Technology services

The outsourcing pyramid didn't invert. It got squeezed from both ends.

Everyone in the industry has been saying for two years that AI is going to break the labour model. I wanted to know whether it already has, so I went through five years of filings for six firms. Here is what the numbers say, including the parts that don't fit the story.

Shwetal Mehta23 September 2026 · about 8 minutes · data through the June 2026 quarter

I ran managed-services P&Ls for a long time, at Infosys and HCL among others. The economics were never complicated. You won a deal on a rate card, you staffed it with a pyramid, a thin layer of experienced people on top of a wide base of freshers, and your margin was the gap between what the client paid per hour and what the base cost you. Every year you hired tens of thousands of graduates at a salary that hadn't changed since 2008, and revenue scaled with headcount because that's what the model was.

So I pulled the fact sheets, annual reports, 10-K and 20-F filings and earnings-call transcripts for TCS, Infosys, Wipro, HCLTech, Cognizant and Accenture, from fiscal 2021 to the latest quarter, and put them in one spreadsheet. The first chart is the whole story in one picture.

Across the four Indian majors, revenue went from $54.0 billion in FY21 to $70.0 billion in FY23 to $75.3 billion in FY26. Headcount went 1.11 million, 1.44 million, 1.38 million. From FY23 to FY26 that is revenue up 7.7% and headcount down 4.2%. The model where revenue scales with people has, at minimum, paused. The interesting questions are when, and why.

1. The hiring machine switched off in FY24. That wasn't AI.

TCS, Infosys, Wipro and HCLTech together added 243,000 people in the year to March 2022. In the year to March 2024 they shed 64,000. Infosys's campus intake went from about 50,000 to 11,900 in one year. Wipro didn't onboard a campus batch at all.

It's tempting to date the AI disruption to that year. It's also wrong. Attrition was running at 17 to 28% in FY22 and the firms over-hired into it. Then the 2023 banking scare and the discretionary spending cuts arrived. JPMorgan downgraded Indian IT in May 2022, before ChatGPT existed. The executives said so themselves. Wipro's CHRO, in April 2024: "we as an industry… all of us are over-hired… post-COVID." GenAI wasn't delivering anything at scale yet.

2. What is AI is that the base was never rebuilt.

FY25 net adds across the four: +9,000. FY26: −6,000. Combined revenue over those two years rose $5.4 billion. Campus intake in FY26 versus each company's own peak: Infosys −76%, Wipro −66%, HCLTech −56%, TCS −44%.

And the FY27 signals are weaker still. TCS made 25,000 fresher offers against roughly 44,000 hired in FY26. Wipro's CHRO said in April there is no fresher target for the year at all, and campus hiring stayed paused through June. Infosys freshers are now about 6% of headcount, versus 27% in FY22. In the old model you could not run the business that way. In the new one, apparently you can.

3. The pyramid isn't inverting. It's being squeezed from both ends.

This is the part the "inverted pyramid" story misses. The base is thinner, but so is the middle. TCS's 12,000-person cut in July 2025 was aimed "primarily in the middle and the senior grades." Accenture spent $923 million in severance over two quarters to exit people "in a compressed timeline where reskilling is not a viable path." HCLTech has run restructuring charges every quarter since mid-2025, and its CEO says people released "due to productivity improvements" can't all be redeployed because "entry level or lower end skills are being addressed through automation." Cognizant's Project Leap is mostly severance.

Meanwhile utilization is at record levels: 93% at Accenture for three straight quarters, around 85% at Infosys and Cognizant. There is no bench left to squeeze. The shape that's emerging is not a diamond or an inverted triangle. It's a column: flat headcount, fewer juniors, fewer managers, and a wage bill per head that keeps rising.

4. The productivity dividend went to the client.

If AI were making these firms more efficient, you'd expect revenue per employee to climb and margins to expand. Revenue per employee is up only 2 to 14% over five years; at Accenture it's flat. People cost as a share of revenue actually rose at five of the six. Operating margins: TCS flat at about 25%, Infosys down four points, HCLTech down four, Wipro down three.

Why? Because the gains are being handed back. TCS's CEO, in July 2026, on renewals: "the productivity gain passed on is around 10% to 15%." Infosys's CFO: "the productivity will get passed back to the client largely." HCLTech's CEO says roughly 40% of the industry's book is "AI disrupted," and, as reported from the April 2026 call, that a $100 million deal is now worth about $80 million. Kotak now models 3 to 3.5% a year of AI-driven revenue deflation through FY28.

Two cautions on this chart. Revenue per employee is partly mechanical: a hiring freeze raises it by definition, and at some firms 7 to 10% of revenue is software resale. The honest reading is modest gains, fully passed through. If you're a client, that's good news. If you're a shareholder, that's why the Nifty IT index had its worst first half since 2003.

5. So which year did it show up?

FY26, April 2025 to March 2026. That's the year TCS posted its first decline in dollar revenue since it listed in 2004 and cut mid-level staff in the same twelve months. It's the year the firms first put a number on AI revenue: TCS $2.3 billion annualized, Infosys 5.5% of revenue, both disclosed only after being pushed. It's the year HCLTech said "decoupling" on an earnings call. And it's the year the fresher pipeline, three years after the cyclical cut, still hadn't come back.

What they said, and what they did

The transcripts are more interesting than the numbers. Here is one line from each company beside one thing it did.

The honest counter-argument

Cognizant is running the opposite play. It hired about 20,000 freshers in 2025, more than double 2024, and guided to 24,000 to 25,000 for 2026 (softened to "approximately 20,000" by July), on the explicit thesis that AI pushes expert work down to juniors, so you want a "broader, shorter" pyramid. TCS added 9,279 people in the June 2026 quarter, the most in fifteen quarters. Industry-wide, NASSCOM's net additions recovered from 60,000 in FY24 to about 135,000 in FY26, and India's captive centres now employ 2.4 million people.

So the industry is split on the base of the pyramid and united only on the middle. Two years from now, either Cognizant's bet or the TCS and HCLTech "fewer, better-paid freshers" bet will look wrong. I know which way I'd lean, because the deflation numbers are hard to argue with, but the disagreement is real and worth watching.

What I'd do with this if I still owned a P&L

• Stop reporting headcount as a growth metric. HCLTech's CEO is right that the wage bill is the number.
• Price renewals assuming 10 to 15% productivity pass-through is the floor, not the ceiling.
• Treat the fresher pipeline as a strategic asset you are either rebuilding on purpose, like Cognizant, or winding down on purpose, like Wipro. Not something that drifts with the quarter.