The AI Buildout Is Now Running on Debt. Broadcom Is Reportedly Raising Up to $80 Billion, in the Biggest Private-Credit Deal Ever.

According to Bloomberg on 20 August and CNBC on 21 August, Broadcom is in talks with a group of lenders to raise more than 60 billion dollars, by some reports 70 to 80 billion, in debt for an AI-chip financing deal that will benefit Anthropic and other companies. The borrowing runs through a Broadcom-backed special purpose vehicle rather than the company's own balance sheet, and banks including Bank of America, Goldman Sachs and Morgan Stanley are already trading pieces of it. It is being described as the largest private-credit deal ever assembled, now being offered to a wider pool of investors. The headline number is astonishing. The structure is the story.

What is actually being financed

Strip away the size and this is infrastructure finance. The money buys the chips and funds the capacity to deploy them, and a special purpose vehicle, a company created to hold the assets and the debt separately, borrows against the future revenue that capacity is expected to throw off. Anthropic is among the beneficiaries, which tells you what the compute is for: training and serving frontier models at a scale no single lab wants to carry on its own books. This is not a chipmaker selling more chips. It is the plumbing of the AI boom being financed the way toll roads and power plants are, with structured, asset-backed debt.

The quiet shift from equity to leverage

Until recently the AI buildout ran on equity. Mega venture rounds, hyperscaler cash flow, and eye-watering private valuations, xAI at 230 billion, OpenAI reported near 500 billion, funded the compute. What is new is the turn to debt at this scale. Equity and debt are not interchangeable, and the difference is the whole point. Equity absorbs losses; if the bet disappoints, investors simply make less or lose their stake. Debt demands to be repaid on a schedule, whether or not the AI revenue arrives on time. Moving the buildout onto leverage is a sign of maturity, lenders now believe the cash flows are real enough to lend against, and a transfer of risk from optimistic equity holders to creditors who expect to be paid regardless.

Why special purpose vehicles and private credit

The choice of vehicle matters as much as the amount. Putting the debt in an SPV keeps it off the operating company's balance sheet, isolates the risk, and lets specialist credit investors fund it directly rather than through public bond markets. It is the same instinct we saw when Anthropic set up a dedicated vehicle to build and lease its data centers: keep the giant, capital-heavy physical layer separate from the fast-moving model business, and let patient infrastructure capital own it. Private credit, money from funds rather than banks' own balance sheets, has become the pool big and flexible enough to write cheques this size. The result is an AI infrastructure layer increasingly owned and financed by the same institutions that own airports and pipelines.

The historical rhyme worth respecting

Debt-financed infrastructure booms have a long history, and it cuts both ways. The railroads of the nineteenth century and the fiber-optic buildout of the late 1990s were financed with enormous leverage, built genuinely transformative infrastructure that outlasted their backers, and also produced spectacular busts when capacity ran ahead of demand and the debt came due. None of that is a prediction that AI will follow the same arc. It is a reminder that leverage amplifies both directions. When the buildout is funded by equity, a demand air-pocket is a disappointment. When it is funded by debt, a demand air-pocket is a default risk.

The honest risk in one sentence

Here is the part the size of the number can hide: debt is a fixed cost that does not care whether your forecast was right. If AI demand or, more precisely, AI monetisation, lags the pace of the buildout even temporarily, leveraged infrastructure is where the strain shows up first, through refinancing pressure, forced asset sales, and stranded capacity that still owes interest. The models will keep getting better regardless. Whether every financing structure erected around them survives a slow quarter is a genuinely separate question, and the more the buildout leans on debt, the more that question matters.

What it means if you build on AI rather than lend to it

For everyone downstream, this wall of financing is, on balance, good news. It is building the capacity that keeps compute abundant and your inputs cheap, the same dynamic that makes a crowded, converging model race a gift to builders. But do not confuse the infrastructure boom with your own business. The value you capture is not in owning compute or betting on the buildout; it is in the specific, grounded decision you build on top of cheap capacity. Let the lenders take the leverage risk. Your job is to turn the cheap capability they are financing into an outcome only your data and your problem can define.

The bottom line

The AI buildout has entered its leverage phase. Expect more capacity, sooner, and more concentrated financial risk sitting in vehicles most people never see. It is a vote of confidence from the most hard-nosed money in the world, and it is a bet that the revenue will arrive on schedule. Both things are true at once. Watch the debt, not just the models, because for the first time the financing structure, and not only the technology, is part of the AI story.