Jason Cardillo
In this report

Research & Tools / Hypergrowth / Library

Reference

Data, research, and examples

Everything in this report rests on other people's primary work. This page credits it, links to it, and makes it findable. External links open in a new tab.

Credits

Company milestones come from company announcements and founder posts; from Sacra, and ARR Club; from TechCrunch, Forbes, CNBC, and The Information (via secondary coverage); and from teardowns by Product Growth, GTMnow, Growth Unhinged, Startup Riders, nrich, Postbeam, and others below. Cohort benchmarks are the work of Stripe, a16z, ChartMogul, Bessemer Venture Partners, Kimchi Hill, and Shearwater Capital. The era comparison draws on IntuitionLabs, TNW, MUFG, and Fast Company.

What this adds: 62 companies in one place; a label for each figure's reporting basis and source credibility (High = company-confirmed with dates; Medium = press citing sources or reputable estimates; Low = founder-only, contested, or retracted); and an interpolated months-from-$1M-to-$20M wherever two dated milestones allowed it, accurate to perhaps three months. Research completed in the first week of September 2026 and extended after peer review in mid-September. The fastest companies' figures change monthly; some rows are already out of date.

This is a structured collection of reported growth cases, not a predictive study. Most companies were picked after unusually fast growth had already become visible, the historical controls are themselves successful companies, and public revenue milestones vary in definition and credibility. The matrix below identifies recurring mechanisms and useful counterexamples. It cannot establish that any condition causes or predicts hypergrowth. Showing that would need matched companies chosen without knowing the outcome, scored on the same definitions, and followed over time.

A | Dataset

Filter by cohort, model, or credibility; search by name or category. Click a row for milestones, caveats, and sources.

CompanyCohortModelStarting point$100M ARRLatest reportedMo. $1M→$20MVerdictCredibility

B | Matrix

My reading of public reporting against the eight conditions, for supernovas, shooting stars, compounders, cautionary cases, and pre-AI controls. The alternate-mechanism value exists because a public artifact is only one way proof travels: Fin, EliseAI, Datadog, and CrowdStrike get credit for outcomes, references, practitioner mobility, ecosystems, and installed base. The last column is an explanatory variable, not a ninth condition.

Present Present through an alternate mechanism Partly or later Not evidenced
CompanySelf-propagating proofEconomics that expandStep-function opportunityMargin pathNew budget or buyer autonomyFounder as early channelProof before scaleDurable revenue evidenceDistribution at the start

The cautionary rows score strongly on several conditions associated with speed and weakly on the durability evidence. That separation is partly built into the archetype definitions, so treat it as a diagnostic illustration rather than a predictive finding.

C | Benchmarks

D | Teardowns

Product-led and creator-led

Founder-led enterprise and vertical

Infrastructure and marketplaces

E | Cautionary

Jasper

$42.5M year one, $75M peak, then a three-year pivot

Absorbed, then re-segmented to enterprise and AI-search visibility. No independent total ARR published since 2022. Sacra | Contrary | Enterprise ARR claim

Bolt.new

Fastest $0→$20M on record; silent since $40M

Speed without retention or margin. Sacra

Windsurf

$82M ARR, then broken up

Winner-take-most in quarters. Sacra

Character.AI

20M+ MAU, ~$30M run-rate

Usage without revenue. Sacra

Cluely

$7M ARR claim, retracted

The number was the marketing. Coverage

11x

$14M contracted vs ~$3M surviving

Break clauses and 70–80% churn. TechCrunch

Mercor

$2B gross; net about a third

A reminder that marketplace headlines are gross of payouts. Sacra

F | Controls

G | Era sources

The full comparison is on Then and now. These are the sources behind it.

H | Glossary

ARR
The annualized value of active recurring subscription contracts. It should exclude one-time revenue, non-recurring consumption, marketplace pass-through, and unsigned future business. Many private AI companies use “ARR” more loosely; I used company’s label but identified the reporting basis.
Annualized run-rate
The most recent month, week, or other short period multiplied to twelve months. It measures current velocity, not necessarily recurring or recognized revenue.
Contracted ARR
Annualized value of signed contracts, including those still inside trials or break clauses. Can overstate live revenue substantially.
Gross vs. net
Gross is total customer spend; net is what the company keeps after payouts. Marketplaces usually announce gross.
Outcome-based annualized revenue:
A run-rate based on completed outcomes such as resolved support cases. It may be durable, but it varies with activity and should not automatically be compared with contracted subscription ARR.
Supernova / Shooting star
Bessemer's archetypes. Its Supernova cohort averaged about $40M ARR in year one and $125M in year two, with roughly 25% gross margin; its Shooting Star cohort averaged about $3M, $12M, $40M, and $103M across years one through four, with roughly 60% gross margin and stronger customer durability. The shorthand of $100M in about 1.5 years is inferred from those cohort averages.
Compounder
Our label for the solid, often profitable company doubling a year or better on the classic path. Not a Bessemer term.
NRR
Net revenue retention: a cohort's revenue a year later, including expansion and churn, as a share of its starting revenue.
Step-function
A discontinuous jump in what the product can do, usually a new model release, that changes demand rather than nudging it.
Product-led floor
A self-serve entry point that produces users, revenue and qualified accounts without sales. Its absence is the most common structural blocker.
T2D3
Triple, triple, double, double, double: roughly $100M ARR in five years from $2M. Still what a good company looks like.

I | Reading

Books and essays this report leans on, as distinct from the data sources above. Listed because the reasoning borrows from them, not as a general reading list.

Playing to Win
A.G. Lafley and Roger L. Martin. The five-question cascade behind What to do, including the discipline of asking what would have to be true rather than arguing about whether a target is realistic.
Obviously Awesome
April Dunford. Positioning as a deliberate act, which is what sits behind the new name you give the buyer in the conditions.
Marketing 4.0
Philip Kotler, Hermawan Kartajaya, and Iwan Setiawan. The move from campaign-led to advocacy-led growth, which is the older frame for what this report calls self-propagating proof.
Growth Matrix
Elena Verna. A framework for evolving across product-, marketing-, and sales-led growth motions rather than treating the first working channel as the finished model.
PreviousMarketing was happening
↑ Library sections