Broadcom's $60B Financing Play: When the Chip Vendor Becomes the Bank
October 6, 2026 · 5 min read
Broadcom is reportedly lining up $60 billion in debt financing to fund its customer's chip bill. The structure, per Bloomberg, Reuters, and CNBC: $42 billion in Class A senior debt underwritten by a bank syndicate, plus $18 billion in Class B junior debt led by Blackstone — which is putting $9 billion of its own money in. The arrangement matches a disclosure in Anthropic's prospectus: Broadcom agreed to make up to $42 billion available to Anthropic, covering about a third of its five-year, $125.2 billion TPU lease commitment. (via Capital Asia)
Read that again. The chip vendor is financing the buyer so the buyer can pay the chip vendor. Broadcom isn't just selling TPUs — it's arranging the money to buy them. When the seller has to underwrite the buyer's loan, the price tag has outgrown what the customer can pay out of pocket. This isn't a purchase order. It's a bank deal wearing a purchase order's clothes.
Vendor financing, at planetary scale
This playbook has a history, and it's not a comforting one. In the late 1990s, telecom equipment vendors financed carriers' network buildouts: sell the gear, then lend the customer the money to buy it. Demand looked unstoppable — right up until the buyers couldn't pay and the vendors took the write-downs. Broadcom's version is the same song at a vastly larger volume: pull demand forward today, park the credit risk somewhere it doesn't show up on your own balance sheet. It works beautifully until it doesn't, and "doesn't" always arrives on schedule — in the quarter the customer's revenue stops justifying the fixed commitments.
From selling chips to selling chips plus financing
The structure itself tells you how the AI infrastructure game has changed. Nobody writes a $60 billion check for one customer's compute bill, so you slice the risk into layers and find a buyer for each one. Class A senior debt — $42 billion — gets paid first and carries the least risk, which is why a bank syndicate can underwrite it and distribute it to institutional investors. Class B junior debt — $18 billion — absorbs losses first, so it pays more, and that's where Blackstone comes in: leading the tranche and committing $9 billion of its own capital as a signal to everyone else standing behind it. This is structured-finance machinery. Ten years ago a chip sale was a chip sale. Now the chip and the loan are one product.
The math that should make you pause
$125.2 billion. Five years. One customer. Broadcom's disclosed $42 billion covers about a third of it — so where does the other roughly $83 billion come from? That is the question the headlines skip. Anthropic's commitment doesn't flex: it has to be serviced whether or not AI revenue compounds fast enough to cover it. And this is just the Broadcom TPU line. Stack enough fixed, multi-year commitments across enough suppliers and the whole industry starts running on leverage — magnificent while demand compounds, unforgiving the quarter it doesn't. The uncomfortable constant in every leverage story: the debt is fixed, but the future revenue it bets on is not.
What to watch next
Three things. First, whether this becomes a template: if Broadcom can staple financing to its chips, expect every major accelerator vendor to offer the same package, and the contest moves from silicon to capital stacks. Second, the missing two-thirds — who funds the rest of the $125.2 billion, and on what terms, will say more about AI credit risk than any analyst note. Third, the timeline: five-year commitments assume five years of compounding demand. If inference prices keep falling faster than volumes grow, 2027 or 2028 could be this cycle's vendor-financing moment. One caveat to keep in mind: none of this is a Broadcom announcement. It comes via Anthropic's prospectus and press reports — the company itself hasn't confirmed the structure. The era of "just sell the chip" is over. Welcome to "sell the chip, arrange the loan, and hope the customer can pay."