Anthropic opened its books to people who might buy its shares, and the numbers on the pages did the talking.
The artificial intelligence developer told potential investors on October 15 that third-quarter revenue reached $16.8 billion, up from $11.6 billion in the second quarter and $4.7 billion in the first. The company said it was profitable in the quarter on an adjusted basis and generated positive cash flow.
The disclosures precede a listing that could come quickly. Anthropic may begin its initial public offering roadshow in the week of November 9, according to people familiar with the matter, with trading possible before Thanksgiving and a listing by the end of the year at the latest.
For a private company, sharing numbers with prospective buyers this way is standard practice before a listing. It is also a commitment. Once those figures are in the hands of investors, they become the baseline against which every later quarter is measured, and the direction of the line starts to matter more than the level.
The growth rate is the first thing analysts will examine. Revenue roughly doubled from the first quarter to the second, then rose about 45 percent in the third. A decelerating curve is normal as a business scales, and the question investors will ask is whether the slowdown reflects the size of the base or something softer underneath.
The cost side is where the detail gets thin. The Financial Times calculated that the company collects $1.70 of revenue for every dollar it spends on computing. That ratio is one of the few windows into the economics of a frontier lab, because compute is the largest single expense for anyone training and serving models at this scale.
A gross ratio of that kind excludes a great deal. It says nothing about sales and marketing, research salaries, legal costs or the interest on equipment financing. Two companies can report the same figure and end up in very different places after the rest of the income statement is applied.
The same is true of adjusted profitability, which is a measure the company defines for itself. Adjusted figures typically exclude stock-based compensation, restructuring charges and other items that a listed company must show. Positive cash flow is a harder number to work around, but it can also be flattered by the timing of payments from customers and to suppliers.
Investors are already looking past the reported quarter. The Wall Street Journal reported that investors expect the company’s annualized revenue to reach $110 billion by the end of the year. Reaching that figure would require the exit rate in the fourth quarter to be several times what the third quarter produced, unless the company books a large amount of contracted business at once.
Competition frames the arithmetic. Anthropic sells its models both directly to businesses and through cloud partners, and it competes with OpenAI for the same enterprise budgets, the same chip supply and, increasingly, the same investors. Both companies have described spending commitments that exceed their current revenue.
The listing calendar adds its own pressure. A roadshow in November means the prospectus, the risk factors and the audited financials have to be finished in weeks, and the pricing conversation happens against whatever the market is doing that week. Public investors, unlike the ones in the private meetings, can price a stock down on a single disappointing quarter.
That is the trade the company is making. Private capital has allowed Anthropic to grow without answering questions about gross margins in public. A listing converts those questions into a quarterly event, with a share price that moves in real time when the answers change.
Private investors have been able to value the company on contracted future business rather than on reported quarterly results, and the disclosure is a step toward the way public markets price a stock, where the multiple applied to earnings moves faster than the earnings themselves. Institutional investors who buy listings tend to apply the same lens they use for enterprise software: growth, retention and the cost of delivering the service.
Retention is the unknown that the disclosure does not address. Anthropic’s revenue comes from companies that buy model access through an API or through cloud partners, and those customers can move workloads to a competing model when prices or performance change. Switching costs in this market are lower than they look, because most developers keep the code that calls the model thin enough to swap.
Dependence on a handful of suppliers is the other structural fact the listing documents will have to describe. The company’s ability to serve customers at the price it charges depends on access to advanced chips, and that access runs through a supply chain controlled by a small number of vendors and constrained by export rules.
What the disclosure did not include is the mix of that revenue. Contracts recognized over years look different from usage billed monthly, and enterprise agreements that bundle support and access can carry costs that arrive after the revenue. Those details will surface in the listing documents.
The company’s message to the people in the room was straightforward: revenue is compounding, the cost of serving each dollar of it is falling, and the business is funded. The market’s response in November will decide how much of that story it is willing to pay for.
