Hook: The 91% Consensus That Rewrites the Pricing Rulebook
A single data point from Lazard’s latest PE secondaries investor survey has quietly triggered a structural recalibration of software asset pricing: 91% of respondents now identify “proprietary data + network effects” as the primary moat for software companies. This is not a typical 50-70% split—it is a statistical anomaly that signals the end of the old valuation framework. Only 4% of investors have not changed their investment approach. The market has stopped debating whether AI will disrupt software and is now pricing how it will happen. The ledger remembers what the code forgot: the value anchor has shifted from “code” to “data and relationships.”
Context: The Mechanics of a Valuation Vacuum
PE secondaries investors are the canaries in the coal mine for private market pricing. Their collective “wait-and-see” stance—shifting capital to other opportunities—is not a panic; it is a rational response to an unquantifiable risk. The traditional software valuation framework (EV/Revenue multiples based on growth, margins, and net dollar retention) is collapsing because the underlying revenue model is being questioned. The survey reveals that the market is entering a “valuation vacuum”: the old rules are dying, but new standardized metrics for AI exposure, data asset quality, and network density have not yet emerged. This vacuum creates both risk and alpha opportunity for those who can quantify the unquantifiable.
Core: Code-Level Analysis of the Moat Consensus
1. The Technical Boundary of the Moat
The 91% consensus is built on a specific technical assumption: the generative capabilities of large language models (LLMs) are becoming a commodity, so differentiation must come from private data. This is technically valid in the short term. Transformer-based LLMs learn patterns from public corpora. Proprietary data—transaction logs, user behavior sequences, compliance-sensitive records—that never enters the public web cannot be directly reproduced by the model. This creates a structural barrier. However, as synthetic data, federated learning, and context windows expand (from 4K to 1M+ tokens), the “uniqueness” of private data is being eroded. The 91% consensus contains a time-dimension bias: it substitutes today’s inimitability for permanent inimitability. Based on my audit experience, the same pattern appears in smart contract security—what is unbreakable today may be vulnerable tomorrow as the attack surface grows.
2. Network Effects as a Complementary System
Network effects (two-sided markets, data flywheels) are not being replaced by AI; they are being amplified. A platform with more users generates more behavioral data, which feeds AI personalization, which attracts more users. This is a closed-loop moat that combines data, model fine-tuning, and user stickiness. The survey correctly identifies this as a “combinatorial innovation” rather than a simple feature addition. The technical implication: transactional SaaS (simple CRUD tools) without network effects will face 40-70% replacement risk, while vertical SaaS with deep domain data and network density will see only 10-20% substitution. The code is ephemeral, but the relationships and data embedded in the network are not.
3. The Hidden Cost of AI Adoption
Software companies that embrace AI face a structural cost shift: from high-margin (70-85% gross) R&D-heavy models to lower-margin models that include inference costs (API fees or GPU depreciation). The survey’s “wait-and-see” reflects the fear that even if revenue grows, margins will compress. A traditional SaaS company with 80% gross margin and 30% net margin might, after AI integration, see gross margin drop to 60% and net margin to 10%. The valuation multiple must adjust accordingly. The 91% moat consensus implicitly assumes that only companies with “data + network” can maintain pricing power and absorb these costs. The rest will be squeezed.
Contrarian: The Blind Spots in the Consensus
1. The Reliability Moat That Investors Ignored
Only 4% of investors changed their approach, but the survey did not ask about “reliability” as a moat. In enterprise B2B software, the deterministic nature of traditional software (predictable outputs, audit trails) versus the probabilistic nature of LLMs (hallucinations, non-determinism) provides a significant defense window. Compliance-heavy industries—healthcare, legal, finance—cannot afford AI hallucinations. Software companies that offer “guaranteed correctness” may have a moat that the 91% overlooked. This is a classic case of the market over-rotating toward the new narrative.
2. The Open-Source Model Threat
Open-source models (Llama, Mistral) are rapidly catching up to proprietary ones. If any software company can deploy a fine-tuned open-source model on its private data, the moat from “data” becomes dependent on the quality of fine-tuning, not the model itself. The 91% consensus may be underestimating how quickly open-source will commoditize the model layer, shifting the competitive advantage back to execution and distribution—not raw data ownership.
3. The Self-Fulfilling Prophecy of Capital Withdrawal
When PE secondaries investors shift capital away from software, they create the very valuation compression they fear. A 10-15% discount in software asset pricing becomes a self-fulfilling prophecy. The actual disruption may take 2-3 years to materialize, but the pricing adjustment happens in 6 months. This creates a window for contrarian buyers who can identify software companies with genuine data moats at discounted prices. The market is pricing fear, not fundamentals.
Takeaway: The Valuation Vacuum Will Be Filled by Forensics
Silence in the logs speaks loudest. The Lazard survey is not a prediction—it is a snapshot of consensus. The next 12-18 months will see a wave of M&A as large platforms acquire vertical SaaS companies with proprietary data, and PE secondaries investors who held cash will re-enter at lower valuations. The winners will be those who can systematically measure “data moat quality” and “AI exposure” using forensic accounting methods. The ledger remembers what the code forgot: the true value lies not in the AI feature, but in the data that feeds it. Investors who wait for the new standardized framework to emerge will miss the alpha. The vacuum is the opportunity.
