← All posts / Industry

The Lender Is Also the Supplier: Broadcom's $42 Billion Convertible Loan Anchors Anthropic's $125 Billion TPU Bet

Anthropic's IPO prospectus reveals Broadcom has agreed to lend the AI lab up to $42 billion via convertible notes to finance its $125.2 billion, five-year TPU lease — a supplier-as-financier structure that mirrors Nvidia's playbook and concentrates risk in two companies.

The Lender Is Also the Supplier: Broadcom's $42 Billion Convertible Loan Anchors Anthropic's $125 Billion TPU Bet

Anthropic’s IPO prospectus runs to hundreds of pages of partnerships, obligations, and risk factors, but one relationship stands apart from the rest. Buried in the filing, first reported by Reuters on October 1, is a disclosure that redefines how the AI buildout gets financed: Broadcom has agreed to lend Anthropic up to $42 billion — and the chipmaker Anthropic pays for its compute could end up owning a piece of the company itself.

What the filing actually says

The mechanics are straightforward even if the numbers are not. Broadcom would extend the financing through convertible notes — debt instruments that can later convert into Anthropic shares. The facility could finance roughly a third of Anthropic’s $125.2 billion commitment for a five-year lease of tensor processing unit (TPU) computing capacity, with the chips coming online starting in 2027. Broadcom may also bring in a designated financing partner to share the load, and Anthropic stated in the filing that it does not expect any notes to be sold before the IPO completes.

The arrangement makes the two companies inseparable. Anthropic is expected to become the largest customer in Broadcom’s bread-and-butter chip design business next year, even as Broadcom functions as one of its lenders. For context on the scale: Broadcom projects AI semiconductor revenue of about $115 billion in fiscal 2027 and $230 billion in fiscal 2028 — and Anthropic is on track to be the biggest single compute customer in that stack by 2027.

There are strings attached that public investors will want to read closely. Anthropic disclosed that it deposited cash into a restricted account for Broadcom’s benefit in April 2026 and may be required to contribute additional amounts in certain circumstances. More pointedly, the company warned that certain payment or performance defaults could make a substantial portion of its lease obligations immediately due, while simultaneously limiting its ability to draw on the $42 billion facility to cover those very payments.

A structure with a trailblazer

If this playbook feels familiar, it should. Nvidia pioneered the model over the past two years: use a fortress balance sheet to help customers buy your chips, whether through investments, backstops, or guarantees, and let the financing itself become a sales channel. Broadcom is now following suit out of competitive necessity.

“Nvidia is putting in place a massive amount of its balance sheet, and Broadcom is having to follow suit,” Seaport Research analyst Jay Goldberg told Reuters.

The difference here is the equity conversion feature. Nvidia’s arrangements — equity stakes in OpenAI, guarantees tied to SoftBank data centers — have drawn plenty of scrutiny for creating circular, self-referential flows of money. Broadcom’s convertible note goes a step further structurally: the supplier literally holds an option on the customer’s equity, converting a vendor relationship into a potential ownership position just as that customer arrives at a public valuation that could reach $2 trillion.

The conflicts Anthropic itself flagged

To its credit, Anthropic did not bury the concern. The prospectus explicitly acknowledges that Broadcom’s dual role — supplying hardware and acting as a financing partner — creates “potential conflicts of interest” that might affect Anthropic’s ability to access the compute it needs. The filing further warns that Broadcom’s decisions around pricing and hardware could constrain Anthropic’s ability to procure sufficient infrastructure.

This is not boilerplate. The April 2026 partnership that Anthropic announced with Google and Broadcom grants access to multiple gigawatts of next-generation TPU capacity beginning in 2027, built on TPUs that Google and Broadcom have co-developed across several generations. Anthropic’s entire training roadmap for the back half of the decade now flows through a counterparty that is simultaneously its creditor, its supplier, and potentially its shareholder — a counterparty whose incentives on pricing may not always align with Anthropic’s need for margin.

Wall Street’s concentration worry

The structure lands at a moment when analysts are already uneasy about how AI infrastructure spending is being funded. The reciprocal spending pattern — chipmakers financing the purchases of the AI labs that buy their chips, cloud providers investing in labs that commit to their clouds — has animated skeptics all year.

“It feels that there’s quite a concentrated bet right now on two companies being able to generate enough revenues to support all the financing that’s happened,” said Robert Leitao, managing partner of Rothschild & Co.

That “two companies” framing is the uncomfortable truth inside the filing. The AI capital cycle increasingly rests on the assumption that OpenAI and Anthropic — neither consistently profitable, both posting enormous losses — will generate the revenue required to service hundreds of billions in combined obligations. Anthropic alone carries $518 billion in planned infrastructure spending across six partners, with roughly 80% of it non-cancellable or requiring payment regardless of delivery, as Reuters reported earlier this week. The $42 billion Broadcom facility doesn’t reduce that exposure; it institutionalizes it.

Why it still makes sense for both sides

For Anthropic, the logic is blunt: capital at scale is scarce, and a convertible note from a strategic partner is cheaper and faster than raising the equivalent in equity ahead of an IPO that could be the largest in history. It also deepens a relationship with the one supplier whose roadmap — Google-designed TPUs built with Broadcom — offers a genuine alternative to Nvidia’s pricing power.

For Broadcom, the calculus is equally direct. Anthropic is about to become its largest compute customer, and financing that customer secures the demand backlog underwriting those $115 billion and $230 billion revenue projections. If Anthropic thrives, the notes convert into equity in a $2 trillion company. If it struggles, Broadcom holds senior claims and restricted cash cushions. Either way, it has hedged its most important customer relationship.

What to watch

Three signals will determine whether this deal looks genius or reckless in hindsight. First, whether the notes actually convert — and at what valuation — after the IPO. Second, whether Broadcom brings in the financing partner it is permitted to designate, which would dilute the concentration but also signal how much of the risk it wants off its own books. Third, how regulators and public-market investors price the conflicts of interest that Anthropic itself has now put in writing.

The vendor-financing era of AI is no longer an emergent risk flagged by fringe analysts. It is a disclosed, contractual, $42 billion fact in an IPO prospectus — signed by a lender that also sells the collateral.