Time Horizon: 18-24 months. CRE data moat is structural (decades-scale durability). Homes.com execution is the swing variable and requires 4-6 more quarters to resolve. AI displacement risk to portals materializes over 3-5 years.

Base Rate: Large-cap vertical SaaS with proprietary data assets. Reference class: domain-specific data platforms facing AI disruption narrative (SPGI, MSCI, Verisk). Base rate for 18-month survival with V > 2.5: ≈90%. Base rate for outperformance from oversold levels (RSI < 25, -40%+ drawdown): ≈55% over 12 months.

Base rate: Vertical data platform, V > 2.5 → 90% survive AI cycle
Prior odds: 9.0

Alpha vs Beta:

Current pricing: $40.20, Fwd P/E 22.6x, -49% YoY
Analyst mean target: $65 (+61%)
  Market beta (CRE macro recovery):  +10-15%
  Sector beta (SaaS re-rating):      +5-10%
  Idiosyncratic alpha (V-Score):     +5-10%  ← the actual thesis

The V-Score contribution is modest. This is not primarily an AI survival trade — it's a CRE macro + Homes.com execution trade where AI survival analysis provides a floor under the core business.

V-Score Card

TICKER: CSGP                    VERDICT: EMBEDDED
V-SCORE: 3.01                   FAST SCREEN: 1.5/3

  C (Compound Cognition)     4 × 0.25 = 1.00
  E (Irreducible Infra)      3 × 0.22 = 0.66
  U (Ecosystem Breadth)      4 × 0.18 = 0.72
  A (Distribution)           3 × 0.12 = 0.36
  M (Ecosystem Gravity)      3 × 0.15 = 0.45
  F (Friction Penalty)       3 × 0.06 = 0.18
                              ─────────────────
  Raw                         1.00 + 0.66 + 0.72 + 0.36 + 0.45 − 0.18 = 3.01

  Gate 1: E = 3 > 1           PASS
  Gate 2: A = 3 > 1           PASS  (C+E+U = 11 < 12, but A > 1 sufficient)

  V = 3.01

CoStar Group builds and maintains the most comprehensive commercial real estate database in the industry — 35+ years of physically collected property data spanning hundreds of data fields, from tenant credit scores to lease expirations to suite-by-suite occupancy. Two segments: Commercial Real Estate ($1.79B, Adj EBITDA $672M) provides data, analytics, and marketplaces for CRE professionals. Residential Real Estate ($1.46B, Adj EBITDA -$230M) operates Apartments.com (dominant, $1.2B+ run rate) and Homes.com (challenger to Zillow, still burning cash). Total revenue $3.25B, 93% subscription, 58 consecutive quarters of double-digit growth.

Dimension Analysis

C = 4 — Compound Cognition

CoStar has spent "more than 35 years building and acquiring databases of real estate information" (10-K FY2025). The database tracks hundreds of data fields — location, zoning, building characteristics, unit availability, tax assessments, true ownership, sales and lease comparables, rents, vacancies, tenant credit scores, lease expirations, forecasts, and demographics. This is not a flat table. It's a web of cross-referencing relationships where each new data point improves the context of every related record.

Three properties of this cognition compound:

Inter-module dependency. The same centralized database powers CoStar analytics, LoopNet marketplace listings, and residential neighborhood content. The 10-K states explicitly: "Our ability to utilize the same commercial real estate information across our standardized platform creates efficiencies in operations and improves data quality for our customers." CRE research capabilities now produce "original, media rich content of neighborhoods, schools, parks, new home communities" for residential products. Data collected for one product improves another.

Network contribution. CRE professionals "routinely take the initiative and proactively report available space and transactions" back to CoStar through the Marketing Center (10-K). The database grows from its users, not just its researchers. This creates a data flywheel: better data attracts more professionals, who contribute more data.

