The AI quality divide: Why most AI startups will fail the behavioral test
Bessemer Venture Partners just released their State of AI 2025 report, analyzing 20 high-growth AI companies across their portfolio and beyond. Buried in their comprehensive analysis is a revelation that challenges conventional AI investing wisdom.
Their data reveals two distinct categories of AI companies with fundamentally different business profiles. "Supernovas", which include breakouts like certain portfolio companies, average $40 million in annual recurring revenue within their first year of commercialization and reach $125 million by year two. These numbers represent some of the fastest growth rates in software history.
The concerning reality? These same Supernova companies operate at an average of just 25% gross margins, with some running negative margins as they "trade distribution for profit in the short term," according to Bessemer's analysis.
In contrast, "Shooting Stars" follow a more measured trajectory, reaching approximately $3 million ARR in year one while maintaining around 60% gross margins. Bessemer notes these companies "retain and expand customer relationships" and are "beloved by their customers."
As Daniel Kahneman and Amos Tversky demonstrated in their foundational research on prospect theory, humans systematically overweight dramatic, low-probability outcomes while undervaluing steady, high-probability gains. This cognitive bias, documented in their 1979 paper "Prospect Theory: An Analysis of Decision under Risk," creates predictable patterns in investment behavior.
In AI investing, this manifests as what behavioral economists call "salience bias", the tendency to focus on the most visible and dramatic examples rather than representative patterns. A 2019 study by Hirshleifer, Lim, and Teoh in the Review of Financial Studies found that investors consistently overweight attention-grabbing growth metrics while undervaluing sustainable business fundamentals.
The behavioral trap of explosive growth
The attraction to Supernova-style growth isn't just about numbers, it's about psychology. Kahneman and Tversky's prospect theory demonstrates that investors systematically overweight dramatic outcomes while undervaluing steady progression. In AI investing, this manifests as a dangerous obsession with viral adoption metrics.
When rapid growth creates lasting moats: The counterargument
To present a balanced perspective, we must acknowledge scenarios where Supernova-style growth can create sustainable competitive advantages. Network effects businesses like early Facebook or marketplace platforms often require rapid user acquisition to achieve critical mass before competitors establish alternative networks.
Some AI companies may follow similar patterns. Claude Shannon's information theory suggests that in winner-take-all markets, the first mover advantage compounds exponentially. If an AI product can achieve dominant market position quickly enough, even low initial margins may be justified by eventual pricing power.
However, our analysis of historical software businesses reveals that sustainable competitive advantages in AI require more than just user acquisition velocity. The key distinction lies in whether rapid growth creates genuine switching costs and network effects, or merely reflects subsidized customer acquisition.
This creates what I call "velocity addiction", the behavioral pattern where founders and investors become so focused on growth rate that they ignore the fundamental question: are we building something people truly cannot live without?
Real quality emerges from customers who would resist switching even if a competitor offered superior features. That level of integration takes time, patience, and a business model that prioritizes depth over breadth.
Applying the QAPITAL framework to AI opportunities
Our QAPITAL quality assessment framework reveals why most AI Supernovas fail the behavioral test for sustainable investment. Let me walk through each component:
Quality of earnings: The 25% gross margin warning signal
When we see AI companies operating at 25% gross margins, we're looking at businesses that haven't solved the fundamental value equation. They're subsidizing adoption through pricing that doesn't reflect true value delivery.
Quality AI businesses demonstrate what we define as sustainable competitive advantages:
Economic moats: Gross margins above 60% within 18 months, indicating genuine value creation rather than subsidized adoption
Customer integration depth: Evidence that switching would require significant workflow restructuring, not just product replacement
Pricing power validation: Ability to implement price increases without material churn, demonstrating value delivery exceeds cost
Margin expansion trajectory: Clear path to improved unit economics as scale increases and customer integration deepens
This definition extends beyond simple financial metrics to include behavioral indicators of true customer dependency. Warren Buffett's concept of "economic moats" provides the framework: sustainable competitive advantages that allow businesses to maintain above-average returns over extended periods.
The behavioral pattern we observe in low-margin AI companies is telling: founders are so anxious about competition that they're essentially paying customers to try their product. This isn't sustainable business building, it's venture-funded customer acquisition.
Predictability: Beyond the demo effect
AI products face a unique challenge: they can deliver spectacular demos while struggling with consistent real-world performance. The "demo effect" creates a behavioral bias where both founders and customers overestimate the product's reliability based on controlled presentations.
Quality AI companies show:
Consistent performance metrics across diverse customer environments
Growing usage depth rather than just user count
Customer renewal rates that increase over time as integration deepens
We've learned to ask a simple behavioral question: "What would happen if this AI product went offline for a week?" If customers could easily continue their workflows, you don't have a quality business—you have a nice-to-have feature.
Intrinsic competitive advantages: Real moats vs. model improvements
Here's where most AI investing goes wrong. Investors get excited about superior model performance while ignoring whether that advantage creates lasting competitive moats.
