Sales and Data Are the Real MOAT for Vertical AI Startups
Naoki Shibata (NSBF Wolf Capital) and Masa Tadokoro (Unicorn Farm) argue that proprietary training data—acquired through enterprise sales—is the only durable AI startup moat.
- Startups don’t build LLMs; the moat is proprietary fine-tuning data extracted from enterprise customers through high-trust sales relationships.
- Shibata’s investment test: customer #101 must produce significantly better AI output than customer #11—if accuracy doesn’t compound, a price war is coming.
- Early vertical AI founders must close deals that include data-sharing agreements, not just revenue—effectively strategic account acquisition from day one.
- Building an MVP is now cheap (Cursor, Windsurf); the scarce resource is years of industry credibility that lets a founder walk into a large enterprise and get data.
- Japan’s large enterprises only began setting AI budgets in late 2025 and are currently outsourcing blindly to Accenture—Shibata sees this as the startup entry window.
- Construction vertical AI example: Japan’s Zenshot and US-based OnSite IQ (NSBF Wolf Capital-backed) use 360° cameras to automate site-progress reporting, replacing high-cost supervisors.
- OnSite IQ’s core value is weekly transparent progress reports for commercial building investors, enabling milestone-linked financing and faster staffing decisions.
- Shibata publishes two vertical AI case studies per week on Note—one free, one paid—focused on unicorn-stage companies, not pre-revenue startups.
2025-11-26 · Watch on YouTube
