A services margin, cross-checked against the comparator median of 14.9% (EPAM/Genpact/Cognizant). Constraint m_EBIT,T 15.8 <= m_gross,T 40.9 satisfied.
Risk & exit
Four distinct questions, one field each. A price is not a thesis: the trigger below forces a re-underwrite and freezes further purchases — it is never an automatic sell.
Risk trigger 21% below the memo price
$45.38
Forward E[R] vs a 0% floor
+52.9%
A daily close below $45.38 triggers a mandatory re-underwrite and freezes further purchases. It is not an automatic sell unless a separate fundamental invalidation condition has been breached.
Thesis-invalidation conditions
Fundamental and falsifiable, never price-based. If one is satisfied the thesis is marked dead and the position is retained only by explicit decision.
Top-customer concentration rises rather than falls. Management explicitly guided that the largest customer will be a *smaller* percentage of FY2026 revenue. Q1 already delivered 61% → 56%. If Q2 or Q3 shows concentration going back up, the diversification mechanism — the single load-bearing claim in the growth case — is refuted by the company's own disclosure. FY2026
The ≥40% FY2026 guide is not raised again, or is cut. Q1 annualises to $360m against a $352m guide. A guide that is merely *maintained* through Q2 and Q3 implies sequential stagnation and would break the 12-month target's revenue base. FY2026
The second Big Tech programme under-delivers against ~$51m. ~$36m remains to be recognised in Q2–Q4. It is checkable each quarter against the "another customer generated approximately X%" disclosure.
The unresolved $17.2m accrued-expense build proves to be a cost timing item that reverses, which would halve the apparent Q1 cash generation. First question for the Q2 call.
Further large insider selling at lower prices. $158.3m was sold at ~$97.5 average. Selling continuing at $55–60 would be a different signal entirely.
Impairment case
The named cause is not "AI spending slows." It is: the customer representing 56% of revenue and 65% of receivables reduces, delays or terminates its statements of work, which it may do on 30–90 days' notice without penalty. Three ways that happens, in descending likelihood: 1. Programme completion without replacement. The 10-K risk language is explicit: *"potential inability to replace projects that are completed, canceled or reduced."* This is the base mechanism. 2. In-sourcing. A frontier lab builds internal data operations, or substitutes synthetic data for human-expert annotation. No filin