Original research
The AI-Native Valuation Benchmark
AI broke the comps. When a target's edge is a model, a dataset, or a workflow that did not exist three years ago, the comparable set is thin and the range is too wide to act on. This is how AI-native companies are actually priced, and where buyers over and under-pay.
The headline finding
20x to 225x
Across roughly two dozen late-stage AI-native companies, public revenue multiples ran from about 20x to 225x. When the same business model trades across an 11-fold spread, comparable-company analysis cannot return a number a buyer can act on. That dispersion, not any single multiple, is what breaks the comps.
Range compiled from public AI-startup revenue multiples, Palle Broe (palle.substack.com/p/from-19x-to-225x-the-wide-range-of), corroborated by CB Insights State of AI Q2 2025 and Aventis Advisors; figures are directional, not audited.
Method
How the benchmark was built
This is not a proprietary dataset of measured deals. It is a synthesis of public market evidence, named and cited on each card where a card carries a number, read through the judgment Tomasz Felpel applies as the founder of Value Alpha and a former Fortune 500 M&A operator who has priced more than 1,000 private companies. The sourced figures come from Carta, PitchBook, CB Insights, Crunchbase, and advisory analyses of Crunchbase data; the directional cards, marked with a premium or a discount rather than a percentage, reflect professional judgment on factors a standard multiple cannot see in an AI company: who owns the data, how defensible the model is, displacement risk from the next foundation-model release, and how much of the value walks out the door with a few key researchers. Where no defensible public number exists, the card shows a direction and a driver, not a fabricated precise figure.
The findings
What moves an AI-native valuation
+38% to +193%
AI-ready premium
AI startups out-priced non-AI peers at every venture stage in 2025, with the premium widening from Series A to Series E+. Source: Carta, State of Private Markets: 2025 in Review
Discount
AI-exposure discount
When AI threatens the core business, public software de-rated hard, with median SaaS EV/Revenue compressing toward the low single digits on disruption fears. Source: Meritech Software Pulse; multiples.vc, early 2026
20x to 225x
Comps gap
Reported revenue multiples across late-stage AI companies span an 11-fold range, so a comparable set returns a band too wide to act on. Source: Compiled public AI multiples (Palle Broe); CB Insights Q2 2025
Premium
Proprietary-data premium
Owned, hard-to-replicate training data is the most durable moat an AI company can hold, and it lifts value above the sector median by judgment, not a published multiple.
Discount
AI-washing discount
Roughly 40% of European companies labeled AI startups showed no material AI, and the SEC has since brought enforcement actions, so unverified AI claims earn a markdown. Source: MMC Ventures, State of AI 2019; SEC press release 2024-36
Discount
Displacement-risk discount
A foundation-model release can erase an applied-AI moat overnight, as when Jasper cut its own valuation about 20% after ChatGPT launched. Source: Maginative; Contrary Research, Jasper profile
Reset
Revenue-quality reset
The AI premium is real but compressing as AI becomes table stakes, with one tracked segment's premium falling from +86% in 2024 to +22% in 2025. Source: PitchBook, Q4 2025 E-Commerce AI Premium
Premium
Strategic-acquirer premium
A strategic buyer pays up for capability it would otherwise build, but that raise-stage premium does not survive an outright control sale, where offers come in materially lower.
Discount
Talent-dependency factor
When value rests on a few key researchers who can leave, the asset is fragile and the price reflects that key-person risk.
The read
Why AI breaks the model
A standard valuation leans on comparable companies and precedent transactions. AI-native targets have neither in clean form: the comparable set is too young, the business model is often unproven, and the moat is a model or dataset that a single foundation-model release can strengthen or erase. The market evidence shows both sides of that knife at once. AI commanded a valuation premium over non-AI peers at every venture stage in 2025, widening from roughly +38% at Series A to +193% at Series E+ in Carta's data, while down rounds hit a multiyear high and clustered in AI and machine learning. Big markups and big markdowns in the same cohort are not a stable multiple. They are a signal that these companies are being priced on conviction, and the conviction is what a buyer is really paying for.
The factors that actually separate an AI-ready company from an AI-exposed one are not on a multiples table: who owns the data, how defensible the model is, and what a foundation-model release does to the moat next quarter. That is the judgment a buyer is really paying for. If you are pricing an AI-native target, that is exactly the second opinion I give.
The full benchmark
Get the complete AI-Native Valuation Benchmark
The full breakdown: every premium and discount, the methodology, and how to price an AI-native target when the comps fail. Sent to your inbox, with the next edition when it drops.
For media & citation
Cite the benchmark
Cite as: Felpel, Tomasz. The AI-Native Valuation Benchmark, 2026. tomaszfelpel.com/ai-native-valuation
Media and data requests: [email protected]. Available for commentary on how AI is repricing private companies, on deadline.
Built by Tomasz Felpel, founder of Value Alpha, the AI valuation platform, and a former Fortune 500 M&A operator who has priced 1,000+ private companies.
Pricing an AI target?
Get a defensible number the comps can't give you
If you are buying, selling, or marking an AI-native company and the comparable set is failing you, I give an independent, defensible value range with the rationale, the moat assessment, and the displacement risk spelled out.
FAQ
Questions, answered
Why are AI companies so hard to value?
Standard valuation leans on comparable companies and precedent transactions, and AI-native companies have neither in clean form. The comparable set is too young, the business model is often unproven, and the moat is a model or dataset that a single foundation-model release can strengthen or erase. So comparable analysis produces a value range too wide to act on, and the deal ends up priced on conviction instead of evidence. The factors that actually separate value, who owns the data, how defensible the model is, and displacement risk, are not on a multiples table.
What is the AI-ready premium and the AI-exposure discount?
The AI-ready premium is the uplift a company earns when AI is a defensible advantage: a proprietary dataset, a hard-to-replicate model, or a workflow with real switching costs. The AI-exposure discount is the markdown a company takes when AI threatens its core business, when the thing it sells is the thing a foundation model now does cheaply. The same sector can contain both, which is why a single sector multiple is misleading for AI companies.
How do you value an AI startup with little revenue?
With little revenue, an earnings or revenue multiple says almost nothing, so value is driven by the strength and defensibility of the asset (data ownership, model edge, talent), the size and timing of the opportunity, and what the capability is worth to a strategic acquirer who would otherwise have to build it. The honest output is a range with the assumptions visible, not a single confident number, and the buyer type changes the answer more than any multiple.
Can I cite this benchmark?
Yes, with attribution to Tomasz Felpel and a link to tomaszfelpel.com/ai-native-valuation. For the underlying breakdowns, a sector cut, or commentary on deadline, contact [email protected].
More: the Private Company Valuation Index · valuation calculator · Valuation 101