V-Score Card

V = 0.25(2) + 0.22(2) + 0.18(3) + 0.12(2) + 0.15(2) − 0.06(4)
  = 0.50 + 0.44 + 0.54 + 0.24 + 0.30 − 0.24
  = 1.78

G₁ = 𝟙[E=2 > 1] = 1  (PASS — Finacle saves the gate)
G₂ = 𝟙[A=2 > 1 ∨ C+E+U=7 ≥ 12] = 𝟙[TRUE ∨ FALSE] = 1  (PASS)

V = 1.78 × 1 × 1 = 1.78
Tier: COLLAPSED (V < 2.0)
κ = (1.78 − 3.0)⁺ = 0

Dimension Analysis

C = 2 | Compound Cognition

Infosys has one genuinely crystallized product: Finacle, the core banking platform serving 1B+ end users across 100+ countries with 12+ functional modules and an integrated BankingSLM. Decades of banking domain knowledge baked into software. Re-derivation cost: years. This is real compound cognition.

The problem: Finacle is 4.7% of revenue and declining in absolute dollars — $1,140M (FY23) to $898M (FY25), a 21% drop while total revenue grew 6% [20-F FY2025, line 12520]. The other products (Panaya, McCamish, Stater, Helix) are genuine but small.

The remaining 95.3% is human-delivered IT services — consulting, code, test, maintain. This knowledge is in people's heads, not crystallized in software. Management's own framing: "detailed understanding of how the client technology landscape is set up" [Q3 FY26 transcript]. That's relationship knowledge, not product knowledge. A frontier model re-derives IT consulting methodology in weeks.

Finacle not mentioned in a single earnings call transcript. The company's growth narrative is Topaz (an integration layer) and agent deployments (500+ built) — output metrics, not compounding knowledge. The Cognition partnership confirms INFY doesn't build the agent IP itself.

Revenue-weighted C ≈ 0.047 × 3.5 + 0.953 × 1.0 = 1.12. Rounded to 2.

E = 2 | Irreducible Infrastructure

This is the kill dimension. E=1 triggers G₁=0 and V collapses to literal zero.

The Bustamante screen scores 0/3: no proprietary data that can't be synthesized locally, no regulatory mandate requiring routing through Infosys, not a transaction rail. The 20-F is explicit: "Our clients typically retain us on a non-exclusive, project-by-project basis" with contracts "terminated with or without cause, with a notice period" [20-F, lines 1291–1299]. Benchmarking provisions allow clients to exit without penalty if pricing is unfavorable [20-F, lines 1367–1378].

The RPO ($12.3B) looks like infrastructure. It's not. It's cancellable optionality. The client IP ownership clause seals it: "our clients usually own the intellectual property in the software we develop for them" [20-F, line 3914]. INFY doesn't even accumulate what it builds.

What saves E from 1: Finacle creates genuine switching costs for banks running it as core banking. Migration costs $100M+ and takes 3–5 years. McCamish (insurance policy admin), Stater (mortgage platform), and the NHS $1.6B deal create real embedded infrastructure where switching is measured in years. These are tasks where c_ℓ is genuinely high regardless of model capability — not because the task is incomputable, but because switching involves regulatory approval, data migration, and counterparty re-coordination.

E=2 holds, but the margin is zero. The entire non-zero V rests on 4.7% of revenue.

U = 3 | Ecosystem Breadth

The strongest dimension. 7 industry verticals (none >28% of revenue), 18 BPM service lines, 8 IT service lines, 5 enterprise solution practices, operations in 59 countries across 292 locations [20-F]. Financial Services 27.7%, Manufacturing 15.5%, Retail 13.5%, EURS 13.3%, Communications 11.7%, Hi-Tech 8.0%, Life Sciences 7.3% [20-F, lines 3797–3808].

This is genuine breadth but it's table stakes for Tier-1 IT services. TCS, Cognizant, Wipro offer comparable coverage. Breadth doesn't differentiate when every peer has it. More critically, these are modular service lines — a client can engage INFY for SAP implementation without touching BPO, or use cloud services without consulting. No superlinear switching cost from breadth; φ_switch is additive, not multiplicative.

A = 2 | Distribution

Agents don't route through services firms. INFY's distribution is relationship-driven enterprise sales — valuable for winning $1.6B government contracts, useless for agent-first discoverability. No API marketplace, no self-service signup, no developer ecosystem.

The CEO was directly challenged on AI revenue quantification: "HCL Tech and TCS have been giving us concrete numbers. Why does Infosys refrain from doing so?" [Q3 FY26 transcript, analyst Ritu Singh]. He pivoted without answering. If the number were impressive, they'd disclose it. AI Day (Feb 2026) revealed 5.5% of quarterly revenue — early but not differentiated.

The Cognition partnership is the tell: INFY partners with the agent builder and deploys in client environments. The agent IP belongs to Cognition. INFY is the system integrator, not the platform.

