AI hasn't cut IT staff costs — so who is keeping the dividend?

Indian IT has changed how it delivers software without changing its wage bill. The productivity dividend is real, but it is being passed to clients as a 2 to 3 per cent price cut at renewal rather than released as savings or pay. Cognizant shows both halves: half its code is AI-assisted and its headcount still grew. The squeeze falls on the fresher while GCC demand concentrates in AI and data roles facing a 38 to 42 per cent supply gap. I suggest a funded apprenticeship pipeline, a reskilling credit tied to audited employer spend, and disclosure of AI-linked headcount change.

On 6 November, The Economic Times reported a fact I had to read twice. "Although artificial intelligence (AI) and automation are changing the traditional labour-intensive model of IT services, they have yet to bring down employee costs" — not at TCS, Infosys, HCLTech, Wipro, Tech Mahindra or LTM. New delivery model, same wage bill.

So who is keeping the dividend? Nobody inside the firm, I think. It has already been handed over at the renewal table. The HinduBusinessLine (1 November 2026) puts AI-led pass-through at 2 to 3 per cent of traditional services revenue a year. One line there is the whole story: "The talent model isn't going away, but is being repriced."

Cognizant proves both halves at once. Its 29 October results (The Hindu) showed 359,900 employees, up 10,100 in a year, with $46 million of separation costs in the quarter and about $200 million of Project Leap savings expected in 2026 — while its chief executive said over 50 per cent of its code is now AI-assisted. Six days later it placed 25,000 people in an OpenAI delivery group (EdTech Innovation Hub, 5 November). Half the code machine-written; the payroll bigger.

Its own survey, reported on 3 November 2026 (700 employees and 75 executives in India), explains the gap:

• 97 per cent of employees report productivity gains, averaging 15 per cent
• 84 per cent of leaders say AI has not cut costs; 91 per cent see no revenue impact
• 37 per cent have paused or dropped AI initiatives over unclear returns

The squeeze point is the fresher, not the mid-career engineer. Quess Corp's Q2 FY27 data (EconoTimes, 6 November 2026) has GCC tech demand down 3.5 per cent quarter on quarter, while AI, data and cybersecurity rose to 30.3 per cent of demand from 25.8 per cent — against a 38 to 42 per cent supply gap in AI, data and analytics.

Mint reported on 6 November that the third PM Internship Scheme round drew about 40,000 candidates — 2.5 times the first two rounds combined — against over 150,000 postings, after the stipend rose to ₹9,000. Pay for the entry pipeline and it fills.

In Optimism Over AI Job Loss (May 2026) I accepted Vinod Khosla's warning that "slowing down could kick up a different set of problems". I still do. These numbers add a second clause: speeding up without sharing the gains kicks up its own — a 15 per cent productivity gain arriving as a 2 to 3 per cent discount.

In Graduates Without Jobs (March 2026) I had asked for structured entry-level roles. Concretely, I would suggest:

1. Fund apprenticeships as infrastructure, tracking completion and conversion, not just postings.
2. A reskilling credit paid only against audited entry-level training spend.
3. Mandatory disclosure of AI-linked headcount change, by band, in annual reports.

I am not asking anyone to slow down. But a dividend that never reaches a payslip, or a fresher's first offer letter, is only a discount.

With regards,

Hemen Parekh
www.Iam-Immortal.ai / www.IndiaAGI.ai / www.YourContentCreator.in
6 November 2026