Cognizant

IThe business

Cognizant is an American company in form and an Indian one in practice. Its headquarters is in Teaneck, New Jersey, and 256,900 of its 351,600 people at the end of 2025 were in India.⁴ It builds and runs software for large companies: it modernizes old systems, writes new applications, tests them, keeps them running, runs back-office processes and, more lately, installs AI.⁴

Revenue was $21,108M in 2025 and $21,891M in the twelve months to September 2026 C (calculated). It comes from four industry groups, split by who the client is.⁴¹⁴

Health Sciences serves hospitals, health insurers and drug makers, and it is the most profitable group. Financial Services serves banks, payments firms and insurers, and it is the one that is growing. Products and Resources covers manufacturers, retailers, travel firms, utilities and, since the Belcan purchase in 2024, aerospace engineering. Communications, Media and Technology is the smallest and the most uneven; the filing says its largest customer drove the growth in technology clients, and does not name it.^1441417

What it sells, and how it is paid

Consulting and technology services, meaning projects, were $13,533M of 2025 revenue, 64%. Outsourcing, meaning application upkeep, infrastructure and business processes run for a client, was $7,575M, 36% C (calculated).⁴ By contract type, $10,010M was fixed price, $9,149M time and materials, and $1,949M was paid by volume or transaction C (calculated).⁴ Fixed price matters now. On a fixed-price project Cognizant bears the risk that the work takes more hours than planned. If AI tools cut the hours, the client may expect the price to fall; if they do not work as promised, the cost lands on Cognizant. The auditor flagged the estimates of labor still to come on fixed-price contracts as the one critical audit matter in 2025.⁴

Contracts are short in practice. The company says many can be ended on short notice without penalty. It reports $6,224M of work contracted but not yet done, of which about 35% is due within a year, and that figure has been flat for a year ($6,279M at the end of 2025).⁴¹⁴ A business with a twelve-month order book of $2.2B on $22B of revenue has to sell most of next year’s revenue during the year C (calculated).⁴

The part of growth that is its own

Cognizant’s reported growth is not all earned the same way. Since 2024 it has bought Thirdera, Belcan, 3Cloud ($728M) and Astreya ($670M), and agreed on 21 October 2026 to buy DyFlex for about $210M.⁴¹⁴ It also resells other companies’ software and hardware inside its bundled deals, which it calls integrated offerings, and the company reports how many points of growth each source added. Take both out and the growth in its own services is small.

Own-services growth was about 1.4 points in each of the first two quarters and 0.8 in the third, 1.2 for the nine months C (calculated).¹⁶¹⁵¹⁴ Resold products carry little margin, which is why the company says they dilute it.¹⁴ In Communications, Media and Technology, resale added 6.5 points and acquisitions 4.0 to a 6.2% nine-month rise, leaving about −4 for the rest C (calculated). In Financial Services, the group management points to, resale added 3.0 points and acquisitions 1.25 to 11.4%, leaving about 7 C (calculated).¹⁴ So Financial Services is real growth. The rest is thinner than the headline.

I do not know how much of the resale is a one-off. The company does not give the margin on it, and I treat it as growth that adds revenue and little owner cash.

IIThe moat

What stops a rival taking the work at a lower price? Cognizant has some protection, and less than its size suggests. The protection comes from four things.

• Knowledge of the client’s systems. Many contracts run for years, and the people who keep a bank’s loan system alive know how it works. Moving that to another firm is slow and risky.⁴
• Scale in India. 256,900 people in India, 22 million square feet of delivery space, about 90% of it in India, and a hiring machine that took in 489 graduates from 58 business schools in a single program this autumn.⁴¹⁷
• Domain work in a few industries. The TriZetto software for health insurers is its own product, and it won a jury verdict against Syntel in a dispute over it; the judgment, entered in April 2026 at about $298M, is under appeal and is not on the balance sheet.¹⁴
• Standing with large clients. It signed 28 deals above $100M in 2025 and trailing bookings were $28.7B in September 2026, 1.3 times revenue.²⁰¹⁷

