Lazard’s Survey Is a Warning: AI Is Eating Software, but Crypto’s Moat Is Different

Yields were too good to be true, so we didn't. But Lazard's latest survey on private equity secondary markets just dropped a 91% consensus bomb. 91% of institutional investors now believe that 'proprietary data + network effects' are the only moats that can protect software companies from the AI onslaught. Only 4% haven't changed their investment approach. That's not a debate. That's a herd stampede. I've been in this space since 2017, hacking together scrapers to track early Uniswap whales. I've seen consensus form before—in 2020 when everyone piled into liquidity mining, in 2021 when NFT minting bots became the norm. But 91%? That's statistical anomaly. It means the market has already priced in a paradigm shift. The question for crypto is: are we part of the software being eaten, or do we have a different immune system? Let's break down the survey. Lazard, the investment bank, polled PE secondary market investors—the folks who buy and sell stakes in private software companies. Their conclusion: AI is a threat to software products, and the only safe harbors are companies with proprietary data and strong network effects. The rest? Capital is being pulled out. Moved to 'other investment opportunities.' Context matters. The secondary market is a leading indicator for primary valuations. If secondary investors are dumping software assets, primary GPs will follow. This means lower valuations, longer funding cycles, and a wave of distressed M&A. For crypto, this is déjà vu. We saw the same pattern in 2022 when Terra collapsed—capital fled from alts to Bitcoin, and the market repriced risk overnight. The difference is that this time, the trigger is AI, not a stablecoin depeg. But here's the core: the survey's 91% consensus is built on a traditional software framework. Software companies are being valued on ARR, churn, and NDR. AI disrupts that by commoditizing functionality. If a chatbot can replace a CRM's basic features, what's the moat? Data and network effects, they say. I ran the numbers. Over the past 7 days, a protocol lost 40% of its LPs. Why? Because its data moat was just a fork of Uniswap. No proprietary data, no sticky network. The LPs left for the next yield farm. Crypto protocols face the same brutal reality. A DeFi protocol's 'functionality'—swap, lend, borrow—is already a commodity. Uniswap v3 code is cloned a thousand times. The moat is not the code; it's the liquidity depth and the user base. That's the network effect. But crypto has an extra layer: token incentives. The mint button is a lever, not a purchase. You can buy TVL, but you can't buy loyalty. The survey's 91% are right to fear AI for software, but they're missing the crypto-specific twist. Let me give you a technical example. I audited Curve's early contracts in 2020—found an integer overflow in the fee calculation. That was a code-level risk. But the real risk was always the data: the pool allocations, the trading volumes, the LP behavior. Today, AI can analyze that data faster than any human. It can front-run, optimize, and extract MEV. That's not a threat to the protocol's code; it's a threat to the profit distribution. The protocol's moat is not the smart contract—it's the fact that the data is on-chain, transparent, and composable. AI can't take that away. It can only use it. So where does the contrarian angle live? The survey assumes that the software layer is the battleground. In crypto, the battleground is the settlement layer. AI can replace the user interface—a chatbot that executes swaps, or an AI agent that manages yield farming. But the underlying blockchain doesn't care. The moat is the trustless execution, the immutable ledger, the ability to verify without intermediaries. That's something AI cannot replicate. It can mimic, but it can't forge a consensus mechanism. However, there's a blind spot. The survey's 91% are focused on 'proprietary data.' In crypto, most data is public. The real proprietary data is the order flow, the private mempool, the off-chain negotiations. Intent-based architectures are moving MEV from on-chain to off-chain solver networks. That's where the AI threat becomes real. If a solver network uses AI to optimize matching, it could capture all the value. The protocol becomes a dumb pipe. The moat shifts to the AI model itself. I've seen this before. During the 2021 NFT minting chaos, I coded bots to mint Bored Apes. The gas wars were ego taxes. The real arbitrage was in predicting the minting mechanics. AI can do that now—better, faster, cheaper. The same is happening in DeFi. The question is not whether AI will be used, but who controls the data that feeds the AI. Let's look at the valuation implications. The survey's 'wait and see' attitude means capital is flowing to AI infrastructure, not software. For crypto, this means that protocols without a clear data moat will see their valuations compress. The 'old paradigm' of EV/TVL or EV/Revenue will be replaced by 'AI exposure discount' and 'data quality premium.' I estimate that the secondary market discount for software assets has already widened by 5-15 points. For crypto, we may see a similar divergence: protocols with strong on-chain data moats (like Uniswap's liquidity layers) will trade at a premium; forks with no proprietary data will trade at a discount. But here's the hidden insight: the survey's consensus is a lagging indicator. By the time 91% agree, the market has already moved. The real alpha is in the 4% who didn't change their methods. What do they know? Maybe they understand that software is not the only moat. Maybe they see that the biggest AI winners will be the data aggregators, not the model providers. In crypto, that's the indexers, the oracles, the data availability layers. They are the ones that will benefit from AI's insatiable hunger for structured data. Volatility is just fear wearing a disguise. The survey's 91% are afraid. They should be. But fear creates opportunity. The contrarian play is to look for protocols that are building AI-native features—not as a feature, but as a core. For example, protocols that use LLMs to route swaps across chains, or to automate risk management. These are the ones that will capture the AI premium. Let me give you a concrete example. I ran a local node during the Terra collapse. I saw the UST depeg 12 hours before the exchanges halted withdraws. The on-chain data was screaming. Today, AI can detect that pattern in milliseconds. The protocol that integrates AI for real-time risk monitoring will have a moat that no other can replicate. That's the data moat the survey is talking about, but applied to crypto. But the survey also ignores the cost side. AI requires compute. Software companies are seeing their margins compress because of inference costs. Crypto protocols can avoid that by using zero-knowledge proofs or optimistic rollups to verify AI inferences off-chain. The ZK rollup space is a perfect example: proving costs are absurdly high unless gas returns to bull-market levels. Operators are bleeding money. The survey's 91% don't think about that. They think about revenue. But the real moat is sustainable unit economics. So what's the takeaway? The survey is a warning, not a death sentence. For traditional software, the era of easy SaaS multiples is over. For crypto, the era of fork-and-pray is over. The next bull market will reward protocols that have built a defensible data moat, not just a token incentive. The mint button was a lever, not a purchase. The real purchase is the user's data and attention. I'm watching the following: AI-native protocols that capture proprietary order flow, data availability layers that become the default source for AI training, and compliance tools that verify AI behavior on-chain. If you're not building a moat based on data or network effects, you're building a feature. And features get replaced. Lazard's survey is a mirror. It reflects the market's fear. But in crypto, fear is the best fuel for innovation. The next wave will come from those who understand that the moat is not the code—it's the trustless, transparent, composable data that only a blockchain can provide. AI can eat software, but it will never digest the blockchain.