Somebody in procurement is looking at a Claude commitment that runs past 2027 and trying to answer whether the vendor is good for it. The numbers available are unaudited: quarterly revenue passed $11.5bn in August, up more than fourteenfold year on year [12], and no audited figures have been published [13]. Fourteenfold growth on a quarterly line tells you new spend arrived. It does not tell you how much of last year's deployment is still running, or how deep usage goes inside the accounts that renewed.
What the financing sequence does tell you is the shape of the working capital need. A revolver is a standing line rather than a lump sum, drawn and repaid and drawn again, and for a company spending at Anthropic's rate it works as working capital, with the side benefit of telling public investors it will not need to raise again in a hurry [6]. One of Reuters' sources said Anthropic expects a tighter window than most between the analyst meetings and the document, because those analysts already know the business [19]. Firms normally leave a few weeks [4].
Size the facility against the revenue line. $11.5bn in a quarter annualises to roughly $46bn [22]. A $15bn line is about a third of that, or a little over one quarter of revenue [23]. You arrange a facility that size when the gap between cash going out and cash coming in has months of width in it.
Then there is the borrowing around it. Broadcom went looking for more than $60bn in debt in August to fund chips for the company [7]. Apollo and Blackstone shopped a $36bn chip financing in May [8]. AMD raised $4.75bn in its biggest ever bond sale [9], and a $1.3bn loan is building Anthropic a Texas data centre [10]. Add the named figures and more than $102bn of Anthropic-linked borrowing has been sought, shopped or raised, against $15bn the company would carry in its own name, close to seven to one [24]. TNW's read is that none of that sits on Anthropic's balance sheet, which is exactly why a revolver in its own name matters to people buying the stock [11]. It matters to people buying tokens for a different reason: the capacity your workload runs on is financed by counterparties you never signed anything with.
Two dates decide whether any of this should touch your paperwork. The first is the date your commitment prices. The second is the date figures with an auditor's name on them exist. If your integration would take a quarter to move to another model, ask for price-change notice and a capacity commitment in writing before you sign past the roadshow, and expect the vendor to charge you something for both. If you can swap providers inside a sprint, the listing calendar is trivia and the negotiation is better spent on unit price. Public investors are being paid to carry the uncertainty in those unaudited numbers; a buyer signing a long commit carries the same uncertainty with none of the upside. When the document does land, the drawn balance on that facility is the line to read before the valuation.