Against that, three things wear it down. First, every large rival has the same four. Cognizant names thirteen direct competitors, from Accenture to Wipro.⁴ Second, clients can bring work in-house. The 10-K lists the global capability centers that clients build in India as a lower-cost alternative, and Cognizant now sells services to help clients set them up.⁴¹⁷ Third, the product itself is changing. The company writes that some services it performed have been replaced by AI and will be again.⁴

The evidence is in the numbers. A business with real protection raises margins when it grows. Cognizant’s operating margin was 17.0% in 2016 and 16.1% in 2025, with a low of 12.7% in 2020 C (calculated).¹³⁴⁹ Revenue per person is higher, $60,000 against $52,000, but wages in India rose too C (calculated). Over ten years revenue grew 56% and operating profit 48% C (calculated). That is a good business, not a protected one.

What would deepen the protection

If AI makes it cheaper to build and run software, the firm that owns the client’s context, its data, rules and history, may be hard to replace, and Cognizant says so: its strategy is to build that context into AI systems.⁴ That claim is untested. In October 2026 it described a logistics client where a seven-person team using an AI engineer handled a legacy conversion and recorded a 37% net cost saving against the conventional approach.¹⁷ The company gave that as a win. An owner can read it the other way: the saving shows up as a lower bill for the client.

IIIThe competitive advantage

The useful comparison is with the firms that do the same work. Of the firms I could check, Cognizant is the second largest by people and sits in the middle on margin.

Two lines stand out. Infosys has almost the same revenue and people and earns four points more margin on it C (calculated).²⁸ Accenture earns about the same margin as Cognizant but charges $91,000 for each person against $61,000, because more of its staff are in high-cost countries and more of its work is advice.²⁷ Cognizant sits between a premium firm and a cheaper one, which usually means no edge on price or cost. Its edge is that the clients it serves in health and finance are large and sticky.

Growth tells the same story. Accenture grew 5% in local currency and Infosys 5.4% a year over four years; Wipro’s IT services shrank 1.6%.²⁷²⁸²⁹ Cognizant’s own-services growth of about 1% is at the low end of that group, and its reported 4.3% is helped by purchases. I do not call that a verdict. Rivals also buy and resell.

What it does have

• A way into the largest buyers in the US. 75% of revenue is North American, and Financial Services is growing at 11.4%.⁴¹⁴
• Healthcare software it owns. TriZetto runs claims for health plans, and in October 2026 it released AI agents for claims processing on its Facets and QNXT platforms.¹⁷
• A margin that has held. Adjusted operating margin rose from 15.3% in 2024 to 15.8% in 2025 and guided about 16.1% in 2026, seven quarters of year-over-year rise C (calculated).⁴¹⁷

IVManagement

Ravi Kumar has been chief executive since January 2023. He came from Infosys, where he was president and ran its global services business, and before that worked at PwC, Oracle and Sapient.⁴ Jatin Dalal, chief financial officer since December 2023, came from Wipro, where he held the same job for eight years.⁴ Stephen Rohleder, a former Accenture group chief executive, chairs the board.²³ The senior team is largely drawn from rivals, which brings knowledge and also means the plan is the industry’s standard plan.

How they are paid, and what that pays them to do

The 2025 total was 477 times the median employee worldwide ($45,138) and 153 times the median in the United States.²³ The cash bonus paid 169% of target for corporate leaders. It rests on three things: constant-currency revenue growth with acquisitions removed (55%), adjusted operating margin (35%) and a set of strategy goals (10%).²³ Stock units, the larger part of pay, rest on revenue growth against ten rivals (50%), adjusted EPS growth (25%) and shareholder return against the S&P 500 (25%).²³ The 2023 to 2025 stock units paid out at 57%.²³²³

Follow what that pays people to do. It pays them for revenue growth, margin and EPS. It does not pay them for cash, return on capital or the price paid for a purchase. The revenue test is relative to peers, which includes the ones that buy. Adjusted EPS rises when the share count falls, and Cognizant’s fell 6.5% in the nine months to September 2026 C (calculated).¹⁴ A manager paid this way gains from a buyback funded by debt at a low share price, as in 2026, and from an acquisition that adds revenue, even at a high price. I do not say that is what drives their choices. I say it is what the plan pays for.

