The AGM on 9 June opened with the usual showreel — the London Marathon, Jaguar TCS Racing, the Tata AI Pavilion. Tata Consultancy Services (TCS) chairman N Chandrasekaran ran through the customary performance recital and then changed gears.

The purpose of the rest of his speech, he told shareholders, was different. AI, he said, is not a threat to TCS; it is the biggest opportunity in the company’s history.

By 2028 to 2030, all of TCS’s revenue will have an AI component; and the day is not far when the company will have as many AI agents as it has employees — which today means 5,84,000 of each.

It is a grand vision. But the latest annual report reveals several spoilers.

The Truth Behind The Margins

Revenue rose 4.6% year-on-year to Rs 2,67,021 crore in FY26. Sounds healthy, until you notice the rupee fell from 85.44 to 93.92 against the dollar during the year, a 10% slide.

For Indian IT, which bills its clients in dollars but reports in rupees, a sliding rupee flatters the numbers. Reported revenue and margins swell without a single new dollar of business being won, growth that lives in the exchange rate rather than in the order book.

In dollars, TCS’s revenue stood still at about $30 billion. The celebrated 25% operating margin, the best in four years, also leaned on the weak rupee.

The chairman admitted as much in the Q&A, saying the currency gave “a net benefit on the top line and the margin line”.

TCS's AI business is now running at a pace of $2.3 billion a year, up from $1.5 billion two quarters earlier, a fast clip, but still only about 8% of total revenue. The chairman expects it to double every year. But it is still only about 8% of total revenue.

Which begs the harder question of how much of it is new money and how much is old work with a new label?

Clients buy AI to spend less on IT services; therefore, AI revenue can grow impressively while total revenue goes nowhere. In dollar terms, that is exactly what just happened.

TCS’s own management has conceded the mechanics: clients, CEO K Krithivasan has acknowledged that AI-led productivity gains “may lead to productivity gains in existing contracts”, and industry analysts estimate application-maintenance and BPO contracts are being repriced lower as automation absorbs the routine work.

Much of the new AI revenue, in other words, is old revenue coming back at a discount.

Pricing The Prophecy

Now put a price on the vision itself.

What does an AI agent cost? Cognition sells its Devin coding agent at $500 a month. OpenAI has reportedly planned agents priced between $2,000 and $20,000 a month.

Take a modest Rs 4.5 lakh per agent per year. Multiply by 5,84,000 agents — the parity the chairman predicts — and the bill is about Rs 26,000 crore a year.

TCS would not be expected to pay sticker prices. A buyer deploying agents by the lakh would extract enterprise terms. But even at half the rate, the bill is roughly Rs 13,000 crore a year, a quarter of last year’s profit, recurring. And the discount cuts both ways: the same price deflation applies to what TCS can charge clients for agent-delivered work.

TCS’s entire net profit last year was Rs 52,820 crore. Half the bottom line would go on agents, before they earn a single new rupee.

The deeper flaw in the prediction is the unit itself.

An AI agent is not a one-for-one stand-in for a human developer. Software does not come in headcount.

One agent can work round the clock, be copied a thousand times at near-zero cost, or sit idle; a single task may need fifty agents chained together.

Counting agents against employees is like counting software licences against people — it sounds precise and means nothing.

The forecast traps itself. If agents are countable, paid-for units, the bill above applies. If they are not, “as many agents as employees” is not a plan but a slogan.

Could TCS build the agents itself and skip the bill?

TCS owns no AI model of its own. Anything it builds will run on someone else’s intelligence, on someone else’s chips. Its own plans prove the point.

HyperVault, the new data centre venture with TPG, the US private equity firm with more than $250 billion under management, will spend Rs 18,000 crore building over a gigawatt of capacity in partnership with OpenAI and AMD. What this means is that TCS, the textbook asset-light company, is entering one of the most capital-hungry businesses there is.

The split of the cheque matters. TPG comes in as a minority financial partner; the larger share of the equity and the venture’s appetite for debt are TCS’s to carry.

One shareholder asked the obvious question: the asset-light model is what produced the 25% margin, so why the confidence in land and gigawatts?

The timing is odd, besides. TCS paid out Rs 39,799 crore to shareholders last year — 81% of profit — just as its strategy turned expensive. Nor will the strain sit conveniently off the books. Unless TCS structures itself into a minority, HyperVault consolidates: the venture’s capex, depreciation and borrowings will flow through the accounts of a company whose own capital spending has historically run below 2% of revenue.

What Was Said And What Was Filed

The workforce story has the same problem. Asked about jobs, the chairman was firm: “There is no downsizing of staff; that’s not planned at all.” But the same annual report carries a Rs 1,388 crore restructuring charge, taken in July 2025, for the 12,000-plus people who were let go.

That’s what cutting about 2% of the workforce cost. The chairman is predicting a company where agents equal humans. Nobody has provisioned for what that transition would cost the other 98%.

The report also says 2,70,000 employees gained “advanced AI skills” in one year, tripling in twelve months. Skills that 1,80,000 people can pick up in a year are not advanced. The number measures how low TCS has set the bar, not how high its talent reaches.

Where the money actually went is revealing, too.

TCS paid Rs 6,750 crore for two Salesforce consulting firms, ListEngage and Coastal Cloud. Of that, Rs 6,675 crore, or 99%, was booked as goodwill.

In plain terms, TCS bought client relationships, not technology.

The company points to 573 AI patents and a 2,81,000-person hackathon. But its list of highest-paid employees, a statutory disclosure, is full of delivery and sales managers and short of research scientists.

The chairman promised an “AI operating system” with agents for every industry. Who, exactly, will build it?

His frankest moment came on the share price, stuck below Rs 2,200 at multi-year lows. Share prices fall, he explained, either because profits fall or because the market doubts future growth.

In industry transitions, price-to-earnings multiples fall. What he meant as reassurance is a confession. A falling multiple means the market expects earnings to disappoint. Calling it a “transition” accepts the forecast and argues only about the word. As for when double-digit growth returns: “whether in FY27 or in FY28 — we will have to wait and see.”

Why does one company’s AGM matter?

Because TCS is not one company. It is the bellwether of India’s largest services export industry, the business model that built a middle class on the simple trade of Indian talent at global prices. AI reprices that trade.

The AGM showed a company that talks about the transition far more fluently than it has planned for it: flat dollar revenue dressed up by a falling rupee, an agent vision nobody has costed, a capital-heavy pivot funded alongside an 81%payout, and a reassurance on jobs that its own accounts contradict.