As the AI race intensifies across Silicon Valley, venture capitalists are becoming increasingly sophisticated about evaluating which AI startups have genuine technical potential versus those destined for obscurity. Jenny Xiao and Jay Zhao, cofounders of Leonis Capital, shared their critical framework for assessing AI companies seeking investment.
Leonis Capital, founded in 2021, is currently deploying its second $40 million fund focused exclusively on next-generation AI companies. The firm has invested in over a dozen startups since launch, including MaintainX, Motion, and SpectroCloud. Their approach emphasizes deep technical understanding rather than surface-level enthusiasm about AI capabilities.
Xiao’s five critical questions focus on technical defensibility and market positioning. She asks founders what becomes possible with 10-20% model improvements, seeking founders who understand non-linear AI progress and capability thresholds rather than incremental features. Her second question directly challenges whether OpenAI, Anthropic, or Google could simply replicate the startup’s offering as a free feature—a fate that kills most AI companies.
The data question is particularly nuanced: Xiao doesn’t ask how much proprietary data exists today, but rather what data exists only because the product exists. This reveals whether the startup creates natural data moats through embedded workflows. She also tests defensibility by asking what would happen if competitors cloned the product in 30 days, distinguishing cosmetic advantages from structural ones like becoming a system of record.
Zhao’s questions probe founder psychology and strategic thinking. He asks what founders had to unlearn to see their opportunity, revealing intellectual flexibility. His question about what well-funded copycats would get wrong pushes beyond obvious answers about speed or data to uncover hidden operational advantages.
Both investors emphasize that most founders underestimate foundation models, and VCs underestimate them even more. The strongest AI founders, according to their Leonis AI 100 benchmark, build just ahead of the next technical breakthrough rather than chasing current capabilities. They look for founders who can hold unpopular positions, change their minds when evidence demands it, and think probabilistically about external forces beyond their control.
The framework reveals a sobering reality: 95% of founders give weak answers about competitive moats, typically claiming advantages in being “more vertical” or having “more niche data”—claims that rarely withstand scrutiny as foundation models improve and well-capitalized competitors emerge.
Key Quotes
Most AI startups don’t fail because they’re bad, but because they’re building something that OpenAI, Anthropic, or Google can eventually ship ‘for free’ as a feature.
Jenny Xiao, cofounder of Leonis Capital, explains the primary existential threat facing AI startups. This insight reveals why technical competence alone isn’t sufficient—startups must build where foundation model labs structurally cannot or will not compete.
We’ve noticed a consistent pattern: Most founders underestimate foundation models, and most VCs underestimate them even more.
This observation from Xiao highlights a critical blind spot in AI investing. The rapid improvement of foundation models means that advantages based on model quality or basic data accumulation erode faster than most market participants anticipate.
The best AI systems are opinionated by design, deliberately removing degrees of freedom and hard-coding assumptions about workflows, authority, or data flow.
Xiao challenges the common founder instinct to ‘stay flexible,’ arguing that the strongest AI products make deliberate constraints that create lock-in. This counterintuitive insight separates systems that become indispensable from those that remain easily replaceable.
AI moves too fast for people who can’t update, and we’re especially wary of founders who treat every pivot as vindication rather than correction.
Jay Zhao, Leonis Capital cofounder, emphasizes epistemic humility as a critical founder trait. In a rapidly evolving field, the ability to genuinely change one’s mind based on new evidence matters more than initial conviction.
Our Take
What’s striking about Leonis Capital’s framework is how it cuts through the AI hype cycle to focus on structural defensibility rather than technological novelty. The questions reveal an uncomfortable truth: in the age of rapidly improving foundation models, most AI startups are building on quicksand.
The emphasis on what foundation model labs structurally cannot do rather than what they’re currently not doing is particularly sophisticated. It acknowledges that OpenAI, Anthropic, and Google have virtually unlimited resources to replicate features—the only sustainable moats are those that would break their business models or require operational complexity that doesn’t scale.
The 95% failure rate on competitive moat questions suggests the AI startup ecosystem is far less mature than funding levels imply. Most founders are pattern-matching to previous startup eras where data advantages and vertical focus created defensibility, not recognizing how foundation models fundamentally change those dynamics. The investors who internalize these frameworks will likely capture disproportionate returns as the AI market consolidates around the few truly defensible positions.
Why This Matters
This framework matters because it reveals the brutal economics of the AI startup landscape where most companies will fail not due to poor execution but because they’re building features that foundation model labs will eventually commoditize. As AI capabilities advance rapidly, the window for startup differentiation narrows dramatically.
The insights are particularly valuable as venture capital flows into AI reach unprecedented levels, yet most investors lack the technical depth to distinguish genuine innovation from temporary advantages. Leonis Capital’s approach of benchmarking the top 100 AI startups provides rare transparency into what actually constitutes defensible positioning.
For AI entrepreneurs, these questions serve as a reality check against common delusions about data moats, vertical focus, and customer understanding—advantages that sound compelling but rarely survive contact with well-funded competition. The emphasis on structural advantages like becoming systems of record or embedding into compliance workflows points toward the few sustainable paths in an increasingly crowded market.
The broader implication is that AI startup success increasingly depends on understanding not just technology but the incentive structures, pricing models, and operational constraints of foundation model labs—knowledge that separates sophisticated founders from those simply riding the AI hype cycle.
Related Stories
- Meta Q4 Earnings: Zuckerberg Bets Big on AI with $135B Capex Plan
- OpenAI’s Competition Sparks Investor Anxiety Over Talent Retention at Microsoft, Meta, and Google
- Groq Investor Warns of Data Center Crisis Threatening AI Industry
- Big Tech’s 2025 AI Plans: Meta, Apple, Tesla, Google Unveil Roadmap
- Google Workspace’s Robert Frati Left for AI Startup Gemini Amid Tensions Over AI Strategy
Source: https://www.businessinsider.com/leonis-capital-investor-questions-for-ai-startups-2026-1