Infosys, Cognizant and AI: Who Keeps the Savings?

The awkward question for an IT company selling AI is also the obvious one for its customer: if the work takes fewer hours, why should the bill stay the same?

At Infosys, that negotiation is already affecting business.

On its October 23 earnings call, chief executive Salil Parekh said clients were seeking the benefits of AI productivity through pricing. Some demands exceeded the economic returns Infosys was willing to accept.

Six days later, Cognizant chief executive Ravi Kumar S reported that more than half the company’s code was AI-assisted. Its adjusted operating margin also improved.

Those disclosures capture different parts of the same commercial problem. AI can reduce the effort required to deliver a service. Who receives the resulting benefit depends on what was sold, how the price was agreed and what happens at renewal.

For India’s IT industry, the consequential rivalry involves vendors competing to promise customers the most attractive combination of price, capability and future productivity.

The risk is promising tomorrow’s savings before knowing how reliably they can be delivered.

The discount can arrive before the saving

Parekh raised precisely that concern when discussing multiyear contracts. Small changes in productivity assumptions could make substantial differences later in a deal. He characterised Infosys as disciplined about achievable benefits, “as opposed to just imagining what can be realized.”

That is management’s account of its competitive position, rather than independent proof that rivals are bidding recklessly.

The underlying arithmetic deserves attention.

Consider an illustrative contract with a price of ₹100 and delivery costs of ₹80. It produces ₹20 of profit before any other costs. If automation cuts delivery costs to ₹70 while the price holds, that profit rises to ₹30.

If competition pushes the price down to ₹85, the profit falls to ₹15 despite the lower delivery cost. The customer receives the larger benefit.

These are hypothetical figures, but the mechanism explains why technical progress can coexist with commercial pressure. A vendor can become more efficient and earn less from the same assignment.

The difficulty deepens when a bid includes savings scheduled for later years. Signing secures revenue while creating an obligation to deliver the promised economics. The consequences become visible as the work progresses.

A generous productivity commitment can help win a contract. It also sets the hurdle the delivery team must clear.

Why the billing model matters

Cognizant’s regulatory filing makes the contract mix visible. In the September quarter, fixed-price revenue was $2.753 billion and transaction or volume-based revenue was $510 million. Together, they represented about 57.6% of revenue, calculated from the filing.

That gives a substantial part of its business a different relationship with hours worked.

Under a fixed-price agreement, reducing delivery effort can improve the vendor’s economics if the agreed price and scope hold. Billing by time gives the client a more immediate route to the saving when fewer hours are charged.

Transaction pricing introduces another variable: volume. A lower cost per transaction can be attractive to a provider if demand grows sufficiently.

None of these arrangements guarantees that the vendor keeps the gain. Buyers can negotiate lower prices at renewal. Fixed-price projects can suffer overruns. An outcome-linked contract needs agreement on what counts as success and who controls the factors that produce it.

The commercial advantage lies partly in knowing the work well enough to price those risks.

Changing the invoice heading will accomplish little if delivery remains unpredictable.

AI-assisted code is an input measure

Cognizant’s claim that more than 50% of its code was AI-assisted signals widespread adoption. It does not mean half its engineers’ work disappeared, or that delivery costs fell by half.

The company reported September-quarter revenue growth of 4.6% to $5.664 billion. Adjusted operating margin rose to 16.3% from 16%, while GAAP operating margin fell to 15.4% from 16%.

Its filing attributes the adjusted improvement principally to Project Leap savings, net of AI investment, and favourable currency movements, partly offset by compensation costs and acquisitions. The adjusted measure excludes $51 million of Project Leap charges.

That is a broader explanation than AI coding productivity.

Software still needs requirements, integration, testing and acceptance. A faster coding stage may shorten delivery, allow more work within the same budget or move the bottleneck elsewhere. The financial result depends on the whole assignment.

Investors therefore need to distinguish tool adoption from delivery improvement and from profit retained after renegotiation.

The gap between those measures is where much of the investment argument remains unproven.

Infosys has two AI businesses to balance

Infosys reported that AI revenue represented 10.5% of its September-quarter total and grew 29% sequentially. Overall revenue grew 1.5% year on year in constant currency, while operating margin was 21%.

Selling additional AI work and reducing the price of existing services can happen simultaneously.

At its media briefing, management linked its revised FY27 growth outlook to softer volumes, pricing pressure and weaker discretionary spending. It also argued that AI was creating new work, including modernising client environments.

The counterargument to a pessimistic industry forecast is therefore credible: cheaper delivery could make previously uneconomic projects worth commissioning.

A customer might use a saving to modernise another system. A vendor might sell a broader transformation programme. Greater demand could offset some revenue lost on existing assignments.

That sequence requires customers to approve new spending. Savings can also be retained by the buyer.

Parekh described the productivity adjustment as finite while presenting new AI opportunities as continuing to expand. Infosys does not disclose AI margins separately. Investors consequently cannot yet establish how much incremental profit that category contributes.

The useful question is whether profitable new work arrives quickly enough to compensate for pressure elsewhere.

What India should watch

An immediate collapse in aggregate employment is not what these quarterly figures show. Infosys ended September with 332,542 employees, up 4,480 sequentially. Cognizant reported 359,900, an increase of 3,200.

Those totals cannot reveal every change in roles or hiring requirements. They do prevent the coding statistics from being treated as a simple job-loss forecast.

For employees, the longer-term issue is which capabilities remain valuable when a contract allows fewer hours for familiar tasks. For employers, it is whether retraining creates capacity they can sell.

The next useful disclosures would connect productivity to the economics of comparable contracts: delivery costs, price concessions, renewal terms and profit after tools and training.

Company-wide margins remain necessary, but currency, acquisitions and restructuring can obscure the contribution from AI.

Infosys and Cognizant are making a substantial wager that they can sell enough valuable new work while improving how existing services are delivered. Customers have an equally rational reason to demand a share of the improvement.

AI may help finish the assignment sooner. The contract will decide how much that achievement is worth to the company doing it.