*The valuations, revenue figures and multiples below come from published 2026 reporting on companies that are not us. Our own position is disclosed at the bottom, in full.*
"Billion dollar company" gets processed as a quantity. It reads like something you accumulate, in the way a bank balance accumulates. That is not what it is. It is a price, quoted by a small number of buyers, for a slice of a company, on the basis of what they expect it to be earning some years from now.
So the useful first move is to convert the phrase into a figure a founder can hold in one hand and work against this quarter. The conversion is a division problem. The divisor got published, repeatedly, across 2026.
Three prices with both halves of the fraction visible
Most valuation coverage prints the numerator and skips the denominator, which is why the number stays abstract. These three were reported with both.
ElevenLabs raised 500 million dollars from Sequoia in February 2026 at an 11 billion dollar valuation, against roughly 330 million of annual recurring revenue at the close of the prior year. Thirty-three times.
Harvey took 200 million in March 2026, co-led by GIC and Sequoia, also at 11 billion, against a revenue line then reported near 190 million. Something close to fifty-eight times.
Lovable raised 330 million in December 2025 at 6.6 billion, with 200 million of ARR underneath it. Thirty-three times again.
Now perform the division almost nobody performs out loud. Price a billion dollar company off a multiple anywhere in that band and the revenue sitting underneath it works out somewhere between 17 and 43 million dollars, recurring, per year.
That is the entire billion. Tens of millions a year, booked on repeat, at a rate of increase that earns the upper end of the range rather than the lower.
That is the load-bearing line of the whole piece. Not because the outcome turns easy afterwards - it does not, and most attempts finish nowhere near it - but because it replaces an unimaginable quantity with a revenue line you can name, staff, forecast and check on an ordinary Wednesday. You cannot plan against a billion. You can plan against twenty or thirty million of subscriptions, because that has a shape: a customer count, a price point, a renewal rate, a sales motion. The billion is what somebody else does to that number afterwards.
The multiple is paid for the slope, not the level
The cleanest evidence for that claim comes from watching a single company get priced twice.
Lovable was bought at thirty-three times when its ARR stood at 200 million. Later, carrying 500 million, it was valued at 13.2 billion - twenty-six times. Revenue rose by a factor of two and a half. The price paid per dollar of that revenue fell.
ElevenLabs runs the same way. Thirty-three times against 330 million at the point the Series D was struck; by April its ARR had reached roughly 500 million, and measured against the same 11 billion price that is twenty-two times. Again the revenue climbed and the multiple came down.
Two companies, two independent observations, both pointing the same direction. And then the cross-check: Harvey holds the smallest revenue line of the three and carries much the highest multiple.
Nothing about this is irrational. The buyer is not purchasing the current revenue. The buyer is purchasing the rate at which that revenue is changing, and receiving the current figure as an accounting detail. A company still climbing steeply from a smaller base is a wager on where the curve terminates, and it is priced as a wager. A company whose curve has begun to flatten is a different instrument entirely - closer to something you value on its earnings, and valued accordingly.
Sit with what that inverts. The public scoreboard in this industry is a level: the MRR screenshot, the ARR milestone, the announcement that a round figure has been crossed. Every artefact of startup culture points at size. The pricing evidence says size is the less important half of the fraction, and that growing into a bigger number is, on its own, the thing that makes each dollar of it cheaper.
Which means a founder optimising purely for revenue this year is optimising the number that the market weighs least.
What a full trajectory looks like, dates included
The clearest published ladder of this cycle belongs to Anysphere, the company behind Cursor. Founded in 2022. Valued at 9.9 billion in June 2025, then 29.3 billion five months later in November. By the middle of 2026 it was running about 2.6 billion dollars of annualised revenue.
Read the calendar rather than the headline. Even the fastest such run anybody has recorded occupies about three years. The stretch this article concerns itself with - the first twelve months - is not the part that got reported. It is the part that happened before the numbers were worth reporting, and nothing about the later ladder is available without it.
Anybody telling you a billion arrives inside twelve months is describing something that has not happened in the published record. The honest horizon is measured in years. What is genuinely new is how few years, and how few people.
