Anthropic's $11.5B quarter turns on whether cloud-partner sales are booked gross or net

Two revenue numbers are in circulation and they do not describe the same thing. A quarter of $11.5B annualizes to $46B ^[1], while the run rate quoted for the end of July, in figures cryptobriefing.com reports are being shared with prospective backers, was above $65B ^[1], which leaves roughly $19B of the headline figure resting on whatever July alone did ^[2]. That implies about $5.4B of revenue in July against a second-quarter average nearer $3.8B a month ^[3]. Annualizing an exit month is a normal thing to do when the growth is real and the cash is collected; it is also the presentation that flatters fastest, and it matters because the multiple hangs off which denominator the book uses. At a $2 trillion price, the candidates are about 31 times the run rate and about 43 times the annualized quarter ^[4], and the whole distance between them is a choice of window.

The reseller question is a second lever on the same denominator. When Claude is sold through a partner, booking gross means the full amount the customer pays and booking net means Anthropic's share after the cloud provider's cut ^[10], and prospective IPO investors are pressing on which is being used ^[9]. If a slice of that revenue is presented gross and later restated to net, revenue falls while operating profit does not, because gross presentation adds the provider's cut to revenue and to cost of revenue in equal measure ^[8]. So the $559M reported for the second quarter is the durable figure and the 4.9% operating margin it implies against $11.5B ^[7] is the fragile one, which also decides whether the comparable set is software or the distribution of someone else's compute. Enterprises are about 80% of revenue ^[11], and many of the largest reach Claude through AWS Bedrock or Google Cloud's Vertex AI ^[13].

Internal projections put 2028 revenue at $190B to $200B ^[8], which is 2.9 to 3.1 times the current run rate and 4.1 to 4.3 times the annualized quarter ^[5]; at the top of that range, $2 trillion is 10 times 2028 revenue ^[6]. Against that, the same report notes fewer than four quarters of meaningful history from which to model churn, seasonality or customer concentration ^[20]. The $65B Series H closed in late May at $965B post-money ^[6], which is between 50% and 55% of every private dollar the company has ever raised ^[12], so the marginal private holder is a few months old and needs a print above its own mark, and $2 trillion is 2.07 times that mark ^[13].

The counter-thesis is respectable. The cash is the same cash whichever line it lands on, sequential growth of 2.43 times ^[14] is not something a presentation convention manufactures, and the extrapolation to 2028 is a larger hole in the book than the bookkeeping. What would break the accounting thesis is a filing that shows channel revenue as a modest share already presented net, at which point pricing goes back to arguing about the trend line. None of these figures sit in a public document yet, since the June draft was confidential and the public version is only anticipated for late September ^[15]. The denominator becomes checkable then.