Re-derivation cost. Xceligent attempted to replicate the database and was caught with 38,489 stolen CoStar copyrighted photographs. Federal court awarded $500M in damages (CEO Florance, Q3 2025 earnings call). Even with theft, replication failed. Building from scratch would require hiring 1,500+ field researchers, establishing broker relationships across every U.S. market, and physically inspecting millions of properties. Minimum 3-5 years to reach comparable coverage, and historical data (35 years of comps) cannot be recreated at all.

Score 4 reflects deep domain encoding with genuine compounding across subsystems. Not a 5 because the cognition, while deep, is confined to the real estate vertical.

E = 3 — Irreducible Infrastructure

The thermodynamic question: can intelligence flow to local? For CoStar's core CRE data — no. A local LLM cannot produce today's vacancy rate at 200 W Adams St, the tenant who signed a lease at 1 Penn Plaza last Tuesday, or the cap rate on yesterday's sale in Atlanta. This data requires centralized, continuously updated physical collection: field researchers inspecting buildings, canvassing tenants suite-by-suite, photographing properties, and geo-coding locations (10-K). The freshness requirement is the irreducibility — stale CRE data is worthless, and no amount of compute replaces a researcher walking into a building.

Services are distributed under "subscription-based license agreements that typically renew automatically, a majority of which have a term of at least one year" (10-K). Remaining contract backlog is $527M recognized over four years (10-K). Subscription revenue is 93% of total.

Why not E=4: Zero regulatory mandate. No NRSRO designation, no compliance requirement, no government-imposed standard. CoStar is the de facto standard, not de jure. The EU Data Act may force data portability — the 10-K discloses requirements to "facilitate customers switching to other providers" and "remove technical, contractual, and commercial obstacles to service switching." The company's own risk factors acknowledge "rapid advancement of generative AI Technologies may reduce the barrier to entry for new competitors" (10-K). Competitors already exist: REIS/Moody's Analytics for CRE data, CompStak for lease comps, Zillow for residential. The infrastructure is capital-replicable (human researchers, not physics or regulation), just extremely time-consuming.

Why not E=2: This is not a cloud wrapper. There is no local version of current CRE vacancy rates that "functions with less quality." Without the centralized database, you have nothing. Google tried CRE data and abandoned it. Xceligent tried replication and got a $500M judgment. The physical collection requirement keeps E firmly above 2.

This is the lowest-confidence score in the card. E=3 vs E=2 is the call that matters most — 0.22 weight on the largest discriminator between dead and alive companies. E=3 holds on the centralization argument, but it's the soft end of 3.

U = 4 — Ecosystem Breadth

CoStar operates 12+ products spanning commercial, residential, and hospitality real estate, serving 10+ distinct customer types:

Cross-module data flows are real. Matterport 3D tours embedded in property listings generate "nearly 40x listing detailed views" compared to listings without (Q3 2025 transcript). Subscribers with Matterports have 37% higher renewal rates. CRE research feeds residential neighborhood content. The centralized database serves as a shared substrate across all products.

International presence spans 20 countries (10-K), with 9% of revenue from foreign currencies and growing.

Score 4 reflects 10-15 workflows across multiple user departments. Not a 5 because the entire ecosystem is within the real estate vertical — no cross-industry coverage.

A = 3 — Distribution and Discoverability

CoStar lists "perception that the service offered is the industry standard" as a competitive factor in its 10-K. This is accurate — CoStar is the Bloomberg of CRE. Any AI agent performing real estate analysis needs CoStar's data. The company has 115M unique monthly visitors and 560M total visits per quarter across its residential portals (Q3 2025 transcript).

CoStar is building AI into its products: Homes AI Smart Search, developed in partnership with Microsoft, produces 69% more search filters, 37% more listing page views, 5x higher return visit rate, and 51% more leads (Q3 2025 transcript). The company is investing 50% of Homes.com software development toward AI features.

Why not A=4: CoStar is protectionist with data, not distributionist. License agreements "restrict the disclosure and use of our information and prohibit the unauthorized reproduction or transfer of any of our proprietary information" (10-K). An internal antipiracy team uses fraud-detection technology to monitor access. There is no public API ecosystem. CoStar builds AI INTO its products rather than being infrastructure AI agents route THROUGH. GEO (generative engine optimization) traffic is currently sub-1% of portal acquisition — "one large U.S. real estate portal only draws 0.45% of its top-of-funnel traffic from ChatGPT" (Q3 2025 transcript).

