Anthropic Now Sets the Funding Cost of Frontier Compute — And It's 10%

Until this week the AI-financing template ran in one direction: a chip vendor or a private-credit JV issued debt, bought the silicon, and leased it to the AI lab. Broadcom repeated that pattern on October 1 with a $42 billion convertible line disclosed in [Anthropic]'s S-1. Project North Star flips the script — [Anthropic] itself sits between the lenders and the SPV that owns the chips, and its own five-year lease payments service the bonds. The economic risk of the AI buildout is migrating up the stack toward the user, not down it toward the supplier.

The reported [Anthropic]-Blackstone-Broadcom package looks more like an investment-grade CLO than a venture loan. The economics are layered so each tier sees a different recovery claim on the chips, the lease cash flows, and ultimately the equity stub if [Anthropic] IPOs.

If a 7% rate is right for similar-risk public credit, the 300-basis-point gap Blackstone is collecting on its $9 billion has to come from somewhere. The answer sits in [Anthropic]'s S-1: Broadcom's $42 billion line is structured as a convertible that can be exchanged for [Anthropic] shares, and the prospectus flags that the notes aren't expected to be sold before the IPO closes. Translation — the entire package is priced as if the bonds will convert. The 10% coupon is the cash floor; the equity kicker is the upside.

Anthropic's $9 billion of junior cash and the broader $60 billion package don't stay at [Anthropic]. They flow through a special-purpose vehicle that buys the chips from Alphabet (designer) and Broadcom (co-developer and physical packager under a contract running through 2031), then drop them at data center sites operated by Fluidstack and Blackstone-owned QTS. The financing structure ties together three industries at once: cloud silicon, hyperscale real estate, and private credit.

The clearest casualty of Project North Star is the assumption that every frontier AI lab will rent its way to scale on NVIDIA GPUs financed by someone else's balance sheet. Project North Star shifts that one layer up: the lab now finances custom Alphabet silicon on its own credit. [NVIDIA]'s TTM revenue of $302.9 billion — with a 75% gross margin — is built on the assumption that AI labs cannot readily finance an alternative. The more $60 billion packages like this get syndicated, the more leverage AI labs have on total chip-build economics.

When [Anthropic]'s draft S-1 leaked on September 28, 2026 the numbers told one story: revenue grew 12-fold to nearly $4.6 billion in 2025, the operating loss hit $42 billion, and forward compute commitments already totaled $518 billion — 112× last year's revenue. Project North Star's 10%+ junior coupon is the second story. If the full $60 billion package gets priced near the current terms, [Anthropic] carries roughly $1 billion a year of junior interest alone (10% × $9 billion, plus fee carry). The S-1 is now read backward from the funding curve, not forward from the revenue curve.

Near-term, the equity-stub math drives the print. If [Anthropic] prices its IPO above $2 trillion and the Broadcom converts come in at a fraction of a percent equity dilution, the realized return on Blackstone's $9 billion blows past 10%. Long-term, the 144A marketability switch in 2027 is the larger lever — once the bonds trade publicly, the spread between [Anthropic]'s junior coupon and the public-market clearing rate becomes the AI funding benchmark.