The code spoke, but the logic was a lie. Lazard's latest survey on private equity secondary markets claims 91% of investors now see proprietary data and network effects as the only moat for software companies. That number is too clean. It smells of consensus manufactured by a bank that needs to sell advisory services. But the underlying signal is real: the valuation framework for software is collapsing, and the same tectonic shift is about to hit crypto-native applications.
Context
Lazard is an investment bank, not a neutral research house. Its survey of PE secondary market investors—timed to an August 15 release, likely in 2024—was designed to position the firm as a thought leader in the AI disruption narrative. The core finding: 91% of respondents believe that proprietary data and network effects are the definitive moat for software companies facing AI commoditization. Only 4% have not changed their investment approach. This is not a slow trend; it is a stampede.
For the uninitiated, PE secondary markets are where institutional investors buy and sell stakes in private companies before IPO. These players are the most sophisticated liquidity providers in the alternative asset universe. When they shift their capital allocation, the ripple effects hit venture capital, growth equity, and eventually public markets. The survey indicates they are rotating out of traditional software and into AI infrastructure, data services, and—implicitly—crypto assets that offer similar moats.
Core: The Technical Deconstruction of the Moat Thesis
The 91% consensus is a red flag. In any market survey, 50-70% agreement is normal; 91% signals either a leading question or a self-reinforcing echo chamber. Lazard likely framed the options to steer respondents toward the "data+network" answer. But even if the number is inflated, the direction is undeniable: investors are abandoning the old valuation metrics—MRR multiples, growth rates, net dollar retention—and replacing them with a fuzzy but potent mix of AI exposure risk and data asset quality.
Let me break this down from first principles. The old software moat was code. A proprietary algorithm, a complex workflow, a polished UI. AI has turned code into a commodity. LLMs can generate functional code in seconds. The differentiation now lies in the data that feeds the model and the network that amplifies its reach. This is technically sound for the short term. As I found in my 2025 audit of an AI-agent protocol, the oracle feed validation lacked cryptographic signatures—a perfect example of a network effect without a data integrity layer. The protocol had users, but the data was manipulable. The moat was a mirage.
Investors in the Lazard survey are betting that proprietary data is hard to replicate. They are correct, but only if the data is truly private, continuously generated, and legally protected. In crypto, we see the same dynamic: DeFi protocols that rely on price oracles have a moat only if they control the oracle nodes. Uniswap's moat is not its code—it's forked daily—but its liquidity network and the transaction data it generates. That data is public, but the network effect is sticky. The parallel is exact.
However, the survey misses a critical boundary condition: the time horizon of data moats. Transformer architectures are expanding context windows exponentially—from 4K to 1M tokens in two years. Synthetic data, federated learning, and inference-time reasoning are eroding the exclusivity of private datasets. In 2021, I spent 400 hours dissecting Luno's staking contract and found a reentrancy vulnerability. The team begged me to stay quiet. I published anyway. That experience taught me that what looks like a solid moat today can be a sieve tomorrow. The same applies to data: today's proprietary dataset can be tomorrow's public training sample.
Data does not lie, but it does not care. The 91% consensus is a lagging indicator, not a leading one. By the time everyone agrees on a moat, the moat is already being mined. The real value lies in identifying which data assets are genuinely defensible—those that are real-time, personally identifiable, or governed by regulation—and which are just noise. In my due diligence work, I evaluate data quality by three criteria: uniqueness, decay rate, and legal barriers. Most software companies fail on at least two.
Contrarian: What the Bulls Got Right
The contrarian angle is that the Lazard survey is correct in its direction but wrong in its magnitude. The 91% figure is a consensus, but consensus is the enemy of alpha. The true signal is the 4% who have not changed their methods. Those are the investors who understand that the old metrics—revenue growth, gross margin, churn—still matter, but they need to be weighted differently. A software company with mediocre data assets but a strong sales force and a loyal customer base may survive the AI disruption better than a data-rich company with no distribution.
Moreover, the survey implicitly assumes that AI will commoditize all software functions. This is false. Enterprise software requires reliability, compliance, and audit trails. AI models hallucinate; deterministic code does not. The "trust but verify" ethos of blockchain—where smart contracts are immutable and auditable—is a stronger moat than any dataset. In a world where AI can generate plausible but false outputs, the demand for verifiable computation will rise. That is a tailwind for crypto-native applications, not a headwind.
They built a palace on a fault line. The Lazard survey's palace is the narrative that data moats will save software. The fault line is the assumption that all data is equally valuable. In crypto, we have seen projects hoard on-chain data as if it were gold, only to find that the data is public, easily replicated, and worthless without a network effect. The same will happen in traditional software. The winners will be those who combine data with network effects and regulatory capture—the same formula that makes Bitcoin and Ethereum durable.
Trust is a variable you cannot hardcode. The survey's investors are trying to hardcode a new valuation framework. They will fail because the market is still in a "valuation vacuum"—the old model is dead, the new one is not yet standardized. This creates a window for contrarian bets. For example, I see value in vertical SaaS companies that own high-friction, compliance-heavy data (e.g., medical records, legal documents) because those datasets are not easily extractable by AI. They are regulated. They are sticky. The network effect is secondary.
Takeaway
The Lazard survey is a mirror for the crypto industry. The same forces that are reshaping software—AI commoditization, data as a moat, network effects as a defense—are reshaping blockchain applications. The protocols that will survive are those that combine proprietary data (e.g., order flow, identity attestations) with a verifiable execution layer. The ones that rely solely on code or hype will be picked apart by the market's cold logic.
Investors who are waiting for the AI dust to settle are making a mistake. The dust will never settle. The correct response is to lean into the uncertainty, apply forensic technical analysis, and bet on projects where the data is truly defensible and the network is genuinely sticky. The code spoke, but the logic was a lie. The truth is in the data. And the data does not care about your consensus.
