Time Horizon: 12-18 months. C3.ai has ≈$622M cash burning ≈$155M/year. The restructuring (26% layoff, Feb 2026) buys time but doesn't change the structural E=1 problem. The relevant window is whether the company can demonstrate revenue stabilization before cash runway becomes existential.

Base Rate:

Reference class: Enterprise software cos with >30% YoY revenue decline + CEO transition + major restructuring
Base rate: ≈20% stabilize within 18 months, ≈35% acquired, ≈45% continued decline
Prior odds: 0.25 (stabilization)

Historical analogs: Cloudera (pivoted, eventually taken private at depressed multiple), MicroStrategy (pivoted to Bitcoin, abandoned software thesis), Palantir pre-2022 (burned cash for years before inflecting — but had E=4+ infrastructure). The critical difference: Palantir had irreducible infrastructure (Gotham/Foundry data ontology). C3.ai does not.

Company Overview

C3.ai sells an enterprise AI platform — middleware that sits between customer data and AI model deployment. Founded in 2009 by Tom Siebel (of Siebel Systems fame), IPO'd December 2020. The product family includes the C3 Agentic AI Platform (development/deployment runtime), 131+ pre-built industry applications (supply chain, reliability, demand forecasting, CRM), and C3 Generative AI (enterprise RAG/knowledge management).

The business model shifted from large subscription contracts to consumption-based pricing via "Initial Production Deployments" (IPDs) — 3-6 month paid trials intended to convert to recurring contracts. As of Q2 FY2026, 394 cumulative IPDs had been signed, with only 269 still active — a 32% attrition rate.

Current state as of Q3 FY2026 (Jan 31, 2026):

• Revenue: $53.3M (-46% YoY, -29% QoQ)
• Subscription gross margin: 11.4% (vs 55.9% prior year)
• Net loss: $354.8M for 9 months on $198.7M revenue
• Cash + securities: $621.9M (burning ≈$155M/year on operations)
• RPO: $225.4M (declining three consecutive periods)
• 26% workforce reduction announced Feb 24, 2026
• CEO transitioned from Tom Siebel to Stephen Ehigian (Sep 2025)

Short interest stands at 30.7%, with 43% sell-side bearish ratings. Enterprise value ≈$578M after netting $622M cash.

V-Score Card

TICKER:         AI (C3.ai, Inc.)
V-SCORE:        0.00
GATE 1 (E>1):   FAIL
GATE 2 (A>1):   PASS (A=2)
FAST SCREEN:    0.5/3

  C (Compound Cognition)       w=0.25    Score: 2
  E (Irreducible Infra)        w=0.22    Score: 1  << GATE FAILURE
  U (Ecosystem Breadth)        w=0.18    Score: 2
  A (Distribution)             w=0.12    Score: 2
  M (Ecosystem Gravity)        w=0.15    Score: 2
  F (Friction penalty)         w=-0.06   Score: 3

ARITHMETIC:
  0.25(2) + 0.22(1) + 0.18(2) + 0.12(2) + 0.15(2) - 0.06(3)
= 0.50   + 0.22   + 0.36   + 0.24   + 0.30   - 0.18
= 1.44 (raw)

  Gate 1: E > 1?  E=1, FAIL -> multiplier 0
  Gate 2: A > 1?  A=2, PASS -> multiplier 1
  V = 1.44 x 0 x 1 = 0.00

Dimension Analysis

C = 2 | Compound Cognition

C3.ai has 17 years in market and 35 US patents (plus 30 international) covering its model-driven architecture — a type system that lets developers describe AI applications as conceptual models rather than writing raw code. Patent US 12,111,859 covers generative AI agentic orchestration specifically. The company has shipped 131+ industry-specific applications across oil & gas, utilities, manufacturing, healthcare, defense, financial services, and aerospace.

