Research & Tools / Hypergrowth / Conditions

Four patterns cover the product and its economics, four cover go-to-market. The Supernovas had nearly all of them at once. None is new to the AI era, and each has a pre-AI twin. That shows the patterns predate the current cycle; it does not prove that any one of them causes hypergrowth.

Read them as heuristics. These companies were selected because their outcomes were already known, and most of the controls are successful companies from an earlier era rather than matched non-winners.

Product

1. Self-propagating proof

A recurring pattern in the set, and the one most often defined too narrowly.

Sometimes the proof is a public artifact. Gamma marks every free export and attributes more than half its growth to word of mouth. Cursor grew on developers posting screenshots. Suno and Midjourney users publish the output as a matter of course.

But a public artifact is only one way proof spreads. Fin, EliseAI, Datadog, and CrowdStrike show that it can move through measurable outcomes, customer references, practitioners changing jobs, ecosystem partners, and installed-base expansion. Nothing is created that anyone would post, but the mechanism still works.

Pre-AI twin: Hotmail’s signature line, Dropbox’s referral storage, Slack’s cross-company invites. Figma, Canva, Notion, and Calendly all predate generative AI and run on the same mechanism.
Test: Name the specific thing that reaches the next buyer, and the path it travels. If you can’t, this condition is an assumption.

2. Economics that expand

The question is whether revenue can grow materially faster than sales capacity, through usage, outcomes, additional products, tier upgrades, deployment breadth, or larger contracts.

Per-seat pricing constrains this when customer headcount is the only multiplier, but it does not rule hypergrowth out on its own. Credit and consumption ladders let a single account grow 10-fold or 50-fold with no conversation: Lovable, Cursor, ElevenLabs, fal.ai, and Suno all work this way. Replit’s inflection followed a move from flat-rate to usage billing, which took gross margin from -14% to +23% in a year. In the enterprise the equivalent is expansion inside the account, where Synthesia reports net revenue retention above 140%.

Pre-AI twin: Atlassian without a sales force; Twilio priced by the message.
Test: Model your largest account growing 10x. If that requires a renegotiation, expansion is a sales project rather than a property of the system.

3. A step-function opportunity

The fastest curves bent on a factor the company did not control.

Agents became capable enough to build working software, which let Replit ship its Agent and repoint its go-to-market at citizen developers. ARR went from $10M at the end of 2024 to $100M by June 2025 (Growth Unhinged). Cursor reached roughly $4M annualized in April 2024, about $48M by October, and passed $100M in January 2025, though the public record does not pin that acceleration to a single model release. n8n compounded quietly for five years and then ran $40M to $100M in nine and a half months.

The opening is external. Whether the product, the infrastructure, the positioning, the proof, and the brand are ready when it arrives is not.

Pre-AI twin: Wiz and cloud migration; Deel and remote work; Zoom, whose revenue rose 326% from a much higher baseline when the pandemic made remote communication the default.
Test: Name the capability or market shift your plan is betting on, and say what happens if it lands six months late.

4. A margin path

Several of the fastest companies became profitable early or were unusually capital-efficient relative to their revenue. Gamma reports being profitable since 2023 with a team of roughly 52 at $100M ARR. Photoroom reached $20M ARR on about $2M invested, Rilla reports being cash-flow positive since 2022, and Midjourney took no outside money.

Replit improved reported gross margin from -14% to +23% alongside a move toward usage billing. Bolt reported roughly 40% gross margin before public revenue updates stopped. Higgsfield cut infrastructure costs 45% with a cloud migration and kept growing, but has never disclosed a gross margin. These cases make a margin path worth testing; they do not establish that margin caused the later outcome.

Pre-AI twin: Salesforce and Amazon eventually normalized weak early margins; Webvan and Kozmo did not.
Test: State the gross margin today and the margin the model assumes in 24 months, then name the mechanism that closes the gap.

Go-to-market

5. New budget or buyer autonomy

Either the buyer can act alone, or the money is being created rather than reallocated.

Product-led winners sold to a person with a credit card. Sales-led winners sold into budgets that did not previously exist. Sierra reached $100M in seven quarters and 40% of the Fortune 50 in 26 months. Harvey and Legora closed seven-figure law-firm contracts in year one. Glean’s $1M-plus segment nearly tripled during its run from $100M to $200M.

Displacement deals are the opposite case. They carry the incumbent’s renewal date with them, and a champion who has to defend the switch.

Pre-AI twin: Wiz selling into a cloud-security line that did not exist in 2019; Deel into global payroll nobody had funded before 2020.
Test: Ask where the money comes from. If the honest answer is another vendor’s contract, your ceiling is that contract’s renewal calendar.

