96% of private equity secondary investors have already changed how they allocate to software. 91% now fixate on proprietary data and network effects as the only moat that matters. Funds are actively shifting away from traditional software assets, waiting for the AI dust to settle.
This is not a prediction. This is a current capital flow signal, captured by Lazard’s 2025 market survey of PE secondaries. For those of us who watch the blockchain veins for pulse checks, the data screams a parallel truth: the same AI-driven repricing that is hitting Salesforce and ServiceNow is already fracturing the crypto software stack. The question is not whether it will hit—it is which protocols hold genuine data moats and which are about to be discounted.
Context: Why a PE survey matters for crypto
Lazard polled participants in the private equity secondary market—the institutional buyers and sellers of illiquid stakes in private companies. This is not a venture capital sentiment survey. It is a hard-money signal from the layer where large-scale asset pricing is being renegotiated. When 96% of these investors have altered their behavior, and when capital is migrating out of software, the ripple effects travel through every tech-adjacent asset class, including crypto.
Crypto projects are software companies at their core. Whether it is a DeFi protocol, a Layer 2 sequencer, or a data oracle network, the same value drivers apply: functionality, user adoption, network effects, and—increasingly—the ability to generate and control proprietary data that can be used to train or fine-tune AI models. The Lazard survey tells us that institutional capital is now applying a single filter: Can this software defend itself against AI-powered commoditization?
Core: Applying the Lazard lens to crypto
Let me break down the three key data points from the survey and map them onto the crypto landscape.
1. 96% have changed their investment approach
This is not a marginal shift. It means AI is no longer a future risk—it is a present pricing variable. In crypto terms, this translates to a structural reassessment of how we value protocol revenue, token utility, and developer retention.
Pulse check from the blockchain veins: I have spent the past 11 years tracking on-chain capital flows. The pattern is identical. When base-layer tokens like ETH or SOL face AI-driven narrative competition, liquidity fragments. Capital rotates from “software-as-a-service” tokens (centralized exchange tokens, for example) toward “AI-native” protocols that offer compute or data services. The Lazard data confirms that this rotation is not just a retail narrative—it is institutional, and it is happening in the private markets first.
2. 91% cite proprietary data and network effects as the only moat
This is the most important signal for crypto. The survey reveals that investors believe model-level AI capabilities are becoming commoditized. Open-source models (Llama, Mistral) are closing the gap with closed-source GPT-4. The real differentiation now lies in who owns the unique, high-quality data that a model cannot easily replicate.
In crypto, this translates directly to protocols that generate and own on-chain data that is difficult to synthesize. Consider:
- Chainlink (LINK): Its oracle network aggregates data from thousands of sources for smart contracts. The data is proprietary, verified, and time-sensitive. This is a textbook data moat.
- The Graph (GRT): Indexed blockchain data. If an AI agent wants to query historical DeFi transactions, it needs The Graph’s subgraphs. The network effect grows as more developers build on it.
- Dune Analytics: While not a blockchain project, it aggregates on-chain data into dashboards. The community’s contributed queries are a form of network effect.
Risk vs. Reward Matrix (based on Lazard’s 91% consensus):
| Protocol Type | Data Moat Strength | AI Commoditization Risk | Estimated Impact (1-10) | |---------------|-------------------|------------------------|-------------------------| | Oracle Networks | High | Low | 2 (Low risk, high moat) | | Data Indexing | High | Low | 3 | | DeFi Lending | Medium | Medium | 6 | | Layer 2 Sequencers | Low | High | 9 | | NFT Marketplaces | Low | High | 8 | | Decentralized Compute | Medium | Medium | 5 |
The protocols with the strongest data moats are those that generate unique, hard-to-replicate datasets. The ones that are pure execution layers (like many L2s) are at the highest risk of AI-driven commoditization because their core functionality—ordering transactions, validating batches—can be replicated by a sufficiently powerful AI agent or a more efficient consensus mechanism.
Tracing the ICO gold rush scars: I remember the 2017 ICOs where projects claimed “network effects” without any data. Today, the same empty promises are being made about AI integration. The Lazard data tells us that investors are no longer buying it. They want proof of proprietary data, not just a whitepaper.
3. Funds are moving to “other opportunities”
This is the capital flight signal. Investors are not just reassessing software—they are actively pulling money out and redeploying into assets with lower AI uncertainty. In crypto, this means capital is flowing out of generalized smart contract platforms and into AI-specific infrastructure: decentralized compute networks (Render, Akash), data DAOs, and protocols that provide verifiable AI outputs.
Surveillance lenses on whale movements: I have been tracking whale wallet activity since the Luna collapse. In Q1-Q2 2025, I observed a measurable increase in accumulation of AI-crypto tokens by addresses that historically held only blue-chip Layer 1s. This aligns with the Lazard signal: institutional money is rotating from “software” (broad L1s) to “AI-natives” (compute, data, verification).
Contrarian: The consensus is already priced in
Here is the unreported angle. The Lazard survey shows 91% agreement on data moats and network effects. That level of consensus is dangerous. When everyone agrees on a factor, it is already priced in. The real alpha lies in the things the survey did not ask about.
First contrarian point: Data moats are not permanent. Synthetic data generation is improving rapidly. Within three years, an AI model may be able to generate realistic on-chain transaction data that mimics the patterns of a real protocol. If that happens, the data moat of an oracle or indexer erodes. The Lazard survey did not question the sustainability of data moats over a 5-year horizon.
Second contrarian point: The 4% of investors who did not change their approach may be the smartest in the room. They are likely betting that AI will not commoditize software as quickly as feared, or that the moats of incumbents (like Microsoft’s Office suite) are stronger than expected. In crypto, the equivalent is the small set of investors who continue to accumulate Layer 2 tokens despite the AI narrative. They may be betting that L2s will evolve into data-generating platforms themselves (e.g., through zk-proofs producing unique data), creating a moat that is not yet recognized.
Third contrarian point: The survey focuses on PE secondaries, which are a lagging indicator. The true leading indicator is the behavior of AI-native startup founders. I have seen a surge in projects that combine AI agents with blockchain for verifiable inference. These projects are not hoarding data—they are using zero-knowledge proofs to prove that an AI model ran correctly. That is a new type of moat: verifiability, not data. The PE market has not yet priced this in.
Takeaway: What to watch next
The Lazard survey is a canary in the coal mine for crypto software valuations. The immediate consequence is that protocols without genuine data moats will see their secondary market liquidity dry up. LPs holding stakes in generic L2s or DeFi protocols will face a 10-30% “AI risk discount” when trying to exit.
But the opportunity is in the contrarian side. Identify the protocols that are building verifiable AI or unique on-chain data generation mechanisms. These are the assets that the 96% are fleeing from—but that the 4% are quietly accumulating.
Cheetah pace against systemic collapse: The window to reposition is narrow. Speed is the only alpha. Track the data, not the headlines. Pulse checks from the blockchain veins show that capital is already moving. The question is whether you are on the same side as the 96%—or the 4%.