AI Startup Valuations in 2025: The New Reality
Artificial intelligence companies — particularly those building on large language models, computer vision, or AI infrastructure — have seen extraordinary valuation multiples in 2024–25. Understanding how these multiples translate to a 409A context is essential for AI founders and their legal counsel.
AI Multiple Premium: Why It Exists
AI companies command premium multiples over traditional SaaS for several reasons:
- Market size: AI is perceived as a platform shift larger than cloud computing, with total addressable markets measured in trillions of dollars
- Growth rates: Leading AI companies have demonstrated revenue growth rates of 200–500% YoY, far exceeding any prior software category
- Winner-take-most dynamics: Data network effects and model quality advantages create strong competitive moats
- Strategic value: Big Tech (Microsoft, Google, Amazon, Meta) valuations are partially driven by AI positioning, pulling up private AI company comparables
Revenue Quality for AI Companies
Not all AI revenue is created equal in a 409A context:
- Annual subscription ARR: Highest multiple — predictable, recurring, similar to SaaS
- API consumption revenue: Medium multiple — variable but growing, benchmarked against public API companies
- Professional services / model customization: Lowest multiple — non-recurring, low margin, services business characteristics
- Government / enterprise contracts: Medium-high — large and sticky but concentrated
Infrastructure Cost Adjustments
AI companies face substantial GPU and inference infrastructure costs that affect margin and thus multiples:
- Gross margin for AI inference companies is often 40–60%, vs. 75–85% for traditional SaaS
- A lower gross margin justifies a proportionally lower ARR multiple
- The appraiser must normalize the multiple selection for margin differences vs. public comps
409A for LLM and Foundation Model Companies
Companies building proprietary large language models (pre-revenue or early revenue) present unique challenges:
- No revenue history — asset approach plus market comparables from funding rounds
- Value is primarily in model weights, training data, and team — intangible assets
- PWERM scenarios include: API monetization, enterprise licensing, acquisition by Big Tech, and model deprecation
- DLOM is typically lower than comparable revenue-stage companies due to high acquisition interest