Ledger update: Capital is fleeing.
A new survey from Lazard, the investment bank, reveals a seismic shift in institutional thinking. 91% of private equity secondaries investors now identify proprietary data and network effects as the only durable moats in software. Only 4% have not altered their investment methodology. This is not a slow drift—it is a consensus that has formed with the speed of a flash crash. The crypto software sector, often overlooked in these discussions, is now standing on the same fault line.
Alpha dropped: Follow the money.
Context: The Lazard survey and its crypto mirror
Lazard's survey targets traditional PE secondaries—investors who buy and sell stakes in private software companies. Their sudden, near-unanimous agreement on data and network moats signals that the old valuation playbook (EV/Revenue, growth rates) is being discarded. In crypto, we have long operated under a different paradigm: open-source code, token incentives, and community governance. But the underlying software logic is the same. A DeFi protocol is a software company. An NFT marketplace is a software company. The AI wave that threatens traditional SaaS is now crashing into crypto's shores.
Why now? Large language models have commoditized code generation. Smart contracts can be written by GPT-4. The unique value of a crypto application is no longer its code—it is the data it accumulates and the network it sustains. The Lazard survey quantifies what many in crypto have felt intuitively: the era of 'code as moat' is over.
Core: The data-network feedback loop in crypto
Based on my experience auditing DeFi protocols during the 2020 liquidity trap, I saw firsthand that protocols with only code were fragile. Uniswap survived because its liquidity network created a data flywheel—each trade generated price data that made the next trade more efficient. That is a network effect. But the Lazard survey adds a new layer: AI now amplifies that data advantage.
Consider a lending protocol like Aave. It holds years of borrowing and liquidation data. An AI model trained on that data can predict default probabilities with higher accuracy than any generic model. That data is proprietary—not because it is hidden, but because the protocol's specific user behavior distribution is unique. No public dataset can replicate it. The 91% consensus is betting that such data moats will become the primary driver of valuation.
Risk Assessment: Protocols without data moats—simple DEX aggregators, generic yield optimizers—face a 40-70% valuation compression within 12 months. The capital rotation has already begun.
But there is a subtlety. In crypto, on-chain data is transparent. How can it be proprietary? The answer lies in the 'computation layer.' Owning the raw data is not enough; you need the AI infrastructure to process it. Protocols that combine on-chain data with private, off-chain models (e.g., using federated learning) will create a truly uncopyable moat. I have seen this in the AI-crypto convergence projects I analyzed in 2025—those with verifiable compute and exclusive data pipelines outperformed by 3x.
Contrarian: The consensus is a trap
Here is the counter-intuitive angle: the 91% consensus might be exactly wrong for crypto. The Lazard survey respondents are traditional PE investors. They think in terms of barriers to entry, switching costs, and data exclusivity. But crypto's nature is open, forkable, and composable. A network effect built on liquidity can be copied by a vampire attack. Data on Ethereum is public, so anyone can train a model on it. The real moat in crypto may not be data or network, but community governance and composability—things that AI cannot easily replicate.
The trap is sprung: the herd is running toward data moats, but the real alpha is in protocols that own the user relationship through AI-native interfaces.
Furthermore, the survey's high consensus (91%) is itself a contrarian signal. When everyone agrees, the marginal insight is already priced in. The opportunity lies in the 4% who did not change their methods—they may see something the others miss. Perhaps they understand that AI will democratize data analysis, eroding the very moats the 91% cherish. Or they recognize that in crypto, the most valuable data is not the transaction history, but the intent data from user interactions—something that is still poorly captured on-chain.