That's the bull case for C=3. Here's why it's C=2:

The 32% IPD attrition rate is the tiebreaker. If the crystallized knowledge were truly at a C=3 level ("months to re-derive, lose the edge cases"), customers wouldn't walk away at a 32% rate after experiencing it firsthand in a paid production deployment. They try the platform, a third shrug, and leave. That's C=2 behavior.

The non-GAAP gross margin trajectory confirms it: 54% in Q2 FY2026, collapsing to 17% GAAP in Q3 FY2026. Subscription gross margin hit 11.4% — meaning $0.89 of every subscription dollar goes to delivery costs. This isn't reusable crystallized software. It's bespoke services wearing a subscription label.

Each industry application is also largely independent. There's no evidence of cross-module data flows where supply chain insights feed reliability models feed demand forecasting in a compounding loop. Compare to SAP, where finance feeds supply chain feeds manufacturing feeds HR in an interconnected system where the whole exceeds the sum of parts. C3.ai's 131 apps are 131 standalone tools, not a compounding system.

Sources: 10-K FY2025 line 1386 (patent count), 10-Q Q3 FY2026 income statement (gross margin), Q2 FY2026 transcript (IPD attrition), 10-Q line 2170 (agentic patent)

E = 1 | Irreducible Infrastructure (GATE FAILURE)

This is the kill variable. Dead companies average E=0.4; alive companies average E=4.2.

C3.ai's core function is enterprise AI application development and deployment. The question: can this go local? Can intelligence flow around it?

Yes, overwhelmingly. Three independent paths confirm E=1:

Path 1 — Hyperscaler native alternatives. AWS Bedrock + SageMaker, Google Vertex AI, Azure AI Foundry each replicate C3.ai's core value proposition: data integration, model orchestration, application deployment, enterprise compliance. C3.ai's own 10-K Risk Factors section lists "public cloud providers offering discrete tools and micro-services with data management, ML, and analytics functionality" as a primary competitive threat. When a company's own SEC filing names its infrastructure providers as competitors for the same function, E=1.

Path 2 — Open-source orchestration. LangChain, LlamaIndex, CrewAI, and Hugging Face provide open-source AI orchestration that replicates C3.ai's core pipeline: data ingestion, model selection, agent coordination, retrieval-augmented generation. The "model-driven architecture" is a proprietary abstraction, but the underlying capability is commodity.

Path 3 — Revenue collapse proves it empirically. Q3 FY2026 revenue fell 46% YoY. RPO declined three consecutive periods ($244M to $235M to $225M). NRR is not disclosed — itself a signal, since companies with strong retention trumpet it. Infrastructure irreducibility prevents revenue collapse by definition. C3.ai's revenue collapsed. The market is confirming E=1 in real time.

The strongest counterargument is federal/defense: air-gapped deployments, FedRAMP compliance, intelligence community contracts. These create real switching friction — but they're table-stakes certifications (Palantir, hyperscaler GovCloud all have them) and represent perhaps 15-20% of revenue. Even within federal, the core capability is replicable.

Benchmark comparison: Snowflake at E=2 had NRR declining from 158% to 125% — customers shrinking but renewing. UiPath at E=2 had NRR declining from 119% to 107%. C3.ai doesn't disclose NRR, has revenue in freefall, and its flagship partner (Baker Hughes) terminated exclusivity. C3.ai is worse than both E=2 benchmarks.

Sources: 10-K FY2025 line 2265 (cloud competitors), 10-K line 2484 (Baker Hughes exclusivity ended), 10-Q line 1961 (RPO $225.4M declining), 10-Q Q3 revenue ($53.3M vs $98.8M)

U = 2 | Ecosystem Breadth

On paper, C3.ai has impressive breadth: 131+ applications spanning supply chain, asset reliability, CRM, energy management, fraud detection, healthcare, and more, plus the Agentic AI Platform (development runtime) and C3 Generative AI (knowledge management). In September 2025, they launched C3 AI Agentic Process Automation, expanding into the RPA market.

