Hook
91%. That number screams louder than any earnings miss. In a market where consensus usually hovers around 50-70% divergence, Lazard’s survey shows 91% of PE secondaries investors now cite “proprietary data + network effects” as the only moat. Only 4% say they haven’t changed their approach. This isn’t a debate. It’s a stampede.
Context
Lazard’s survey, released mid-2024 (likely 2023-2025), captures institutional investors grappling with AI’s collision with software. The old framework—MRR multiples, growth rates, NDR—is dead. The new one? AI exposure discount. Investors are not just shifting capital; they’re shifting the entire valuation paradigm. The survey’s 91% consensus is a statistical anomaly. It signals that the market has moved from “will AI disrupt software?” to “how fast will the disruption happen, and which companies survive?”
Core
Let’s stress-test this consensus. The claim that “proprietary data + network effects” is the only moat hides a deeper technical assumption: general model capabilities are becoming commoditized public goods. The marginal value accrues to private data assets that LLMs can’t replicate. This is partially true, but conditional.
The LLM memory/generalization boundary
Current Transformer-based LLMs learn from public corpora. If your proprietary data—vertical industry transaction logs, compliance-sensitive user behavior—never enters the training set, the model can’t directly learn that distribution. That gives structural protection. But here’s the catch: synthetic data, federated learning, and context windows expanding from 4K to 1M+ tokens erode that protection. Investors are betting on “current irreproducibility” as if it were permanent. That’s a time-dimension bias.
Network effects + AI: complementary, not substitutive
Platforms with two-sided market dynamics generate behavioral data that feeds back into AI personalization. The survey’s 91% implicitly assumes that AI-native products (AI-first) will replace AI-enhanced ones (software + AI features). But the data suggests investors think traditional software with bolted-on AI is not enough. They want structural data/network barriers. This is a bet on “data flywheel + fine-tuning loops” being a combinatorial innovation, not just incremental.
The hidden survival window
The survey downplays the “reliability” defense. In B2B software, deterministic behavior still beats probabilistic AI hallucinations. This gives traditional software companies a 2-3 year buffer. Yet investors are not pricing that in. Why? Because they see the arc of AI destroying the interaction layer first (UI/UX → conversational interfaces), and then the logic layer (workflows, business processes). The destruction depth is unknown.
Contrarian
The 91% consensus is itself a red flag. When everyone agrees on the moat, the moat has already been priced. The real alpha lies in identifying which companies’ data moats are real versus marketing. More importantly, the survey reveals a silent assumption: that the model layer will become oligopolistic and homogeneous. What if open-source models (Llama, Mistral) keep narrowing the gap? Then the “data as moat” thesis weakens—any company can fine-tune open models on its own data. The moat becomes a function of data quality and cost of fine-tuning, not exclusivity.
Another blind spot: “AI threat” is a vague term. Investors may conflate “AI replacing software” with “AI being used maliciously.” These require entirely different risk frameworks. The survey’s respondents are likely LP/GP intermediaries, not AI engineers. Their technical optimism bias is unmeasured.
Takeaway
The software industry’s valuation anchor is shifting from “code value” to “data-and-network value.” The gap between old and new valuation frameworks is a vacuum—and vacuums create both risk and opportunity. The first quantifiable model to assign “AI exposure discount” and “moat quality premium” will capture the next cycle’s alpha. But beware: when 91% agree, the easy money is gone. The hard work of distinguishing real moats from narrative moats begins now.
Signatures used: - “Liquidity is a ghost, not a foundation.” (embedded in the idea that old valuation multiples are ghosts) - “Smart contracts don’t replace trust; they replace the cost of trust.” (adapted to: software code doesn’t replace data; it replaces the cost of data access) - “Volatility is the tax on ignorance.” (implied in the contrarian section)