Their own money

Kumar held 113,011 shares outright in April 2026, worth $7.4M at today’s price, plus 647,130 unvested units. The five named officers together held 350,012 shares, about $22.8M.²³ Kumar’s guideline is six times salary, $7.8M, and the company counts him as in compliance, using some unvested units.²³ The 20 directors and officers together hold 541,662 shares, under 0.2% of the company C (calculated). BlackRock owns 9.0% and State Street 5.3%; no founder or family controls the stock.²³ These are people with real stakes in pay, not owners in the Buffett sense.

What they did with capital over a full cycle

In the ten years to 2025 it spent $12.8B buying back about 187M shares, and the count still fell only 21%, from 609M to 479M, because it issued about 57M shares to staff and for purchases C (calculated).³² At $68 an average, those shares are worth less today than were paid, by 5% C (calculated). The 2026 purchases look better: $1.84B at $55, including $500M through an accelerated purchase at $51.54, bought largely in the months when the stock fell.¹⁵ On 18 May 2026 the company raised its 2026 buyback target from $1B to $2B, drew $1B on its credit line, and the chief executive said the share price significantly undervalued its prospects.²¹ The shares rose 9% that day to $51.40, then fell 25% to the $38.73 low of 30 June C (calculated).¹ Buying into a falling stock with borrowed money is a bet on the answer to the AI question. It may prove right. It also moved the company from $1.3B of net cash at the end of 2025 to none C (calculated).^{41411854161514}

Five of the last eight years carry a charge, $795M through 2024 and about $215M more in 2026 C (calculated).^1096514 An owner should count that as a running cost, not as a one-off excluded from adjusted margin. I do: my owner cash is after the cash cost of these programs.⁹⁴¹⁴

Acquisitions and the past

Cognizant bought $6.4B of companies from 2017 to 2024 and $1.3B in the first nine months of 2026, and goodwill rose from $2.6B in 2016 to $8.1B C (calculated).³²¹⁴ Goodwill is 55% of equity. The company says its goodwill tests show headroom, and after the June fall in the share price it assessed qualitatively and found no impairment, but it warned that a further fall before its annual test on 31 October 2026 could require a full test.¹⁵

The past also includes an internal investigation into payments tied to Indian facilities, which the SEC and Justice Department closed in 2019; a securities class action it settled for $95M, mostly paid by insurers; and a derivative suit dismissed in September 2026.⁹¹⁴ Two matters are open: a jury found for plaintiffs in a US case on how non-South Asian staff were let go from the bench, and the company has recorded nothing; and a trial date of 8 December 2026 is set for related individual claims.¹⁴ I cannot estimate either.

VFinancials

Ten years of results show a business that grew steadily, kept its margin and returned most of its cash, with revenue up 56% and earnings per share up 79%, the second because the share count fell C (calculated).¹³⁴³²

Two notes. 2017 net income carries US tax reform and 2025 carries a $390M non-cash tax charge from the 2025 US tax law, which is why net income fell while operating profit rose.⁴ And 2016 cash from operations was low and 2020’s high; 2024’s was cut by a $360M deposit with Indian tax authorities.⁴ The TTM columns add the nine months of 2026 to 2025 and subtract the nine months of 2025.^{131211109876541432}

What an owner can take out

Reported net income over the last twelve months was $2,573M. The company’s free cash flow, $2,363M, is cash from operations less purchases of property. Take out stock pay of $193M and what is left, $2,170M, is my owner cash: 84% of reported net income, 62% of adjusted operating profit C (calculated).¹⁴¹⁷⁴ Capital spending at $331M is close to depreciation of property (about $332M in 2025), so there is no hidden gap there C (calculated).⁴

Three things sit behind that number. Receivables and unbilled work rose by $340M in nine months and the days-sales figure went from 81 to 89, which means clients are paying more slowly or contracts bill later C (calculated).¹⁴ Incentive pay for 2025, the highest bonus funding since 2018, was paid in 2026 and took accrued pay down from $1,490M to $1,228M.¹⁴²³ And the company says lower cash tax payments partly offset the pay-out this year.¹⁴ The first two are costs I think will ease. The third is a benefit I would not count on repeating. They roughly offset, so I use the TTM as it stands, and I use a conversion of 65% of adjusted operating profit forward against 68% on average from 2021 to 2025 and 62% now C (calculated).

