Cognizant Says AI Creates Jobs—Its $320 Million Severance Plan Says Otherwise

Cognizant spent $84 million in the three months ended June 30 shutting down jobs it no longer wanted — $56 million of it pure severance — under a program it calls "Project Leap." That is a company-scale cost, and it is only the first installment of a 2026 bill expected to run $230 million to $320 million. The same company's chief executive, Ravi Kumar, sat at a Fortune summit in June and told industry peers that artificial intelligence will create more jobs than it destroys, dismissing the warnings of Sam Altman and Dario Amodei as "fearmongering". In early September he put it in writing: "We believe AI will create significantly more jobs than it displaces."

Read those two facts side by side and the story stops working. A company that believed AI creates far more jobs than it destroys does not normally spend a third of a billion dollars to remove them. The reconciliation is in the headcount, and it is the difference between the jobs story Cognizant wants investors to hear and the arithmetic its own filings supply.

The jobs are being swapped, not created

Look at the numbers the company itself put out. Cognizant says it plans to hire 24,000 to 25,000 fresh graduates in 2026, about 20% more than the roughly 20,000 it hired the year before. Ravi Kumar cites that 20,000-plus gross hiring figure whenever he makes the "AI creates jobs" case. Now look at total headcount: 356,700 people on June 30, 2026. That is a decrease of 900 from three months earlier and an increase of just 12,900 from a year before.

Hire twenty thousand-plus people and add fewer than thirteen thousand over the trailing year. The difference — the several thousand who left or were shown the door, the ~1% of the workforce the early Project Leap announcements put at roughly 4,000 roles, with outside reports sketching a far larger 12,000 to 15,000 — is the part of the sentence Ravi Kumar omits. This is not a growing pyramid. It is a pyramid being re-laid with cheaper stones.

The company's own explanation confirms the mechanism. Under Project Leap it is creating two new job families it calls "Frontier Certified Engineer" and "Frontier Business Operator," roles it openly says do not require technical backgrounds — a history major who can spot agentic work, a biology major, an HR accountant. It plans to scale these combined roles to 15,000 people. Those are low-cost, entry-level, trainable positions. On the other side of the ledger sit the mid-level engineers and managers that AI output cheapens, the people drawing the higher wages in the over-250,000-strong India workforce that is expected to absorb the largest share of the cuts.

So both claims survive only because they measure different things. "AI creates jobs" is true in the gross sense Cognizant hires twenty thousand freshers. "AI displaces jobs" is true in the net sense that headcount barely moved while the company paid workers to leave and replaced them with cheaper ones. The jobs created are not the jobs cut, and the timing is not coincidence — the hiring, the severance, and the margin guidance are one program.

What the severance is really buying

Strip the AI-era language out of Project Leap and the remaining shape is a labor-arbitrage trade. The cost is real: the Q2 charge of $84 million, the $230 million to $320 million expected for the full year, a hit management sizes at roughly $0.50 to $0.70 per share before tax in 2026. The promised return is the thing investors can actually track — margin.

The adjusted operating margin for the second quarter was 16.0%, up 40 basis points year over year, and management held full-year guidance at 16.0% to 16.2%, still 20 to 40 basis points of expansion on top of the year-earlier level. It simultaneously lifted full-year adjusted EPS guidance to $5.70 to $5.82, growth of 8% to 10%. The revenue underneath it, worth noting, is genuinely moving: $5.5 billion in the quarter, up 4.5%, with financial services up 12% — and the quarter closed $1.39 billion of AI-focused acquisitions. This is not a sick book of business being rescued; it is a profitable one being squeezed for margin.

That is the crux for an investor, and it is worth naming exactly. A restructuring that swaps cheap entry-level workers (and AI tooling) for expensive mid-level ones produces one-time-looking margin gains. Whether they endure is the entire question. If the added margin is real operating leverage — AI letting a smaller, flatter workforce bill and deliver as much — then a stock trading around 12 times trailing earnings, down roughly 29% on the year, with a ~2.3% dividend yield and about $2.6 billion in trailing free cash flow, is cheap. If it is one-time labor arbitrage that every competitor from Accenture to the Indian services giants can copy with the same AI tools, the credit fades as fast as it appeared — and the share price's slide reflects the market already suspecting exactly that.

Three ways this ends

The evidence here is not a fraud allegation; it is a documented discrepancy between a public claim and the company's own footnotes, confirmed across Cognizant's press releases, its CEO's interviews, and its quarterly filings. That is worth stating plainly, because it changes how much weight the "AI creates jobs" slogan should carry: it is narrative, not arithmetic, and the arithmetic is a cost program.

The genuinely unresolved question — the one the next earnings reports will settle — is which of three outcomes runs. In the friendly one, the flattened pyramid holds, attrition and severance stop consuming the gains, margin settles a step higher, and the mid-teens multiple on $5.70-plus of EPS proves too gloomy. In the persistent-but-lawful one, this becomes a treadmill: hire fresh, shed trained, capture a few margin points, watch competitors do the same AI arbitrage, repeat next year — a story that stays true in the press release and stays mediocre in the stock. In the worst one, the automation that justifies the layoffs also compresses the billable rates and delivery headcount AI was supposed to augment, and the savings from cutting people get handed back in lost projects.

The settling documents are already on a schedule: each quarter's operating margin against the 16.0% to 16.2% guide, the attrition number (13.0% for tech services over the trailing year, already drifting up), and — most tellingly — the net headcount. Ravi Kumar will keep saying AI creates more jobs than it displaces. The shareholder invoice is in the line that says how many people, on net, are actually still on the payroll.