The numbers are stark. 96% of private equity secondary market investors have changed how they invest in software. 91% now believe that proprietary data and network effects are the only real moats. And most tellingly, capital is actively flowing out of software assets into other opportunities. These aren't predictions from a crypto Twitter thread. They come from a 2025 survey by Lazard, the investment bank, targeting the PE secondary market — the segment where institutional investors trade stakes in private companies. The survey signals a paradigm shift that has already been priced into the ledger of traditional finance. But the same structural forces are now reshaping crypto assets, and the market is only beginning to feel the tremor.
I have spent the last decade in the trenches of smart contract architecture — from reverse-engineering the 2x2 DAO's flawed voting logic in 2017 to stress-testing Aave v2's liquidation curves across 500+ simulations during the 2020 DeFi Summer. Each time, I saw the same pattern: the market chases the narrative first, and the math catches up later. The Lazard survey is that math. It reveals that the capital allocation engine has already decided that AI is not a future threat but a present pricing variable. The question for crypto is: which protocols are the software stocks of this cycle, and which are the AI-native assets that will thrive?
Context: The Survey's Mechanics and Its Crypto Reflection
Lazard's survey polled investors in the PE secondary market — a universe where liquidity is thin, information is asymmetric, and leverage is often brute. The finding that 96% of respondents have altered their software investment approach is not a soft signal. It is a hard capital reallocation. The 91% consensus on "proprietary data + network effects" as the defining moat is a direct challenge to the traditional SaaS valuation model based on ARR growth and gross margins. In crypto, the equivalent would be a shift from valuing a protocol by its TVL or fee revenue to valuing it by the exclusivity of its data and the stickiness of its user network.
Consider the parallel: in DeFi, liquidity is the data. A protocol like Uniswap has a massive data moat — its order flow, its routing intelligence, its user behavior patterns. But the market currently prices Uniswap's token largely on fee accrual and governance control, not on the proprietary value of its transaction data. The Lazard survey suggests that investors should start assigning a "data asset multiple" to protocols that accumulate unique, hard-to-replicate on-chain signals. Logic holds until the ledger bleeds. The ledger of traditional finance is bleeding into AI-native assets, and crypto's ledger will follow.
Core: Deconstructing the Data Moat Thesis Through a Crypto Lens
Let me apply the survey's framework to three crypto archetypes: the data oracle, the L1 settlement layer, and the AI-native dApp.
First, the oracle. Chainlink's network of node operators and its cross-chain interoperability protocol (CCIP) generate a continuous stream of unique, high-value data: price feeds, verifiable randomness, and proof of reserve attestations. This data is not easily replicated by a fork because the trust network and the integration agreements with hundreds of dApps constitute a network effect. The Lazard survey would classify Chainlink as a high-moat asset. Yet its token is currently priced at a discount to its potential, because the market still values it based on staking yields and speculation, not on the irreplaceability of its data pipeline.
Second, the L1. Ethereum's rollup-centric roadmap generates a massive amount of blob data post-Dencun. But that data is public and permissionless. The true moat is not the data itself but the composability ecosystem — the network effect of developers building on top of existing infrastructure. The survey's 91% consensus would suggest that L1s with strong developer network effects (like Ethereum, Solana, or Base) have a durable moat, but L1s that rely solely on performance or low fees without a differentiated data layer are at risk.
Third, the AI-native dApp. Projects like Bittensor or Akash Network are building decentralized marketplaces for machine learning. Their moat is not just data but also the network of compute providers and the models trained on that network. The survey's emphasis on "hard-to-replicate" data aligns with the thesis that decentralized AI protocols will outcompete centralized ones in the long run, because their data is both permissionless and verifiable.
"Trust is a variable, not a constant." In my audit of the Terra-Luna collapse, I watched the circular dependency between LUNA and UST destroy billions in value because the market trusted the algorithm's promise of stability. The Lazard survey reveals a similar trust breakdown: investors no longer trust that software companies can sustain their margins without a proprietary data moat. In crypto, the same trust breakdown is happening for protocols that lack genuine network effects. The capital is moving to assets that demonstrate a defensible data advantage.
Contrarian: The Blind Spots in the Consensus
The 91% consensus is itself a risk. When a signal becomes the consensus, it is already priced in. The real alpha lies in the counter-intuitive corner: the factors that the survey did not capture. For instance, the survey ignored the role of synthetic data. Generative AI can now produce realistic synthetic data at scale, potentially eroding the exclusivity of proprietary datasets. If a protocol's data moat is based on transaction history that can be simulated by a foundation model, that moat is not durable.
Second, the survey did not distinguish between different dimensions of network effects. In crypto, the most valuable network effects are not just user count but composability — the ability of smart contracts to interact atomically. An AI-native dApp that composes with existing DeFi protocols can leverage the network effect of the entire ecosystem, not just its own user base. Code compiles; people break. The code of composability is a moat that is harder to replicate than raw data volume.
Third, the survey's investors are moving capital out of software, but they are moving into other opportunities. The question is: are those opportunities AI infrastructure or are they non-tech sectors? If capital is flowing into AI infrastructure (GPUs, data centers, model providers), then crypto projects that are building decentralized compute or data storage will benefit. But if capital is flowing into energy or healthcare to avoid AI uncertainty, then crypto as a whole might face a liquidity drought. The survey's silence on this direction is a blind spot.
Silence is the only audit that matters. The silence in the Lazard survey is about the time horizon. Investors are changing behavior now, but they are not specifying when the AI disruption will hit. In crypto, the time horizon is even more compressed. A protocol that fails to integrate AI capabilities within the next two years will see its valuation gap widen. The market is already discounting that risk.
Takeaway: The Vulnerability Forecast and the Crypto Opportunity
The Lazard survey is a wake-up call for crypto project teams and investors. The traditional software valuation model is being rewritten by AI, and the same rewriting will happen for crypto tokens. The protocols that will survive are those with genuine data moats — not just tokenized user bases but proprietary, hard-to-replicate on-chain data that feeds into AI models. The protocols that will fail are those that rely on feature-based value, like a simple DEX clone or a generic L2 with no unique data pipeline.
My prediction: within the next 18 months, we will see a wave of token repricing as the market adopts the "data asset multiple." Protocols like Chainlink, The Graph, and Arweave will see their valuations expand as investors recognize their data moats. Meanwhile, generic DeFi protocols without network effects will be discarded. The capital that left software will eventually find its way into crypto, but only into assets that pass the data moat test.
The algorithm saw the crash, not the pain. The survey's algorithm saw the capital reallocation, but it could not see the pain of the software companies that will be left behind. In crypto, we have the chance to avoid that pain by building protocols with true data durability. The question is not whether AI will disrupt crypto — it already has. The question is whether we are building the assets that will survive the disruption.