← Home
The Twenty Minute VC · August 6, 2024 · 24m

Why 90% of AI Startups Will Fail

Stebbings argues that the AI startup landscape is repeating the 2021 SaaS bubble: too much capital chasing too few defensible businesses. He estimates 90% of current AI startups will fail — not because AI doesn't work, but because they have no moat.

Canon

Stebbings observes that AI startups built around GPT-3's limitations (hallucination correction, context window management) were killed when GPT-4 fixed those limitations natively. Each model improvement resets the baseline and destroys the companies built on the old one.

Highlights

Most AI startups are 'thin wrappers' around foundation models — and thin wrappers have no moat
Stebbings estimates that 70% of AI startups are building products that could be replicated in a weekend by anyone with API access to GPT-4 or Claude. Without proprietary data, unique workflows, or distribution advantages, these companies have zero defensibility.