Every week we report figures: annualised revenue, margins, profits, valuations. They are all accurate and they do not measure the same thing. Some describe a reality, others a projection, others an accounting choice. Knowing how to tell them apart is what allows you to read an announcement without being spun a story.
This morning we reported a gross margin announced at over 80%. Here is how to place that kind of figure in context.
Annualised revenue: a projection, not a result
It is the most frequently cited figure and the most misleading. It is calculated by taking the last month's revenue and multiplying it by twelve.
Its value is real for a fast-growing company: last year's revenue says nothing about the current state of things. Its limitation is just as real: it assumes the pace holds, which is not a fact but a bet.
An annualised revenue of 30 billion does not mean the company took in 30 billion. It means it took in around 2.5 billion last month. The difference is considerable, and the phrasing erases it.
Gross margin: what is left after direct cost
It measures what remains of one euro of revenue once the cost of producing what was sold has been paid.
For an AI lab, that cost is essentially inference: the electricity and hardware needed to serve requests.
It is the most informative indicator of the health of the product. A high margin means selling a request is profitable. A low or negative margin means you are selling at a loss, which is only sustainable temporarily.
The trap: what you put under direct cost is a choice. Whether or not you count server depreciation, reliability teams, model fine-tuning shifts the margin by tens of points without changing anything about reality.
It deducts operating expenses from gross margin: salaries, research, sales, administration. This is what Anthropic announced in August at around 559 million dollars. It is a markedly more solid indicator than gross margin, and it still leaves out interest, taxes, and above all long-term commitments. A company can post a comfortable operating profit while having signed twenty-year infrastructure leases that do not yet appear in that calculation.
The three questions to ask in front of a figure
Is it reported or projected? A published quarterly result, a projection given to investors and an annualised pace are three different objects. The first is binding, the other two are not.
Who produced it, and for whom? A figure presented before a funding round or an IPO is aimed at someone you are trying to convince. That is not grounds for rejection, it is grounds for vigilance about the scope.
What is not in it? That is the most profitable question. An indicator is defined as much by what it excludes as by what it includes.
What the prospectus will change
All of the above explains why we keep saying that an IPO will be the most illuminating event of the year.
A prospectus is filed under legal liability. It requires publishing the exact scope of costs, the breakdown of revenue, customer concentration, and above all off-balance-sheet commitments — those leases and compute purchase contracts that appear in none of the indicators above.
It also contains a section where the company must honestly list what could go wrong. It is often the most instructive part of such a document.
What to take away
None of these indicators is dishonest. They describe different things, and the confusion comes from presenting them as interchangeable.
The simple rule: the more flattering a figure is, the more you need to look at its scope. A high gross margin is excellent news about the product and says nothing about the company. A spectacular annualised revenue describes one month. And an operating profit, even a real one, can coexist with commitments that exceed it tenfold.