Where the cash went

Over ten years it spent $23.7B, $12.8B on buybacks, $4.5B on dividends and $6.4B on purchases, against owner cash of $19.4B C (calculated).³²⁴ It funded the difference from cash it had built up. In the first nine months of 2026 it spent $3.7B against free cash flow of $1.6B, using a $1B draw on its credit line and its cash C (calculated).¹⁷¹⁴ Debt is $1,552M against cash of $1,552M, with the new $550M term loan and $1,850M credit line due in October 2031.¹⁴²² That is not a stretched balance sheet. It is a change in direction from the cash cushion of the past decade.

This year, and the guidance

The top of the revenue range fell from 6.5% to 4.4% in nine months while the EPS range barely moved. That is margin and buybacks doing the work. Q3 revenue of $5,664M rose 4.6%; bookings in the quarter fell 7%; trailing bookings of $28.7B are 1.3 times revenue.¹⁷ Headcount rose 10,100 in a year to 359,900 while attrition was 12.6% and utilization 86%.¹⁷¹⁸ The stock fell 5.8% on results day, 29 October, and rose 4.1% the next day C (calculated).¹²⁰

VIValuation

First, how this goes badly wrong

Before the numbers I ask what would most likely ruin the business or the investment, and work back from there. I see three ways.

1. AI makes the work cheaper and clients keep the savings. If the hours needed to build and run software fall by a third and clients expect prices to follow, revenue shrinks faster than Cognizant can cut its people. 47% of revenue is fixed price, utilization is already 86%, and revenue per head moves slowly. The first sign would be headcount growing faster than revenue; it is not yet, 10,100 added against 4.6% revenue growth in a year, but own-services growth of 0.8% leaves little room C (calculated).⁴¹⁷
2. Borrowed money buys the wrong things. A company that spends $3.7B in nine months against $1.6B of free cash flow, buying stock after a fall and firms at prices that add goodwill, can destroy value without a single bad project. Goodwill is $8.1B, 55% of equity C (calculated).¹⁷¹⁴
3. A cheap price for a shrinking business. This is the one that fits the investor most. The shares can look cheap on cash for years while the cash falls. If owner cash fell 3% a year for ten years and then stopped growing, the shares would be worth about $48 at my rate, 26% under the price, even though they would show a 7.5% cash yield on the way down C (calculated).

The first is the risk to the business. The second is a risk from management. The third is the risk to a buyer who confuses a low multiple with safety.

What I cannot judge

I cannot judge how far AI will cut the hours needed for software work, nor how much of the saving a client will demand. Nobody can yet; the evidence so far is a handful of case studies the company chose to publish. I cannot see the margin on resold products or the revenue of recently bought firms by themselves. I cannot estimate the Indian tax case ($360M on deposit, none reserved), the discrimination case, or the effect of the India-Mauritius treaty change, which entered into force in August 2026 and could raise the tax rate.¹⁴ And I cannot pin the discount rate: the beta of the stock against the market is 0.77 over three years with a standard error of 0.55, 0.86 over five and 1.00 over ten C (calculated).²⁶ Each of these is a range. I show ranges.

The model

I value owner cash over ten years and then at a steady growth rate. Each year is revenue times adjusted operating margin times a conversion to owner cash. The conversion, 65% in the base, is what is left after tax at about 25%, working capital, capital spending, the cash cost of restructuring and stock pay, set from the five years to 2025 (68% on average) and the last twelve months (62%) C (calculated).¹⁴²³ Revenue starts at the $22.1B middle of the guidance for 2026.¹⁷ I value at 30 September 2026, where cash of $1,552M offsets debt of $1,552M, and divide by 447.0M shares. I do not add unvested units to the share count, since stock pay is already taken out of cash.¹⁴ I leave out three items that could go either way: the $360M deposit in India, the Syntel judgment of about $298M under appeal, and the $210M DyFlex purchase.¹⁴

The bear is a business whose prices fall, whose people cannot be cut as fast as work disappears, and whose margin returns to the 2020 level. The shares touched that value in June. The base is 2026 repeated: growth near the 2025 organic pace, a margin that rises 0.3 point and a tax rate that does not move. The bull is Cognizant growing like Accenture, with a margin within two points of Infosys’s.

