The median public SaaS company trades at 6–8x forward revenue. OpenAI is valued at ~$157B on ~$3.4B ARR. That's 46x revenue. Anthropic sits at $61.5B. Perplexity crossed $9B in early 2025 with a fraction of those numbers.
This isn't irrationality. Investors aren't pricing AI companies on revenue — they're pricing them on infrastructure monopoly potential. The logic is specific, defensible, and importantly, falsifiable. Here's how it works.
AI Companies Valuation: The 2025 Snapshot
Current private market valuations for the leading AI companies, as of early 2026:
| Company | Valuation | Est. ARR | Revenue Multiple |
|---|---|---|---|
| OpenAI | ~$157B | ~$3.4B | ~46x |
| Anthropic | ~$61.5B | ~$1.5–2B | ~35–40x |
| xAI (Grok) | ~$50B | <$1B | 50x+ |
| Mistral | ~$6B | ~$50–100M | 60–120x |
| Perplexity | ~$9B | ~$100M | ~90x |
| Cohere | ~$5B | ~$100M | ~50x |
| Databricks | ~$62B | ~$2.4B | ~26x |
| Median Public SaaS | — | — | 6–8x |
Sources: PitchBook, Bloomberg, company disclosures. ARR estimates as of Q1 2026.
The 5 Drivers Behind AI Companies Valuation Multiples
Infrastructure Monopoly Potential
AWS took 15 years to build a $100B revenue business. Investors pricing OpenAI at 46x revenue today are betting it does the same for AI infrastructure. The multiple isn't on today's ARR — it's a probability-weighted bet on 2035 revenue. If there's even a 30% chance OpenAI becomes the AWS of AI, a $157B valuation is defensible math.
Strategic Capital Distorts Market Pricing
Microsoft invested $13B in OpenAI. Amazon committed $4B to Anthropic. Google put $300M into Anthropic and builds Gemini internally. When trillion-dollar hyperscalers buy into these companies, they're not doing venture math — they're pricing strategic optionality and defensive positioning. Their tolerance for premium pricing creates a floor that pure financial investors wouldn't accept.
Revenue Velocity Is Unprecedented
OpenAI went from $1B to $3.4B ARR in roughly 18 months. Anthropic went from zero to $1B+ in under two years. No SaaS company in history grew this fast at scale. When growth rates are 200–300% year-over-year, even a 40x multiple can look cheap on a forward basis if the trajectory holds.
Winner-Take-Most Platform Dynamics
Every developer who builds on GPT-4 is a switching cost. Every enterprise deal that standardizes on Claude creates stickiness. The first generation of AI-native applications — built on specific APIs — creates lock-in at a platform layer that's harder to dislodge than a point solution. Investors are pricing moat formation, not current cash flows.
Technical and Regulatory Moats Are Compressing
Training a frontier model requires $1B+ in compute, specialized talent pools measured in the dozens globally, and months of infrastructure buildout. NVIDIA GPU access alone gates entry. These barriers aren't permanent — but they give leading companies a 2–3 year window to cement distribution advantages before the technical moat compresses.
Where the Multiple Expansion Logic Breaks Down
The infrastructure monopoly thesis has a critical flaw: it assumes one or two winners. The 2022–2025 SaaS correction happened because growth slowed and rates rose. AI faces a different risk — commoditization of the core model capability.
Model Commoditization
If Llama 4, Gemini, and Claude converge on benchmark performance, the moat shrinks to distribution and pricing — much less defensible than technical superiority.
Inference Cost Collapse
API pricing has dropped 80%+ in 18 months. Cheaper inference is great for adoption but terrible for revenue per query. Revenue growth must come from volume, not price.
Open-Source Disruption
Meta's open-source strategy commoditizes the model layer. Any company with a business model predicated on charging for model access faces sustained pressure from free alternatives.
Customer Concentration
Many AI startups have 1–3 hyperscalers or large enterprises as 80%+ of revenue. That's a venture round, not a business. Multiple concentration risk is severely underpriced.
How to Think About AI Valuation as an Investor
The framework I use when evaluating AI companies valuation today: don't compare to SaaS multiples, compare to platform infrastructure investments.
- →Does the company own a proprietary data advantage that compounds over time? (Scale AI does. A wrapper does not.)
- →Is the company building the infrastructure layer or the application layer? Infrastructure warrants infrastructure multiples.
- →What's the customer acquisition dynamic — pull (developers seek you out) or push (sales-heavy)? Pull compounds; push doesn't.
- →Who are the strategic investors and what's their incentive? Strategic capital at high valuations creates a floor but also misaligned incentives at exit.
- →What does the revenue look like under model commoditization? If the moat is the model, test what happens when the model is free.
Track live AI company valuations, revenue multiples, and funding rounds on the AI Valuations Dashboard at Value Add VC.
AI companies valuation multiples aren't a bug in investor logic.
They're a bet that the foundation model layer becomes the most valuable infrastructure in the history of software — and the premium reflects the probability weight on that outcome.
Monitor AI company valuations and funding rounds in real time on the AI Valuations Dashboard. Originally published in the Trace Cohen newsletter.