A $10bn Nvidia anchor would fill a tenth of Anthropic's book before demand is tested

An anchor allocation is a fixed slice of an offering, bought before the book is marketed to anyone else ^[4]. Up to $10bn into a raise of up to $100bn is a tenth of that book ^[1], and at a valuation near $2 trillion the same $10bn buys half a percent of the equity ^[2]. According to cryptopolitan's account of the Reuters reporting, Nvidia would enter not only as a financier but as a supplier to one of its major customers ^[7].

The November commitment and the IPO cheque may be the same money. Nvidia's IPO investment could be that partnership money routed into an offering, or a fresh $10bn on top of it. The difference is $10bn.

Anthropic owes its suppliers much more than either figure. In April it said it planned to spend more than $100bn with Amazon Web Services over ten years and to use more than 1 million Trainium2 chips ^[10], and it has agreements with Google and Broadcom for several gigawatts of TPU capacity ^[11]. Set $10bn against the $30bn Azure commitment plus the $100bn AWS plan and it covers about 8% of $130bn ^[3]. Anthropic is also building an internal chip team working on custom hardware for Claude, to get more control over computing costs ^[16].

The multiple depends on which revenue figure you divide by. Cryptopolitan reports an annualized run rate above $65bn by July against $9bn in December, a factor of about seven in seven months ^[13][8]. Against that run rate, $2 trillion is roughly 31 times revenue ^[4]. Against the $190bn to $200bn for 2028 that Reuters earlier reported Anthropic projects, the midpoint is closer to 10 times ^[15][5], and part of the $2 trillion figure rests on those projections ^[14]. The May round priced the company at $965bn post-money ^[12], so the IPO talk is about 2.1 times a valuation set a few months ago ^[6].

I think a pre-committed anchor makes the eventual print less informative than its size suggests, and there are two ways that view fails. An anchor buys at whatever price the book produces, so it takes size off the book at a level someone else sets. And half a percent of a customer is small enough to look like ordinary strategic money from a firm that did the same for Arm. The more interesting version is the denominator: $100bn is about 73% of the $137bn that US listings excluding SPACs raised through the end of August, on Dealogic's count ^[17][7].

Either a raise that prices well under $100bn or a book that fills without an anchor at all would break the thesis. The listing is forecast before the November midterm elections ^[18].