In practice, each application is standalone. An enterprise deploys C3 AI Demand Forecasting. It works (or doesn't). There's no architectural reason that success leads to adopting C3 AI Reliability or C3 AI CRM. The platform is the shared runtime, but the applications don't feed each other's intelligence.

The company itself acknowledged this implicitly: the February 2026 restructuring narrowed product focus to three verticals — asset performance, supply chain, and procurement — in four sectors (energy, manufacturing, healthcare, defense/gov). That's a company admitting 131 apps across 7+ industries wasn't working and cutting to the subset that actually retained customers.

The IPD model reflects U=2 dynamics: customers enter for one use case, get value or don't within 3-6 months, and either convert or leave. Only 14 new IPDs were signed in Q3 FY2026 versus 50 in the prior year — a 72% decline. Breadth isn't attracting. The company is concentrating, not expanding.

Sources: Q1 FY2026 transcript line 25 (131+ apps), 10-Q line 2076 (product narrowing), 10-Q line 2001 (14 vs 50 IPDs), 10-Q line 2139 (Agentic Process Automation launch)

A = 2 | Distribution & Discoverability

C3.ai has real distribution infrastructure. The "AI" ticker provides brand recognition. Microsoft Azure marketplace integration means Azure sellers receive quota retirement and compensation when transacting C3.ai solutions. AWS partnership produced 9 joint agreements in Q2 FY2026 with 172% YoY pipeline growth. The joint Microsoft pipeline generated $130M+ in Q2, growing 146% YoY. Booz Allen joined the strategic integrator program. Forrester named C3.ai a Leader in AI/ML Platforms (Q3 2024).

89% of Q2 FY2026 bookings came through the partner ecosystem. That's distribution reach.

But distribution and discoverability in an AI-agent world are different from human-mediated sales. Agents don't browse the Azure marketplace looking for middleware. They don't attend C3 Transform conferences. The hyperscaler partnership is simultaneously C3.ai's distribution channel AND its replacement vector — when an Azure seller closes an AI deal, they can recommend C3.ai or Azure AI Foundry. C3.ai's distribution depends on partners who sell competing products.

The "strategic integrator program" launched in Q1 FY2026 — an implicit admission that existing distribution needed supplementation. No AI-specific revenue metric is disclosed, unlike ServiceNow ($600M+ AI-attached ACV) or Palantir (AIP-driven growth metrics). When a company in the AI space doesn't break out AI revenue, the number is either small or declining.

Sources: Q2 FY2026 transcript (89% partner bookings, $130M pipeline, 172% AWS growth), 10-K line 998 (Azure marketplace), 10-K line 1262 (Forrester Wave), Q1 transcript line 27 (integrator program)

M = 2 | Ecosystem Gravity

Revenue was $389M in FY2025, supported by ≈60 large-scale customer engagements, federal contracts across HHS, DoD, intelligence community, Army, Marines, and NAVSEA, plus a multi-year Microsoft global alliance and the Baker Hughes partnership (total historical commitments of $495M).

None of it held. Revenue collapsed 46% YoY in Q3 FY2026. RPO declined three consecutive periods. Baker Hughes is no longer an exclusive reseller. 32% of IPD customers walked away. The company cut 26% of its workforce.

Ecosystem gravity means customers can't leave even if they want to — the cost of switching exceeds the cost of staying. C3.ai exhibits the opposite: customers leave readily, flagship partners downgrade exclusivity, and the company itself is in survival mode with mass layoffs.

No counterparty network effects exist. Unlike SAP (where a manufacturer's suppliers and customers also run SAP, creating multi-party lock-in), C3.ai serves individual enterprises with no inter-enterprise dependency. Customer data stays with the customer — no data gravity accumulates at C3.ai.

One customer accounted for 17% of Q3 FY2026 revenue. Three customers represented 41% of accounts receivable. The customer base is narrow, concentrated, and demonstrably willing to leave.

