Two of the largest AI financings of the year landed within days of each other and took opposite routes. One
company sold stock to the people selling it chips. The other opened its accounts to everybody.
| Meta | Anthropic |
|---|
| Instrument | New stock: ~$35bn underwritten, ~$60bn at-the-market | IPO, Form S-1 filed 26 Oct |
| Total sought | Up to $95 billion | Share count not yet determined |
| Share class | New non-voting Class C | Class A common, ticker ANT |
| Notable buyers | Nvidia, AMD, Broadcom - up to $5bn | Public markets; GS and MS leading |
| Disclosure added | None beyond existing public filings | Full S-1: revenue, losses, compute, headcount |
| Context | 2026 capex guidance $135-145bn | Up to $120.5bn compute commitments |
The circularity in Meta's version
Nvidia, AMD and Broadcom taking up to $5 billion of a customer's equity raise, where that customer's $135 to
$145 billion capex flows substantially back to those same vendors, is a loop worth naming. It is not improper and
it is not hidden. It does mean the demand signal and the funding signal are partly the same signal.
Meta's choice of non-voting Class C shares for the larger $60 billion tranche keeps control
untouched while raising the money. Anthropic's route costs more in disclosure than in control - the S-1 obliges it
to publish revenue, losses and compute obligations every quarter from now on.
Which structure tells you more
If you are trying to judge whether AI capex is sustainable, Anthropic's filing is worth more than Meta's raise.
The filing shows what compute actually costs against what it earns - $5.6 billion of Q2 compute cost against a
$974 million operating profit. Meta's raise shows only that the money is available.
Sources
26 October daily brief. Anthropic newsroom. Yahoo Finance on the accounting.