The specific thing that changed
Here is the figure that makes the rest of it reachable. AI companies arriving at a billion dollar valuation get there with a median of around 203 employees. For companies outside AI reaching the same mark, the median is roughly 414.
That ratio is the fact worth stopping on. Not the revenue - revenue at these levels has existed for decades. The change is how many people it takes to carry it. A revenue line that used to require a large organisation, with the hiring plan, the management layers and the burn that a large organisation implies, now sits inside a company half that size.
For a founder in year one this is the whole difference, and it is worth stating precisely. It does not mean revenue arrives more easily. Customers are no easier to find than they were, and the bar on the product is higher because everybody's product got better at once. It means the revenue level that corresponds to a billion dollar price - the tens of millions we divided out earlier - no longer implies a headcount you cannot afford before you have raised. The target became reachable by a team of the size that can actually exist at the start.
Year one, as a consequence rather than a wish
Four things follow directly from the arithmetic above. None of them is a promise about outcomes, and the base rate on all of it is unkind.
**Aim at the divided number, not the headline one.** The objective for a first year is a credible path toward tens of millions of recurring revenue, in a market where that ceiling exists. A market that tops out at three million of total spend cannot produce this outcome regardless of execution quality, and you can check that before you write code.
**Instrument the slope from the first invoice.** If buyers are paying for the rate of change, then measuring that rate is a second product, shipped to a second audience. Dated cohorts, retained revenue, month over month, kept honestly, including the months that went sideways. A steep slope you cannot evidence is worth roughly what an unevidenced forecast is worth.
**Start charging early enough for the rate to have a denominator.** A growth rate needs a starting revenue and a start date. Twelve months of unpaid pilots produce neither, and they produce no information about willingness to pay, which is the thing the whole edifice rests on.
**Treat a plateau as the expensive event.** The compression from thirty-three times to twenty-six, and from thirty-three to twenty-two, is the published cost of a curve easing off at scale. The same fact applies at a smaller scale in year one, where a flat quarter is cheap to fix and a flat year is not.
The parts that are not true, said plainly
None of these companies has listed on a stock exchange under its own name. What happened instead, in the one case that reached a full exit, was an acquisition: in June 2026 SpaceX agreed to buy Anysphere in an all-stock deal valuing it at 60 billion dollars, with every Anysphere share converting into SpaceX Class A stock. Reported as the largest startup acquisition on record, and expected to close in the third quarter.
That is worth reading carefully, because it is neither the IPO story nor the story I would have written a week ago. The route to public equity for this cohort ran through being bought by a company that had already listed. Not through listing. Any claim of a founder taking one of these businesses public on a one-year or two-year horizon still describes an event that none of them has produced.
The billion is a valuation. An investor set it by pricing a future they may have read wrong. It is not earnings. No cash entered anybody's account because of it, and no founder gets to spend it. Valuations are marked down as readily as they are marked up, and the same arithmetic that turns 30 million of ARR into a billion at a high multiple turns it into far less the moment the rate softens - which, on the evidence above, is what routinely happens as a company grows.
And the fastest published runs belong to companies with unusual founders, unusual timing and unusual access to capital. The arithmetic is general. The outcome is not.
Our own numbers, so nothing here can be inferred about us
What we run here are automated forecasting agents. The company is bootstrapped and pre-seed, it has a single founder, and there is no revenue to report. No ARR line exists on this side of the page, no valuation, no round, and no client whose figures I could quote even if I wanted to. Everything cited above belongs to other people's companies and other people's reporting, labelled that way deliberately.
We do publish one thing in the same category this article argues for: a rate, tracked over time, with the failures left in view. Predictions are committed before the event and scored once it settles, the misses sitting where anybody can count them against the hits. That running record is the only performance claim we make, and it is built to be the least flattering version available.
The arithmetic is the useful part to take away. A billion is a multiple applied to a revenue line, the multiple is bought with a rate, and the rate is built in the year when nobody is watching. Everything else is a press release about somebody else's round.
Educational content - not financial advice.