For scale comparison: Palantir serves the same federal AI market with $2.87B TTM revenue growing 36% YoY. C3.ai is running at ≈$265M TTM pace, shrinking.

Sources: 10-Q line 959 (17% concentration), 10-K line 2484 (Baker Hughes exclusivity), 10-Q line 1961 (RPO), 10-Q line 2066 (26% layoff), 10-K line 1054 (Baker Hughes $495M)

F = 3 | Ecosystem Friction (Penalty)

Implementation friction is moderate. The IPD model requires 3-6 months before production deployment. Professional services represented 16% of FY2025 revenue. A Center of Excellence (COE) support engagement is required. Sales cycles are "long and unpredictable" per the 10-K. The platform requires integration with "a variety of hardware and software platforms."

The Q3 FY2026 subscription gross margin of 11.4% is the structural tell. When subscription revenue costs $0.89/$1.00 to deliver, the "software" is actually labor-intensive deployment services. This is friction made visible in the income statement.

Counterpoint: consumption-based pricing lowers the entry barrier, the model-driven architecture claims to reduce code (10-K line 526), and IPDs are pre-configured to deliver "production-grade Enterprise AI application within weeks." This prevents F=4 (SAP-level multi-year implementations). F=3 reflects standard enterprise complexity — not trivial, not crippling, but enough to extract a -0.18 penalty on the raw score.

Sources: 10-K line 306 ("long and unpredictable"), 10-K line 5416 (16% pro services), 10-Q Q3 income statement (11.4% sub GM), 10-K line 1147 (COE required)

Thermodynamic Summary

Nothing prevents intelligence from flowing around C3.ai. The company sells enterprise AI application deployment — the exact capability that every hyperscaler now offers natively and every open-source framework replicates for free. There is no petabyte-scale data asset, no regulated exchange, no physical constraint, and no regulatory mandate that forces intelligence through C3.ai's infrastructure.

The Q3 FY2026 results are the thermodynamic proof. Revenue doesn't fall 46% YoY at a company with irreducible infrastructure. Customers don't attrit at 32% from a platform they can't replace. RPO doesn't decline for three consecutive periods when switching costs exceed the pain of staying. The market is adjudicating the E-score question empirically, and the answer is E=1.

The February 2026 restructuring — 26% layoffs, product narrowing to 3 verticals, organizational flattening — is a survival response, not a growth pivot. The company is retreating to defensible positions (federal/defense, energy, manufacturing) and abandoning the breadth thesis that justified the "enterprise AI platform" positioning.

Kill Zone

Primary threat: Hyperscaler native AI services (AWS Bedrock, Google Vertex AI, Azure AI Foundry). These platforms replicate C3.ai's core orchestration capability, are sold by the same salesforce that distributes C3.ai (89% of bookings), and cost less because they're bundled with the underlying cloud infrastructure. C3.ai is an intermediation layer between enterprises and their cloud providers — and the cloud providers have eliminated the need for intermediation.

Secondary threat: Open-source AI orchestration (LangChain, CrewAI, LlamaIndex, Hugging Face). These frameworks provide the same agent orchestration, RAG pipelines, and model coordination that C3.ai's platform offers, at zero license cost. Enterprise AI teams increasingly build on open-source rather than purchasing proprietary middleware.

Tertiary threat: Palantir. In the federal/defense segment — C3.ai's strongest surviving market — Palantir is the dominant player with 10x the revenue ($2.87B TTM), deeper government relationships, and genuinely irreducible infrastructure (Gotham/Foundry data ontology, V=3.18). C3.ai's federal business is contested by a competitor with structural advantages at every level.

Durable vs Exposed Revenue

Durable (≈15%): Federal/defense air-gapped deployments with multi-year contract cycles and government procurement inertia. FedRAMP compliance and security clearance requirements create 2-3 year switching friction. The COTS (commercial off-the-shelf) mandate provides a tailwind as government moves from bespoke to commercial. But even this is contested by Palantir and hyperscaler GovCloud.