Model performance is easily commoditized. OpenAI, Anthropic, and others will continue pushing the frontier forward, making today's cutting-edge capabilities tomorrow's baseline. Quality AI businesses build advantages that compound independently of underlying model improvements:
Proprietary datasets that improve with customer usage
Integration depth that creates switching costs
Workflow embedding that makes the AI indispensable to daily operations
Network effects where value increases with user adoption
Team and behavioral assessment: The founder psychology test
The behavioral patterns of AI founders reveal crucial insights about long-term value creation potential. We've identified several psychological profiles:
The Technology Optimist: Believes superior AI will automatically create a superior business. Often struggles with practical go-to-market execution and customer retention.
The Growth Sprinter: Obsessed with viral adoption metrics. Makes pricing and product decisions that optimize for short-term usage rather than long-term value.
The Quality Builder: Focuses on deep customer problems and sustainable competitive advantages. Willing to grow more slowly in exchange for better retention and margins.
Quality AI founders exhibit specific behavioral traits: they speak more about customer problems than model capabilities, they prioritize retention over acquisition metrics, and they demonstrate patience with the business model development process.
The mispricing opportunity in AI investing
This behavioral analysis reveals a systematic mispricing in AI markets. Investors are bidding up companies with unsustainable growth while overlooking businesses building genuine competitive moats.
The opportunity exists because of three psychological biases:
Availability bias: Recent examples of explosive AI growth dominate investor attention, making Supernova trajectories seem more common and achievable than they actually are.
Status quo bias: Investors apply traditional SaaS metrics to AI businesses without adjusting for the fundamental differences in customer adoption patterns and competitive dynamics.
Narrative bias: The AI story is so compelling that investors focus on the technology narrative rather than business quality fundamentals.
For quality-focused investors, this creates systematic opportunities to:
Acquire stakes in AI Shooting Stars at reasonable valuations
Partner with founders who prioritize sustainable business building
Build concentrated positions in AI companies with genuine competitive moats
Practical assessment framework for AI opportunities
Based on our analysis of Bessemer's data and established investment frameworks, we've developed specific behavioral tests for AI opportunities. These complement traditional due diligence by examining the psychological and competitive dynamics unique to AI businesses:
The Substitution Test: Survey customers about alternatives if the AI product became unavailable. Quality products create dependency that extends beyond feature functionality to workflow integration. We look for responses indicating switching would require 3+ months of restructuring rather than simple product replacement.
The Pricing Power Test: Analyze customer response to price changes over time. Companies like Salesforce built sustainable businesses by gradually increasing prices as customer dependency deepened. Quality AI should demonstrate similar pricing flexibility without triggering churn.
The Integration Depth Test: Assess how deeply the AI product embeds into customer operations. Surface-level integrations suggest "nice-to-have" products; core workflow embedding indicates "must-have" solutions. We evaluate this through customer interviews and usage pattern analysis.
The Founder Conviction Test: Examine whether leadership optimizes for sustainable business metrics or fundraising-friendly growth indicators. This behavioral assessment reveals long-term value creation orientation versus short-term venture capital appealing strategies.
The long-term view on AI quality
The AI sector is experiencing a classic early-market dynamic where growth attracts more attention than sustainability. This creates temporary mispricings that quality-focused investors can systematically exploit.
Over the next 24 months, we expect to see:
Significant consolidation among AI Supernovas as unsustainable unit economics become apparent
Continued steady growth from AI Shooting Stars as their competitive moats strengthen
Increasing investor sophistication around AI business model evaluation
The winners will be companies that use AI to solve genuine business problems with sustainable unit economics, not those that achieve viral adoption through unsustainable pricing.
For founders building in AI: resist the temptation to optimize for venture-friendly growth metrics. Focus on building something customers truly cannot live without, even if it means growing more slowly than your competitors.
For investors evaluating AI opportunities: apply the same quality frameworks that work in traditional businesses. Technology may change, but the fundamentals of sustainable value creation remain constant.
Bottom line
The AI quality divide isn't just about different growth trajectories, it's about fundamentally different approaches to building lasting businesses. While markets remain distracted by explosive growth stories, quality-focused investors have a systematic opportunity to build concentrated positions in AI companies with genuine competitive advantages.
The behavioral patterns are clear: investors are systematically overvaluing velocity while undervaluing sustainability. That's exactly the kind of market inefficiency that creates superior long-term returns for those willing to think differently.
Sources and further reading:
Bessemer Venture Partners, "State of AI 2025" (2025)
Kahneman, D., & Tversky, A., "Prospect Theory: An Analysis of Decision under Risk," Econometrica (1979)
Hirshleifer, D., Lim, S. S., & Teoh, S. H., "Driven to Distraction: Extraneous Events and Underreaction to Earnings News," Review of Financial Studies (2019)
Buffett, W., Berkshire Hathaway Annual Letters (various years)
What behavioral patterns have you observed in AI investing? Share your perspective in the comments below.
For more insights on behavioral investing and quality growth opportunities, subscribe to The Behavioral Investor newsletter.