Hook: A 91% Consensus That Feels Like a Funeral
In the quiet corridors of private equity, a survey landed like a stone in still water. Lazard’s latest report on secondaries investors revealed that 91% of respondents now believe the only durable moat in software is “proprietary data plus network effects.” Only 4% said they hadn’t changed their investment approach. The number is so lopsided that it stops being a data point and becomes a signal—a collective confession that the old way of valuing software companies is dead. I’ve been sitting with this number for a week, and it still feels heavy. A transaction is just a promise frozen in time, but this survey is a promise that the market is rewriting the rules of what a software company is worth.
Context: The Liquidity Map Shifts
To understand the gravity, you need to see the broader liquidity landscape. Private equity secondaries—where investors buy and sell stakes in existing funds—are the canary in the coal mine for institutional sentiment. When secondaries investors start pulling capital out of a sector, it means the forward-looking risk-adjusted return has deteriorated. The Lazard survey, conducted among a broad cross-section of institutional investors (LPs, GPs, intermediaries), captures the moment when the market stopped debating if AI would disrupt software and started debating how fast and which survivors. The global liquidity map in 2025-2026 is still awash in capital, but the vectors are shifting: money is flowing out of generic SaaS and into infrastructure, data layers, and AI-native platforms. This survey is the document that marks the boundary between two eras.
Core: The Valuation Framework Is Being Torn Down and Rebuilt
Let me walk through the mechanics. Traditional software valuation relied on a few sacred multipliers: EV/Revenue based on growth rate, gross margin, and net dollar retention. These metrics assumed that the software itself—the code, the features, the user interface—was the source of value. The Lazard survey says that assumption is no longer valid. The 91% consensus on “proprietary data + network effects” as the only moat is a direct attack on the old framework. It implies that code is now a commodity. The real value lies in the data that the software generates and the relationships it mediates.
Based on my own experience auditing ICO whitepapers back in 2017, I can tell you that the shift from “code value” to “data value” is not new. But the speed of this consensus is. In 2017, everyone was arguing about whether the protocol layer would capture value. Now, the market has decided: the protocol layer is a public good, and the application layer is about data monopolies. The Lazard survey is essentially saying that the only software businesses worth owning are those that have a defensible, proprietary data asset that a general-purpose AI model cannot replicate. This is a profound change in the capital allocation logic.
Let me illustrate with a concrete example. Imagine a traditional SaaS company that provides customer support ticketing. It has a nice UI, good workflows, and a $10M ARR. Under the old framework, it might trade at 8x revenue because of its 30% growth rate. Under the new framework, the investor asks: “Can an AI agent, powered by a foundation model, do this job without your software?” If the answer is yes, the company’s value drops to near zero. The only way to survive is if the company has accumulated a unique dataset of customer interactions that allows it to fine-tune an AI model better than any generic API. That’s the “proprietary data” moat.
But here is where the technical nuance gets interesting. The 91% consensus hides a critical assumption: that foundation model capabilities will continue to improve rapidly, and that the only differentiator will be data. This is a bet on model commoditization. It assumes that the model layer will become a race to the bottom, with a few oligopolists offering cheap, powerful APIs. In that world, the application layer’s only hope is to own the data that the model cannot see. I’ve seen this pattern before—in the early days of cloud computing, when everyone thought the only moat was network effects. But the cloud era also taught us that operational excellence and distribution mattered. The AI era may be different.
Contrarian: The Decoupling Thesis—Why the Consensus Might Be Wrong
Now, let me offer a contrarian angle. The 91% consensus is so strong that it feels like a bubble in itself. When everyone agrees on the same moat, the market prices it in, and the opportunity shifts to the blind spots. I see three overlooked factors.
First, the “reliability moat.” In enterprise B2B software, the biggest barrier to AI adoption is not data—it’s trust. Enterprises need deterministic, auditable, and predictable outcomes. General-purpose AI models are probabilistic and hallucinate. Software companies that can provide a reliable wrapper around AI—guaranteeing accuracy, compliance, and security—may have a moat that doesn’t depend on proprietary data. The Lazard survey ignored this entirely.
Second, the “distribution moat.” Acquiring customers is expensive. The 91% consensus focuses on data and network effects, but it forgets that many software companies have entrenched sales channels, integration ecosystems, and switching costs that are not easily replicated by AI. A CRM with 10,000 enterprise customers and 500 integrations is hard to displace, even if an AI-native competitor has better technology.
Third, the “model evolution risk.” The 91% consensus assumes that the current generation of LLMs (with their memory and generalization limits) will not be able to absorb proprietary data. But we are already seeing models with 1M+ token context windows, and synthetic data techniques that can infer private distributions. The timeline for “data moat” is uncertain. I’ve been studying this for years, and I suspect that within 3-5 years, the gap between public and proprietary data will narrow significantly. Investors may be overestimating the durability of current data moats.
Takeaway: Positioning for the Valuation Vacuum
We are now in a “valuation vacuum”—the old framework is dead, the new one is not yet standardized. This creates both risk and opportunity. The capital flowing out of software secondaries will find its way into AI infrastructure, data platforms, and security. But the smart money will look for the software companies that can survive the transition: those with data moats, yes, but also those with reliability moats, distribution moats, and the ability to pivot to AI-native experiences. The market is not crashing; it’s sighing. And in that sigh, there is a quiet invitation to rethink what we value. A transaction is just a promise frozen in time—but the promise we make today about software’s worth will echo for a decade.