Discount rate. I use 9.0%. The 10-year Treasury yielded 4.94% on 30 October and the implied equity premium is 4.20%, so a stock with a beta near 0.95 needs about 8.9%.²⁴²⁵ I round up. A lower beta from the last three years would give 8.2%, but its standard error is too large to lean on C (calculated).²⁶

What the price needs

Reverse the model. At $65.11 and a 9% rate, owner cash has to change by -0.8% a year for ten years from 2026’s $2.3B, then grow 2.5% a year C (calculated). That is, the market is paying for owner cash that falls slowly and never recovers. At 8% the price needs −2.8% a year, at 7% −5.1%, and at 10% +1.1% C (calculated). My base grows owner cash 3.2% a year to $3,166M C (calculated).

In the base, the cash of the next ten years is worth $18.4B today and the value after 2036 is worth $20.6B C (calculated). That value after 2036 is 53% of the total, lower than in most growth names, because the business pays out cash now.

Even at a 10.5% rate and 1.5% growth my base is $67, just above the price C (calculated). Every cell of the table is above $65.11. That tells me the base is not fragile to the rate; it says nothing about whether the base is right. The key input is the cash flow. A conversion of 55% takes the base to $74, and the bear case, with lower margin and shrinking revenue, takes it to $42.

What is not in the number

Acquisitions are outside it. I model only the business as it stands, and growth of 3% without buying anything. Cognizant has spent $7.7B on purchases since 2017 against $2.2B a year of owner cash C (calculated).³²¹⁷ If half of that spending is just keeping revenue from shrinking, owner cash is about $300M a year lower than I show, and the base falls by $11 C (calculated). I cannot tell from the filings which half it is.

The same dollar elsewhere

A dollar in the 10-year Treasury earns 4.94% with no risk if held.²⁴ My base cash flows return 11.3% a year on the stock, 6 points more, for the risk that I am wrong about AI. The best comparable on this site is ServiceNow, where my base returns about 11% a year at $105 on a 10% rate.³³ The two are the same bet on the same debate: both sell to the same buyers and fall together if AI cuts the budget. They are not two independent ideas, and taken together they are one bet counted twice.

Quality and price are two questions

How good is the business? Ordinary. It has no pricing power I can see, thin protection, a margin that has been flat for ten years and growth that is mostly bought or resold. How good is the price? Better than the business: 7.5% owner cash yield, 25% below my base, a bear case the stock already touched. This looks like an ordinary business at a fair-to-low price, and the answer depends on a trend I cannot see. I would not call it a great business, and I would not call it dear.

Why the market may be right

• The AI risk is real and the company says so: the 10-K states that services it performed have been replaced by AI and will be again.⁴
• Own-services growth is 0.8% in the latest quarter, and the top of the 2026 revenue guide fell from 6.5% to 4.4% in nine months.²⁰¹⁷
• Receivables days rose from 82 to 89 in a year while cash from operations fell $189M in nine months despite higher net income.¹⁴
• Management spent borrowed money at $51 on the idea that $65 is too low, and goodwill is more than half of equity.²¹¹⁴

How the odds shrink

Multiplied out, my base comes true in all its parts about one time in five, so it is a middle, not a promise. The price needs less: about two times in five. That is the simplest way to say why the price is below my base and not far below it.

The margin of safety

My margin-of-safety price is $61, 30% under my base value E (estimate). I use 30% because the key input, what AI does to prices, cannot be known. The price is 7% above it C (calculated). At $61 my base would return about 12% a year E (estimate).

