The Nonlinearity Test

The derating is most sensitive to a single fear: that generative AI severs the historical link between billable headcount and revenue, collapsing a labor-arbitrage model into a shrinking one. Cognizant's own filings concede AI is already replacing some of the work it sells [1]. But the numbers show the link bending, not breaking: revenue per employee rose from roughly $56,000 to $60,000 across 2021–2025, and management is re-pricing toward outcomes. The mechanism a bear fears is also the lever a bull needs.

The link the model depends on

For three decades this business converted low-cost engineering hours into client value: more projects meant more people, and more people meant more revenue. Cognizant's CEO describes the old world plainly — "for much of the last 30 years, IT services grew through a linear model. More people and more projects drove incremental growth" [2]. Revenue tracking headcount is visible in the raw counts: the company ended 2025 with approximately 351,600 employees, 256,900 of them in India [3], against roughly 330,600 in 2021 [4].

The company states the disruption risk in its own risk factors, without euphemism: "Some services that we historically performed for our clients have been and will continue to be replaced by AI or other forms of automation, including our own AI-enabled client offerings" [5]. The 2025 10-K adds the pricing corollary: reductions and replacements "can negatively impact our results of operations if we are unable to adapt our pricing," and it names clients' in-house global capability centers as a lower-cost alternative to Cognizant's services [6]. That is the bear case, stated by the company: if a unit of client value takes fewer billable hours, a headcount-priced revenue line has to fall unless something replaces it.

What the numbers actually show

If AI were breaking the model, the first place it would appear is revenue per employee — falling, as automation strips out the billable hours faster than new work replaces them. It is doing the opposite, modestly.

Revenue per employee = annual revenue ÷ year-end headcount. Revenue from Form 10-K income statements; year-end employee counts from each year's 10-K [7] [8].

The story the crude annual figures tell is a trough in 2022 — the tail of a post-pandemic hiring surge that pushed headcount to 355,300 ahead of revenue — followed by a recovery to about $60,000 per head by 2025, a 7% gain over the full window and roughly 2% a year. The single clearest year was 2024, when headcount fell by about 11,000 while revenue still grew. Management measures the same trend on a trailing-twelve-month basis and reports a steeper slope: revenue per employee up 8% year-over-year and operating income per employee up 10% in the third quarter of 2025 [9], and up 5% on revenue with 8% on margin per person by the fourth quarter [10]. The company's richer figures partly reflect acquisitions and mix; the year-end arithmetic is more conservative. Both point the same way.

Revenue, headcount and derived revenue per employee, FY2021–FY2025, from Form 10-K filings [11].

The pricing model is shifting to match. Fixed-price and success-based contracts are now about half of the total — an analyst put the figure at "about 50%" on the fourth-quarter call, and the CFO did not dispute it, describing a portfolio where "the larger component of delivery risk resides with the service provider" [12]. Management frames the direction as "shifting our economics from labor-based to outcome-based models," noting the combined fixed-price and transaction-based book has grown as a share of revenue over three years [13]. Two concrete proof points sit underneath the framing: business process outsourcing — the service most exposed to automation — grew 9% year-over-year in the fourth quarter, and digital engineering grew 8% [14]. By early 2026 the company said nearly 40% of its code was AI-assisted and it was running "well over 5,000 AI engagements," up from about 4,000 three months earlier [15].

This is a company spending to get ahead of its own disruption rather than waiting for it: it committed roughly $1 billion to generative-AI capabilities over three years back in 2023 [16] and reports having put more than 340,000 associates through AI skilling [17]. The honest read of the evidence: nonlinearity is real but early. Revenue per head is rising a few points a year, not stepping up; the fixed-bid shift is genuine but its margin economics are unproven at scale. The bull claim that AI turns this into a nonlinear growth machine is only faintly visible in the P&L, and the bear claim that it collapses the revenue line is not visible at all.

Where the industry is heading

The disruption question is not Cognizant's alone, and the outside evidence cuts against the extreme bear read. Industry data compiled by NASSCOM, cited in Wipro's FY2026 annual report, has global IT-services spending growing 4.6% in calendar 2025 — with "AI reshaping operating models and talent requirements rather than driving broad displacement" [18]. The nonlinearity Cognizant reports is showing up sector-wide: HCLTech's CEO told investors his firm had "grown 4%-5% and our headcount has not grown" over the prior couple of years [19]. Growth is decoupling from headcount across the industry, which is consistent with margin expansion, not revenue collapse — so far.

Where the incremental growth is headed is the contested part. Cognizant's own answer is a "three-vector" strategy built around what it calls the AI velocity gap — the distance between what enterprises have spent on AI infrastructure and the business value they have realized. It cites internal research putting the prize at "$4.5 trillion in US labor value," to be captured through AI-led productivity, "industrializing AI," and building new agentic workflows that expand total addressable spend [20]. That is a large, self-generated number and should be read as ambition, not forecast. The load-bearing point for an investor is narrower: the same technology that could deflate a per-hour revenue line is also the reason enterprises need a systems integrator to move from experimentation to production — the demand Cognizant is billing against in those 5,000 engagements. Which force dominates will decide the derating, and it resolves in the numbers over the next several quarters, not in the narrative.

Bending, not breaking

The evidence points to a model bending toward nonlinearity, not breaking under it. Revenue per employee is rising, the pricing mix is moving to outcomes, the most automatable service line (BPO) is growing fastest, and the industry body and peers describe AI as reshaping the labor model rather than displacing the revenue. The strongest fact against that read is the company's own admission that AI is already replacing services it used to sell [21], combined with the thinness of the proof: a few points of revenue-per-head gain and a fixed-bid book whose realized margins the CFO says are "very close" to plan but which has not been tested through a full technology transition [22]. What would change the read: revenue per employee stalling or reversing while headcount holds, or fixed-price contracts underrunning their assumed cost curves — either would mean the productivity Cognizant is pricing away to clients is not coming back as margin. Neither has happened yet. The nonlinearity thesis is intact but unproven — a state consistent with the stock's ~8x forward multiple rather than proof of it.