By Liquiditrax ResearchPublished
Broadcom lost about 6% of its value on Friday on a day it announced nothing. No earnings, no guidance cut, no lost customer. An analyst put a number on how its AI business is financed, and the market decided that number mattered more than the business.
Key finding: the selloff was triggered by an estimate of debt that Broadcom does not owe.
Bank of America analyst Tom Curcuruto flagged that Broadcom's chip-financing vehicle could reach $370B of senior debt by mid-2029 at 20-gigawatt scale, including roughly $150B of new issuance in 2027 alone. The stock fell 5.94% to $392.99 (24/7 Wall St.; Yahoo Finance).
Read the structure carefully, because the headline number and the actual obligation are two different things. Per Broadcom's latest 10-Q, an investor partner assumed the purchase and lease agreements, and Broadcom agreed to backstop lease payments for five years, with maximum exposure of up to $29B on the initial transaction. The vehicle raises capital and leases custom AI accelerators to customers. It launched in June when Apollo and Blackstone led a $35B financing of Broadcom's AI XPV Platform, funding more than 1 gigawatt of compute for Anthropic, with the platform sized to support more than 20 gigawatts through 2028. Blackstone has already sounded out investors for a further transaction exceeding $30B.
So there are two numbers. $29B is disclosed, contractual and Broadcom's. $370B is a projection of a third party's senior debt at a scale that has not been built. Both are true. The share price spent Friday arguing about the distance between them.
Why it matters: the demand side was never in question.
Broadcom's fiscal Q2 2026 revenue was $22.19B, up 47.9% year on year, with AI chip revenue up 143%, and management guided to $16B of AI semiconductor revenue for the current quarter. Prediction-market pricing cited alongside the selloff gave a 94% chance of topping $15B of AI revenue this quarter and 78% of exceeding $16B. The stock was up 21% year to date through the prior close, which is not a stretched expectation problem.
The control experiment ran the same day. NVIDIA went public this week with a plan to collectively finance roughly $500B of AI computing deals alongside Goldman Sachs, Blackstone and Apollo, with KKR, BlackRock and Brookfield added days earlier. Jensen Huang said NVIDIA's support would cover as much as 25% of an opportunity, and no deals were signed at announcement. NVIDIA did not sell off.
Same structure. Different disclosure. Opposite reaction. That is the whole finding.
And this was not the AI trade de-rating. The iShares Semiconductor ETF fell only 0.7%, Intel fell 2%, and AMD rose 5.13% to $507.77 on Baird doubling its price target to a Street-high $1,250 from $625, modelling $147B of AI GPU platform revenue by 2030 on 15% share of the data-centre accelerator market. On one screen, the market marked AI leverage down and the AI addressable market up.
The connecting thread: leverage is accumulating where the usual dashboards cannot see it.
We wrote on 5 August that AI capex was becoming a credit story and said to watch investment-grade and high-yield spreads and Oracle's CDS. The thesis was right and the indicator was wrong. Spreads and CDS see funded debt at the issuer. They do not see a special-purpose vehicle's senior debt or a five-year vendor lease backstop. A Goldman Sachs note cited Friday says hyperscalers are using lease commitments and other debt structures to fund expansion, so this is a pattern across the theme rather than one company's quirk.
Which means the repricing does not drift, it steps. It happens on the day someone publishes a number.
There is a macro rhyme worth noting. On the same day this financing structure was repriced, the Treasury sold 30-year bonds at 5.216%, described as the highest in 25 years. These vehicles are funded by private credit and alternative managers whose cost of capital lives at that end of the curve.
So what for an allocator (ABC).
The screen for AI exposure now needs three axes, not one. Funding source: does the growth come out of operating profit, debt or new shares. Cash conversion: does the earnings beat show up as free cash flow, the question Applied Materials failed on Thursday and kept failing Friday, down another 5.59% to $504.67. Financing disclosure: is the buildout financed on a balance sheet you can read, or in a vehicle you cannot.
A vendor that arranges financing for its own customers is booking revenue against a contingent liability. Revenue growth and credit exposure become the same line, read twice.
For Alpha, the specific thing to underwrite in Broadcom is the backstop, not the revenue, because the backstop is the only number the company actually owes. If that disclosed $29B maximum exposure grows materially as the platform scales toward 20 gigawatts, the liability is compounding alongside the sales.
For Beta, this is a reminder of what index exposure quietly contains. Semiconductors are a large, concentrated block of a market-cap-weighted index, and a financing risk that nobody is pricing is still inside your broad-market sleeve whether or not you have an opinion on it.
For Cash, no action. This is a research finding, not a stress event. The VIX fell on the day.
Takeaway: when a company finances its own customers, its revenue growth and its credit risk are the same number written in two places. Read the filings for the backstop, not the press release for the beat.
Analytics & education, not advice. DYOR.