This posture is rational for now (protect the data moat) but creates long-term discoverability risk as AI agents become primary research interfaces. If CRE professionals increasingly work through AI agents that can't access CoStar data, the moat becomes a wall that locks customers out, not competitors.

M = 3 — Ecosystem Gravity

Scale: $3.25B revenue, 8,000+ employees across 20 countries, 100+ issued patents (96 in US, mostly Matterport technology). 35 years of accumulated historical data that cannot be retroactively created. Apartments.com has 87,000+ multifamily properties, 99% monthly renewal rate, 93 NPS (Q3 2025 transcript). Net new bookings of $308M in FY2025, up 23% from $250M (10-K). Subscription revenue is 93%, auto-renewing with 1yr+ terms.

Why not M=4: No counterparty network effects. Unlike an exchange (where both sides must be on the platform) or supply chain ERP (where your partners must use the same system), CoStar customers don't depend on each other's participation. A CRE broker can switch to REIS without their clients or counterparties needing to switch. Competitors maintain viable alternatives — Zillow (≈$2.3B revenue), REIS/Moody's, CompStak, Yardi, RealPage. Patents are explicitly "not materially dependent on any single patent" (10-K). Homes.com ARPU is declining — the 10-K discloses residential revenue growth "partially offset by a reduction in average price," indicating volume-discount pricing pressure.

F = 3 — Ecosystem Friction (Penalty)

CoStar's consumer products have low friction: Apartments.com NPS of 93, Homes.com bounce rate of 24% (down 64% YoY), AI Smart Search improving discoverability. Simple monthly subscription pricing, web-based delivery.

But the CRE enterprise products — which are 55% of revenue — have standard enterprise friction. The 10-K describes researchers "training our clients to use CoStar Group products." Service demonstrations are "our most effective sales method" (10-K), implying the product isn't self-explanatory. Field sales teams operate from offices across six countries. Visual Lease and CoStar Real Estate Manager require enterprise implementation for lease administration. There is no public developer API.

The system as a whole — 12+ products from 5+ acquisitions being integrated — has inconsistent interfaces and no programmatic access layer. Consumer portals (45% of revenue) are F=1-2 territory. CRE enterprise (55%) is F=3. The blend is F=3.

Thermodynamic Summary

Intelligence cannot flow around CoStar's CRE database because the data requires physical collection — field inspections, broker relationships, tenant canvassing — that no local model can replicate. The moat is atoms, not bits. AI enhances the product (better search, automated data validation, content generation) without undermining the data collection barrier. Every competitor who has tried to replicate this database has either failed (Xceligent, $500M judgment) or settled for narrow scope (CompStak for lease comps only, REIS for analytics only).

The residential portal business is a different story. Homes.com and Apartments.com are marketplace businesses where the moat is traffic, brand, and listing volume — all potentially vulnerable to AI-powered listing aggregation over 3-5 years. But even here, Apartments.com's 99% monthly renewal and 93 NPS suggest deep workflow embedding that doesn't evaporate overnight.

Kill Zone

AI-powered listing aggregation that disintermediates consumer discovery. If AI agents can aggregate rental and for-sale listings directly from MLSs, property managers, and public records — bypassing portal intermediaries — the marketplace businesses (Apartments.com, Homes.com, LoopNet) face structural pressure. This affects ≈55% of revenue (all marketplace/portal products).

The CRE data/analytics core (CoStar, STR, Visual Lease, Debt Solutions) is outside the kill zone. Physical data collection, historical comps, and proprietary benchmarking cannot be synthesized by AI models.

