TCS's $2.6 Billion AI Business Has to be Re-sold Every Quarter

The most revealing exchange on TCS's June quarter earnings call was a boast and a confession delivered in the same breath. 

The boast: TCS's AI business is now billing at a pace of $2.6 billion a year, April-June's AI revenue multiplied by four, up 13.6% from the $2.3 billion of the March-ended quarter. The confession, from the chief executive himself: these engagements run one to two quarters, carry none of the multi-year annuity that underpins traditional IT services, and must be continuously replaced simply to keep the number where it is.

That second sentence deserves more attention than the first, because annuity is not a detail of the IT services model. It is THE model. TCS's economics rest on application maintenance, infrastructure management and BPO contracts that run three to five years, renew at high rates, and generate cash predictable enough to support a policy of paying out 80 to 100% of free cash flow. 

The market, in turn, pays a premium multiple for that visibility. What management described on the call is a business with the opposite character: project work that ends when the project does.

A yearly pace of $2.6 billion means roughly $650 million a quarter, or about Rs 61 billion at the current exchange rate, some 8.5 % of quarterly revenue. If engagements last one to two quarters, then between half and all of that book expires every ninety days. 

To merely stand still, TCS must close $325 to 650 million of fresh AI contracts each quarter. To grow at the rate it just did, it must do meaningfully more. And the burden compounds with success: at $5 billion a year, the replacement bill doubles. This is a business that gets harder to sustain the bigger it becomes, which is precisely the inverse of the annuity book, where scale begets stability.

Multiplying one quarter by four dresses up as yearly income, a stream that, by the CEO's own account, cannot be assumed to survive the year. Companies annualise when a number is too new to impress on its own terms. Lumpiness follows mechanically, and the company said as much: quarterly AI revenue will swing with the timing of contract closures rather than with underlying demand. 

That cuts both ways for readers of future filings. A weak AI print will not necessarily signal fading adoption, but by the same logic the 13.6% sequential jump just reported carries less information than it appears to. An annual figure assembled from such short engagements is a snapshot of recent sales execution, not a base.

The cost side is no kinder. Short-cycle work carries the consulting economics of continuous origination, proposal engineering, presales benches and partner ecosystems, all of which sit in overheads rather than billable delivery. It is at least suggestive that other expenses grew 26% year on year in the June quarter, twice as fast as revenue, and that management attributed the subsidiary network's operating drag partly to investment in scaled AI infrastructure. The AI business, on the visible evidence, is being won expensively and delivered on assets the company has only begun to depreciate.

The generous reading is that this is a phase. Early enterprise AI adoption naturally starts with pilots and proofs of concept. The land-and-expand argument holds that today's eight-week agent deployment becomes tomorrow's long-term managed operations contract, rebuilding the annuity at a higher technology layer. 

That may prove right, and TCS's incumbency across thousands of client estates positions it as well as anyone. But management has offered no conversion metrics, no disclosure of how much AI revenue has rolled into recurring contracts, and no view on when the mix might turn. 

Which leads to the uncomfortable valuation point. Indian IT majors trade at a premium to global peers substantially because their revenue is visible years out. If the fastest-growing line in the portfolio is explicitly non-recurring, the correct response is to apply a lower multiple to that revenue stream, not the higher one the AI narrative invites. A rupee of project revenue that must be re-won every quarter is worth less than a rupee of contracted annuity, however fashionable its label.

The number to watch from here is therefore not the headline figure, which will rise as long as the sales engine keeps closing, but the mix. The quarter TCS can report multi-year AI contracts with annuity characteristics is the quarter the $2.6 billion becomes a foundation. Until then, it is a treadmill reading, taken while running.