6. A founder as an early channel

Anton Osika posts daily and uses Lovable’s growth numbers as content. Kareem Amin spent years on LinkedIn talking to a buyer who had no job title, then gave them one, GTM Engineer, and the people who adopted it became Clay’s distribution. Andrej Karpathy coined “vibe coding” in February 2025; Amjad Masad and Replit recognized it early, amplified it, and made Replit one of the products most associated with the behavior. Bret Taylor’s history got Sierra its first four design partners without cold outreach.

You do not always have to coin the language, but you do have to inhabit it credibly. Marketing can lead the language that gives buyers a name for what they are doing; whether the founder becomes an early channel remains the founder’s and CEO’s decision.

Pre-AI twin: “No Software,” “inbound,” “revenue intelligence.”
Test: Is the founder willing to be the primary channel for a year? If not, budget for the substitute, and expect it to be slower.

7. Proof before scale

Customer evidence precedes large investments in headcount and in paid distribution. The same sequence appears in enterprise and self-serve companies, which is why it is one condition rather than two.

Harvey launched with an exclusive at Allen & Overy (3,500 lawyers, roughly 40,000 beta queries) and a PwC alliance a month later; that was the go-to-market for a year. ElevenLabs built a self-serve-to-enterprise ladder so employees were already power users before procurement arrived. Cursor hired its first salesperson in late 2024, according to Sacra, and a President of Revenue much later.

On the paid side: Lovable spent about $2M on marketing on the way to $30M ARR, under 10% of first-year growth from paid, and turned on search and video only to accelerate a loop that already converted. Gamma put 70% of its creator budget into micro-creators on performance terms. Cursor reached several hundred million having reported no paid-marketing spend. Bolt ran under 10 people in go-to-market to $40M.

Pre-AI twin: Salesforce’s early logos; Wiz publishing its $100M milestone as a sales document.
Test: Work out which functions already drive the loop, then hire and invest to scale those. Adding headcount ahead of proof sets a cost base to a curve you have not demonstrated.

8. Durable revenue evidence

Retention you can measure, alongside an ARR number you can defend. This is the condition that separates the Supernovas that lasted from the Cautionary Tales, and it is invisible in the growth curve itself.

The public evidence is uneven. Some companies disclose direct measures such as paid-cohort retention or net revenue retention: Lovable reports 85% day-30 paid retention, and Wispr Flow reports 70–80% in year one. For others, continued acceleration after $100M and expansion within large accounts are encouraging but indirect signals; new-logo growth can produce the same revenue curve. Treat disclosed retention as evidence, later revenue growth as a proxy, and silence as unknown rather than automatically as failure.

Pre-AI twin: eBay versus Webvan. The same measurement, and only the speed at which the gap appears has changed.
Test: Can you produce 90-day cohort retention and a one-sentence definition of the ARR basis, today, without a project? If not, that is the first thing to build.

Friction

The mirror image, organized by the mechanism that slows a company down. Very few of these are absolute. Legora, Sierra, EliseAI, Abridge, and CrowdStrike all grew fast through one or more of them. Most excellent companies are on this list somewhere.

The buyer cannot self-serve

Committee purchase, procurement, security review, a pilot before a contract. Adoption runs at the buyer’s pace rather than the product’s. Abridge is the reference case: an excellent product, and years from pilots to deployments.

Proof stays private

Back-office automation, security, data infrastructure, regulated workflows. These companies need another path for proof to travel: quantified outcomes, trusted references, practitioner mobility, analyst validation, ecosystem partners, or installed-base expansion.

Displacement budget

Winning means unseating a line item, so the cycle includes the incumbent’s renewal date.

Regulated-domain friction

Healthcare, financial services, legal, and the public sector add validation, security, procurement, and deployment delay. This usually produces a Shooting Star or Compounder path rather than preventing exceptional growth. OpenEvidence took the other route and sold nothing to the institution at all: free to physicians, funded by pharma advertising.

Absorbable feature

The model vendor or the platform can ship what you do as a feature. Jasper and Windsurf. This blocker arrives after growth, not before it. Jasper’s answer was to move up-stack into AI-search visibility, a workflow the platforms do not own. That took a new CEO and three years.

Services-heavy delivery

Revenue that scales with people scales at hiring speed. Forward-deployed engineering sits on this spectrum. Mercor’s $2B is gross; net is about a third.

Usage without revenue

Character.AI carried more than 20 million monthly users on a roughly $30M run-rate. Or the reverse, revenue that does not retain. Both are visible within a quarter to anyone who asks for cohorts.

Your own go-to-market structure

Per-seat pricing with no expansion mechanic. Sales-led with no self-serve floor. Partner-dependent distribution. ABM as the only motion. A scorecard built on MQLs while the company needs a loop. If you are measured on form fills, your team will produce form fills.

The condition-by-company matrix →Marketing was already happening →