Revenue Durability

Durable (≈80%):

• CoStar data/analytics + STR + Visual Lease + Debt Solutions (≈$1.4B, 43% of total). Physical data collection moat, 35-year head start, no AI shortcut. CRE professionals need current, verified, field-collected data. AI cannot inspect buildings or verify tenant occupancy.
• Apartments.com (≈$1.0B recognized FY2025, 31% of total). 99% monthly renewal, 93 NPS, 87K+ properties. Deeply embedded in multifamily marketing workflow. While marketplace businesses face long-term AI aggregation risk, the renewal rates and NPS indicate years of runway before meaningful displacement.
• LoopNet (≈$200M of $300M, 6% of total). CRE marketplace backed by CoStar data, partially protected by proprietary listing content. Remaining ≈$100M faces AI aggregation risk.

Exposed (≈20%):

• Homes.com (≈$260M run rate, 8% of total). ARPU declining. Residential EBITDA still -$230M (improved from -$361M). Competing directly with Zillow's entrenched position. The "Your Listing, Your Lead" model is differentiated, but the market hasn't validated it yet. Burning $230M/yr with uncertain payoff.
• Matterport 3D capture (portion of $216M Other CRE). Spatial computing commoditizing through Apple Vision Pro, Google spatial tools. Camera hardware is a declining-margin business.
• Domain/OnTheMarket (≈$100M). Local portal competition in Australia and UK. International portals face entrenched local competitors.
• LoopNet partial (≈$100M). CRE listing marketplace without proprietary data backing, vulnerable to AI aggregation.

Steelman Bear Case

The strongest argument against CSGP is not AI displacement — it's capital misallocation. CoStar is taking a $672M/yr EBITDA engine (Commercial RE) and using it to fund a -$230M/yr bet (Homes.com) against Zillow, the entrenched incumbent with 2-3x CoStar's residential traffic and a decade head start in consumer brand awareness. Homes.com's ARPU is already declining, suggesting the path to profitability requires more agents at lower prices — a volume play against a competitor that already has the volume.

Meanwhile, the CRE data business grows only 9% organically. If the Homes.com bet fails, CoStar is a $1.8B CRE data company valued at $17B (9.4x revenue) with a large cost structure built for residential that has no purpose. The buyback ($700M planned 2026) signals management uncertainty about reinvestment returns.

This bear case is real and NOT captured by the V-Score, which only measures AI survival. The V-Score says the CRE core survives AI. It says nothing about whether Homes.com justifies the capital being deployed.

Kill Criteria

Thesis dies if:
- Homes.com EBITDA loss widens beyond -$300M in FY2026 → capital allocation failure
- Apartments.com monthly renewal drops below 95% → core franchise eroding
- CRE segment organic growth decelerates below 5% → data moat not translating to pricing power
- AI-native CRE data competitor raises >$500M (Google, OpenAI, etc.) → E-score drops to 2

Thesis strengthened if:
- Homes.com reaches EBITDA breakeven by H2 2027 → Zillow challenge validated
- CRE segment EBITDA margin expands above 40% → pricing power confirmed
- CoStar launches public API/data licensing program → A-score moves to 4
- Zillow lawsuits (FTC, Compass, CoStar IP) result in injunctive relief → competitive landscape shifts

LR Signal

LR = 1.3 — Mild bullish.

The market is pricing CSGP at 22.6x forward earnings, down 49% YoY, with RSI 22.5 (deeply oversold). The V-Score at 3.01 (EMBEDDED) confirms that 80% of revenue sits behind an atoms-based moat that AI cannot shortcut. The market selloff is primarily driven by Homes.com execution risk and CRE macro weakness — concerns the V-Score does not address.

The V-Score's incremental contribution is establishing a floor: even if Homes.com fails entirely, the CRE data core ($1.8B revenue, $672M EBITDA) is structurally durable. The market may be pricing some AI displacement risk into the core business that doesn't belong there. But the gap between V-Score insight and market pricing is modest — most of the selloff is fundamental (Homes.com cash burn, ARPU decline, CRE cyclicality), not AI-fear-driven.

Not LR > 2.0 because the V-Score doesn't resolve the primary uncertainty (Homes.com execution). Not LR < 1.0 because nothing in the analysis suggests the market is TOO bullish on AI survival.

Evidence