable 22% operating margin ($22 million in profit).
Enter the AI transformation. The vendor pitches an autonomous incident-resolution system. It works brilliantly. Within 18 months, ticket volume requiring human intervention drops by 65%.
The telecom giant's procurement team immediately calls for a contract restructuring. They demand a 40% price cut, bringing the annual contract down to $60 million. The IT vendor reduces its team to 450 higher-paid AI specialists and absorbs the licensing costs of the underlying cloud LLMs. Even if the vendor manages to maintain a 20% margin on the new deal, their absolute profit drops from $22 million to $12 million.
Multiply that dynamic across 500 Fortune 1000 clients. That is not turning a corner. That is shrinking to survive.
What Enterprise Buyers and Leaders Must Do Instead
If you are sitting on the buying side of these contracts, or trying to navigate this shift as a technology leader, stop accepting the status quo of IT outsourcing. The old playbooks are dead.
- Kill Time-and-Materials (T&M) Contracts Immediately: If you are still paying an outsourcer by the hour or by the headcount for software development, QA, or customer support, you are subsidizing their inefficiency. Why would a vendor deploy an AI tool that finishes a task in two hours if you are paying them for ten? Shift 100% of vendor agreements to strict outcome-based or fixed-price milestones with aggressive year-over-year productivity clawbacks.
- Separate the Plumbing from the Intelligence: Use traditional Indian IT giants for what they do best: massive-scale cloud migrations, legacy mainframe upkeep, and routine ERP rollouts. Do not let them own your core AI model fine-tuning or proprietary data pipelines. Keep that architecture inside your own walls or within your own captive GCC.
- Audit the "AI" in Your Vendor's Pitch: When a services firm pitches their proprietary AI platform, demand to see the underlying architecture. Nine times out of ten, it is a thin UI wrapper calling standard hyperscaler APIs. Do not pay bespoke development rates for commodity API integration.
The Counter-Argument (And Where I Could Be Wrong)
Intellectual honesty requires admitting where this thesis has blind spots. Could Indian IT services pull off a genuine renaissance? Yes, under one specific condition: Jevons Paradox.
In economics, Jevons Paradox occurs when technological progress increases the efficiency with which a resource is used, yet the total consumption of that resource rises rather than falls because demand explodes. If AI makes building custom software 10 times cheaper, global enterprises might suddenly decide to build 50 times more software applications than they ever dreamed possible.
In that world, the sheer volume of custom enterprise apps, regulatory compliance frameworks, and localized AI agents required globally could overwhelm internal IT teams, forcing CIOs back into the arms of Indian outsourcers simply because nowhere else on earth has the sheer density of English-speaking engineering talent at scale.
Even if Jevons Paradox saves their top-line revenue growth, however, it will not save their old business model. The era of effortless 25% margins built on bodies and billable hours is over.
Stop Confusing Activity with Progress
Right now, Indian IT boardrooms are furiously rebranding everything they touch. Legacy automation is now "agentic workflows." Basic data warehousing is now "AI foundation engineering." Investors are eating it up because they desperately want to believe the worst of the tech slowdown is behind us.
Do not mistake a clever PR pivot for a structural turnaround. Until these giants prove they can decouple revenue from headcount—and capture real software margins rather than passing all AI efficiency gains straight to their clients—every "AI-led recovery" headline is just noise masking an existential crisis.
Stop looking at the deal pipeline announcements and start looking at the revenue per employee. The numbers don't lie, even when the press releases do.