AI Business Models: What's Sustainable, What's Not —

Every AI startup claims "AI-powered" until they need a business model. Then reality hits.

Let's talk money:

The Models That Work

1. Usage-Based API

Pay per token, per minute, per call.

Pro: Scales with value delivered

Con: Customer fears unpredictable bills

Examples: OpenAI, Anthropic, Replicate

2. Subscription

Fixed monthly, unlock features.

Pro: Predictable revenue

Con: Hard to justify if usage is low

Examples: Notion AI, GitHub Copilot

3. Hybrid

Free tier (to prove value) + paid tier (at scale).

Pro: Converts after validation

Con: High free tier abuse risk

Examples: Most successful AI SaaS

4. Local + Support

Sell software, not service.

Pro: Predictable, one-time

Con: Support burden, updates required

Examples: Obsidian plugins, offline tools

The Models That Die

"We'll Figure It Out Later"

Revenue that doesn't exist yet isn't a business model.

Feature-First, Revenue-Later

Building cool AI features, charging later.

API Reselling

Marking up OpenAI API isn't sustainable. Gap shrinks.

What's Sustainable

For a solo/small AI business:

Local-first: Lower costs, higher margins

Solve real problems: Not "AI for" but "for + AI"

Start free: Let users validate before paying

Build audience first: Newsletter, content, community

The AI business crash (2025-2026) killed the hype-only startups. The survivors have revenue.

My Recommendation

Start with something:

You would pay for

You can deliver without bleeding

Has clear customer, clear problem

Then add AI. Don't start with AI.

---

Article 8 of 10 - AI Industry Series