Exposed (≈85%): Commercial enterprise AI platform revenue is fully exposed. The 32% IPD attrition rate demonstrates that commercial customers can and do leave after experiencing the platform. Subscription gross margin of 11.4% in Q3 confirms bespoke deployment content, not reusable software moat. The restructuring's product narrowing (from 131+ apps across 7+ industries to 3 verticals in 4 sectors) is an implicit acknowledgment that the broad commercial thesis failed.

Risk Profile:

AI stock beta:       2.18 (high market sensitivity)
Idio vol:            58.6% (high stock-specific risk)
Total vol:           68.7%
Short interest:      30.7%

Bear thesis decomposition:
  Structural return expectation: -30% to -50% over 12mo
  Market beta component (if SPX flat):   0%
  Sector beta component:                 ≈0%
  Idiosyncratic component:               -30% to -50%  << the thesis

  Exposure: beta 2.18 means a +10% market rally = +22% headwind on bearish positioning.
  30.7% short interest creates squeeze risk.
  $622M cash = 4+ years of runway, limiting near-term zero risk.

The bear thesis is almost entirely idiosyncratic (structural V=0 with declining revenue) but beta 2.18 creates significant adverse exposure to market rallies, and 30.7% SI amplifies squeeze risk.

Steelman Bear Case:

The strongest argument FOR C3.ai that this analysis must honestly engage:

The new CEO Stephen Ehigian brings government execution experience (former Acting Administrator of GSA under Trump). Federal AI is a generational tailwind — the AI action plan, COTS mandates, reindustrialization of the maritime industrial base. Federal bookings grew 89% YoY and represented 45% of Q2 bookings. The restructuring focuses the company on exactly the segments where it has traction. With $622M cash and a 26% cost reduction, the company could reach cash-flow breakeven within 18 months on a smaller, federal-focused revenue base.

Response: The federal pivot is real, but insufficient. Federal at 45% of bookings (not revenue) on a shrinking base means perhaps $100-120M annualized federal revenue. Even at breakeven on that base, C3.ai would be a $100M federal AI services company competing against Palantir at $2.87B. The implied valuation for a sub-scale federal IT company would be 3-5x revenue, well below current enterprise value. The pivot extends the runway but doesn't change the destination.

Kill Criteria:

Thesis challenged if:
- Q4 FY2026 revenue > $90M (would signal stabilization)
- RPO inflects upward for 2 consecutive quarters
- NRR disclosed at >100% (challenges E=1 assumption)
- Major federal contract win >$100M (validates federal pivot)

Thesis confirmed if:
- Q4 revenue < $65M (continued collapse)
- Cash burn > $50M/quarter post-restructuring (restructuring not working)
- Another major customer loss or Baker Hughes full exit
- Hyperscaler launches competing "enterprise AI platform" product

LR Signal

LR = 0.85 (mild bearish — market slightly optimistic)

The market already prices C3.ai as deeply distressed: 30.7% short interest, 43% sell-side bearish, consensus target near current price. The V=0 confirms the structural thesis with framework rigor — gate failure at E=1 is a death sentence — but the market has largely arrived at the same conclusion through the revenue collapse.

The 0.85 reflects two residual gaps between V=0 and market pricing: (1) the market still gives C3.ai ≈$578M of enterprise value above its cash, implying some option value on turnaround or acquisition that the V-score framework says doesn't exist at E=1; and (2) the federal pivot narrative may be sustaining a modest premium that the evidence doesn't support given Palantir's dominance.

This is not a high-conviction signal at current prices. The market is already heavily short (30.7% SI), oversold (RSI 29.9), and the $622M cash pile provides years of runway preventing a near-term zero. The structural thesis is correct but largely priced.

Evidence