Where this could be in 5 and 10 years

For the base to hold at five years, revenue has to grow about 3% a year without buying firms, the margin has to reach 16.4% by 2030, and about 65% of operating profit has to reach owners. The value rises from $87 to $97 and then $109 because each year brings the later cash nearer. At ten years the base also needs the second five years to match the first, with owner cash near $3.2B in 2036 against $2.3B now E (estimate).

The cases differ mostly on growth and margin. The bear has revenue falling to 2030 and a 12.5% margin, so its value slips from $41 to $38 and it pays out $37 a share on the way. The bull grows 5% to 6% a year to an 18.5% margin and reaches $174. The discount rate matters less: one point moves the base about $11.

The most likely way the base is wrong is growth. Own-services growth is 0.8% now, and the base needs 3% from somewhere. At 1% a year, with the margin and conversion unchanged, the base is $75 today, $81 at five years and $90 at ten, and the return falls from 13.5% to 11.3% E (estimate). That is still above the price, so the harder test is the margin: if AI savings go to clients, growth and margin fall together, which is the bear.

A reader’s guide: these are estimates of value, not forecasts of the share price. The price can sit above or below value for years.

VIIThe case for the price

What most people miss

Two things in the filings. First, the stock is not a one-way story. The screen that flagged it looked at 5% revenue growth and two double-digit segments. In my numbers, resold products and purchases supply most of that, and Financial Services, up about 7% on its own, is the only large group with real organic growth C (calculated).¹⁴ Second, the company has been the buyer: 33.6M shares for $1.84B at an average of $55 in 2026, 16% below the $65.11 close C (calculated).^1615141

The bull case

The strongest case is management’s. On 18 May the chief executive said the company believed a fundamental shift in IT services was under way that strengthens Cognizant’s position, and that the current share price significantly undervalued those prospects.²¹ In October he said more than half of the company’s code is now AI-assisted, with revenue and adjusted operating income per employee still improving.¹⁷ The company points to 28 large deals in 2025, 12 of them in the fourth quarter, to a book-to-bill of 1.3 and to seven quarters of margin expansion.²⁰¹⁷ I looked for published views on Cognizant from the investors I usually check and found none I could verify, and I leave out sell-side targets I could only see second-hand.

Put plainly, the case is that AI raises the amount of software to be built and run, that firms with client context get paid to do it, and that Cognizant can use its own AI to keep margin while growth returns. If that is right, the owner cash yield of 7.5% is the floor of the return, and growth and a lower discount rate add to it.

Against reality

Today the business does part of what the case needs. Margin is rising, Financial Services is growing and utilization is 86%.¹⁷¹⁸ If it grows revenue 6% a year, lifts margin to 18.5% and holds conversion near 67%, my bull gives $127 and the price is cheap by half E (estimate). That needs own-services growth to go from 0.8% to about 5%, and margin to rise two points when AI savings are meant to go to clients. Today’s numbers are short on growth and long on cost cuts: Project Leap saves about $200M in 2026 and costs up to $250M.¹⁷

What a 10x would take. Ten times today is $651 a share and a market value of about $291B C (calculated). At 20 times earnings, well above today’s 12, that needs about $14.6B of net income. At the 11.8% net margin of the last twelve months, that is $124B of revenue, 5.7 times today and 1.7 times Accenture’s $74B C (calculated).²⁷ Over ten years it means 19% revenue growth a year, against 5.1% for Cognizant from 2016 to 2025 C (calculated).³² Large services firms do not do this. A tenfold rise in a decade for a company of this size is not plausible, and I would not build a case on it.

What they have to hit

Where I come down

The cash yield, the June low and the company’s own buying moved me: my base is above the price. The case for the price did not move my bear, because own-services growth is under 1% and the margin is being held up by cost cuts. A year of own-services growth above 3% with days sales back under 85 would raise my values.

VIIIRisks, and what would change my view

What would change my view

What would raise my base: own growth above 3% for a year, margin held while headcount grows slower than revenue, and acquisitions that stop. What would lower it: negative growth in two quarters, margin under 14%, or another year in which purchases and buybacks exceed free cash flow by $2B. A lower share price would not change my base, but it would raise the return. At the margin-of-safety price the same cash flows would return about 12% a year.