AI Business & Monetization
4 Types of AI Companies: Which to Build for Your Resources
Two startups pitched the same VC in 2022. First: 'We're building a foundation model for enterprise.' Second: 'We're building AI for lawyers - specific, with integrations, we know the industry.' The second got funded. The first closed a year later. Choosing the right company type beats ambition.
- Harvey AI: $0 → $3B in 2 years, started with one niche (legal research)
- Abridge: $50M ARR in medical documentation - GPT doesn't compete because of HIPAA
- CoreWeave: $100M GPU cluster → $23B valuation - right type at the right time
Предварительные знания
Four Types: Characteristics and Economics
**The AI industry divides into 4 clear company types.** Each has its own economics, entry barriers, and exit potential. Choosing the right type before building determines the entire strategic direction.
| Type | Capital | Margin | Defensibility | Exit Potential | Examples |
|---|---|---|---|---|---|
| Infrastructure | $100M+ | 55-70% | Physical assets, scale | IPO / strategic $10B+ | Nvidia, CoreWeave, Lambda |
| Foundation Models | $50M-10B | 20-40% | Model, data, brand | IPO / strategic $1-100B | OpenAI, Anthropic, Mistral |
| Horizontal Platforms | $1M-50M | 40-70% | Network effects, integrations | Acquisition $100M-5B | LangChain, Weights&Biases, Replicate |
| Vertical Applications | $0-5M | 60-85% | Domain expertise, distribution | Bootstrap / Acquisition $10M-500M | Harvey AI, Abridge, Cursor |
**Important:** types are not hierarchical - a vertical application can be more profitable than a foundation model. The choice is driven by resources, goals, and existing advantages - not ambitions.
Weights&Biases (MLOps platform) - what type of AI company is this?
Why Vertical Applications Often Beat Horizontal
**Harvey AI - legal AI.** Starts at $0, first client - Allen & Overy (top-10 global law firm). Two years later: $100M ARR, $3B valuation, clients paying $50K-500K/year. How did a niche product outperform horizontal AI tools?
| Company | Vertical | ARR/Revenue | Pricing | Why They Win |
|---|---|---|---|---|
| Harvey AI | Legal | $100M ARR | $50K-500K/year/firm | Legal data, workflow integration |
| Abridge | Medical | $50M+ ARR | Corporate contracts | Medical accuracy, HIPAA compliance |
| Nabla | Healthcare | $30M+ ARR | Per-provider license | Ambient AI notes, EHR integration |
| Glean | Enterprise Search | $100M ARR | $20/month/user | Corporate data, IT security |
| EvenUp | Legal (PI) | $50M ARR | % of settlement | Unique revenue model for the niche |
**The vertical limitation:** total addressable market (TAM) is smaller than for horizontal. If the entire industry is 50,000 law firms, that's the ceiling. But deep penetration in a niche is often more profitable than shallow coverage of a large market.
Harvey AI charges $200K/year per law firm. A generic AI writing tool charges $100/month per user. Under what conditions is Harvey more profitable as a business?
Decision Matrix: Choosing the Type Based on Available Resources
**Choosing the company type is the most important strategic decision.** Entering the wrong layer with insufficient resources = years of wasted effort. A decision matrix based on real case studies follows.
| Budget | Type | Target ARR (3 years) | Exit Strategy |
|---|---|---|---|
| $0 bootstrap | Vertical niche app | $100K-1M ARR | Lifestyle / small acquisition |
| $10K-100K | Vertical SMB app | $500K-3M ARR | Acquisition $1-10M |
| $100K-1M (seed) | Vertical enterprise | $2-10M ARR | Acquisition $10-50M |
| $1M-10M (Series A) | Platform or large vertical | $10-50M ARR | Acquisition/IPO $50-500M |
| $10M+ | Foundation model or infra | $50M+ ARR | IPO / $500M+ strategic |
**Real actionable advice:** without $50M in capital, the right move is a Vertical Application. Not because it's boring - but because Midjourney ($200M profit), Harvey AI ($3B valuation), Cursor ($100M ARR) all started in the Application layer without enormous capital.
A solo founder has $50K and 6 months. Background: experienced HR manager. What type of AI company is most realistic?
Key Ideas
- 4 types: Infrastructure, Foundation Models, Horizontal Platforms, Vertical Applications
- Vertical Applications: least capital needed, highest margins, most accessible for Indies
- Harvey AI case: niche = $200K/year/client vs generic = $100/month/user
- Decision matrix: budget determines type, type determines strategy
- With $0-1M - Vertical Application. No exceptions.
What's Next
Type chosen. Now the question of funding: bootstrap like Midjourney or raise VC? We break down the AI investment landscape.
- AI Startup Funding — Bootstrap vs VC: when to choose what
Вопросы для размышления
- How does available capital and domain expertise determine which AI company type is viable to build?
- Which verticals have the strongest potential for AI products, and what factors determine this?
- Harvey AI started with one niche (legal research) and expanded. What does the narrow-to-broad expansion strategy look like in practice?