M = 2 | Ecosystem Gravity

Scale without gravity. $19B revenue, 337K employees, 1,800+ clients. But: no data accumulation (client owns IP), no network effects (one client's engagement doesn't improve another's), no counterparty dependencies (low concentration — top 5 = 13.2%). Large deal TCV is strong ($4.8B in Q3 FY26, $11.6B for FY25), but TCV measures current demand, not structural lock-in.

Headcount growing (+13,655 YoY) while TCS cuts 30,000 [Q3 FY26 transcript]. Management claims demand confidence. But if AI delivers 20–40% efficiency, headcount growth at ≈$59,600/employee/year is scaling the cost base that intelligence rerouting will compress.

F = 4 | Ecosystem Friction (Penalty)

High friction works against INFY. Enterprise-only sales, staff rebadging in large deals, consultant-dependent implementations, no self-service. This is friction that makes adoption expensive — it doesn't protect the business, it taxes it.

The CEO's own quantification: "there could be benefits to the clients of 20% to 40%" from AI in customer service [Q4 FY25 transcript]. This is the deflationary pressure stated on the record. The CFO confirmed the self-funding dynamic: "discretionary spend expected to be self-funded through AI-led productivity benefits" [Q1 FY26 transcript]. Clients use AI savings to pay for AI work — net revenue impact is deflationary.

Thermodynamic Summary

Infosys sells computable tasks performed by human labor. The Tool Death Theorem applies directly.

D(INFY) = {code, test, maintain, migrate, consult, manage infrastructure, process BPO transactions}. All computable. As model capability M(t) → ∞, the cost to route each task through a local AI agent c_ℓ(τ,t) = g(τ)/M(t) → 0. The cost to route through Infosys is bounded below by the biological floor: c_s ≥ $59,600/employee/year.

The kill cycle is in progress. The CEO quantifies it at 20–40% for customer service. The CFO confirms clients self-fund new AI work from old AI savings. Revenue growth has decelerated from 25% (FY22) to 3% (FY26 guidance). EPS growth: 0.9% TTM.

Resistance: R(s,t) = |{τ: c_ℓ(τ,t) > c_s(τ,t)}| / |D(s)|. Today R ≈ 0.95 — most tasks still cheaper through Infosys than through AI (especially complex enterprise integration). In the limit, R → 0 because every task in D(INFY) is computable.

Durable revenue (≈5%): Finacle (core banking), Panaya (test automation), McCamish (insurance admin). Real software with real switching costs.

Exposed revenue (≈95%): Human-delivered IT services with 0–90 day terminable contracts where the client owns the IP.

The destination is thermodynamic. The timeline depends on Ṁ/M.

Regime Context

15-week factor regression (2025-12-23 to 2026-04-13):

IR = α̂/σ_idio = −0.17        (near zero — no idiosyncratic signal)
β_sector = 0.88              (moving 1:1 with IT services sector)
%Idio Var = 39.7%            (regime-dominated, below 75% target)
ρ_intra = 0.61               (high pairwise correlation across sector)
PC1 = 67.6% of sector cov   (one-factor selloff)

All IT services names down 21–41% while SPY flat. Pure sector event. ρ_intra peaked at 0.72 during the February Claude Code selloff (Nifty IT −20% in 8 sessions). The market applied a uniform AI-disruption discount — indiscriminate.

IR does not gate the verdict. IR = −0.17 means the measurement window contains no idiosyncratic signal, not that alpha is absent. When ρ → 1, ε_i → 0 for all names. IR measures the regime.

δ = V − V_market: The market treats all IT services as interchangeable AI losers. For INFY, this is approximately correct — V = 1.78 confirms the structural vulnerability. The uniform discount is the right discount. δ ≈ 0.

Conviction & Basket Weight

V(INFY) = 1.78
κ = (V − 3.0)⁺ = (1.78 − 3.0)⁺ = 0
w_INFY = W_S · κ / Σ_j κ_j = 0

Zero conviction. Zero weight. The structural diagnosis produces no basket allocation.

κ is regime-invariant. Even if ρ_intra drops to 0.3 and idiosyncratic signal returns, V = 1.78 doesn't change. Structural properties don't improve because the sector stops correlating.

Verdict: COLLAPSED

What would move the score:

Primary sources cited: 20-F FY2025 (filed 2025-07-01), 6-K Q3 FY2026 (filed 2026-01-14), Q3 FY2026 earnings transcript (Jan 14, 2026), Q4 FY2025 transcript (Apr 17, 2025), Q1 FY2026 transcript (Jul 23, 2025), AI Day disclosure (Feb 17, 2026).

Evidence table:

Combined LR (independent signals: #1–4 from filings, #5–7 from transcripts, #9–10 structural): net bear LR ≈ 150. Prior P(structural vulnerability) = 50% → posterior ≈ 99%.

V = 1.78. COLLAPSED. κ = 0. The product is human intelligence. The price of intelligence is falling.