NVIDIA reported its second quarter on 26 August 2026, and the number everyone quoted was the top line: revenue of 96.221 billion dollars, up 106 percent year on year, with data-center revenue of 89 billion and a guide to 108 billion for the current quarter. “AI has reached its inflection point,” said Jensen Huang. “Now, compute is revenue.”
The demand story is real and it is not the story. The consequential disclosures in that release were about financing, not selling. Read them together and the quarter marks a shift: the AI buildout is moving off corporate balance sheets and into the credit markets.
When the cash king borrows
Start with the tell hiding in NVIDIA’s own balance sheet. This is the most cash-generative company in technology, with more than 99 billion dollars in cash and marketable securities and 21.3 billion in free cash flow in the quarter alone. It does not need to borrow.
It borrowed anyway. NVIDIA’s long-term debt rose from 7.469 billion dollars at the end of January to 32.366 billion by late July, a near-quadrupling in six months, with 24.896 billion in new debt raised during the period per the cash-flow statement. When the company with the least need for outside capital in the entire sector taps the bond market, the signal is not distress. It is that debt has become the preferred instrument for financing compute, even for those who could pay cash.
NVIDIA went further than borrowing for itself. In the same release it disclosed plans to help mobilize more than 500 billion dollars of third-party capital, through what it called independent compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, subject to definitive agreements. The chip vendor is now an orchestrator of private capital for the buildout, standing between the balance sheets that fund data centers and the silicon that fills them.
The circular financing loop
There is a second structure worth naming. NVIDIA’s non-marketable securities, largely stakes in the customers and partners that buy its chips, rose from 22.251 billion dollars to 51.157 billion over six months. The company is increasingly an investor in its own demand. Capital flows out to a buyer, the buyer spends it on GPUs, the revenue flows back. None of that is improper, and all of it makes the reported demand harder to read as fully independent.
The buildout hits the bond market
Zoom out from one company and the pattern is a wave of issuance. By Vanguard’s count, corporate bond issuance from the five largest cloud operators reached roughly 132 billion dollars in 2026 through 31 July, against 93 billion for all of 2025 and an annual average near 35 billion across 2020 to 2024. Seven months of 2026 already exceed the whole of the prior year.
That total includes one multitranche offering of roughly 53 billion dollars, among the largest corporate bond sales on record. Some of this debt carries maturities that stretch a century out. The buildout is being termed out over horizons longer than the useful life of any chip it will buy.
The part you cannot see on the balance sheet
The bond issuance is the visible half. The other half is engineered to stay off the income statement. In October 2025 Meta closed a roughly 27 billion dollar joint venture with funds managed by Blue Owl Capital to build the Hyperion data center in Louisiana. The Blue Owl funds own 80 percent, Meta 20 percent, and the debt was placed privately with bond investors including PIMCO. The economics are Meta’s. The debt is not on Meta’s balance sheet.
The Bank for International Settlements has a name for this. It calls the structures “shadow borrowing: obligations that are economically akin to debt but largely reside outside corporate balance sheets,” built from minority stakes plus long-term leases and offtake agreements. The obligation is real. The reported leverage understates it.
The stress is already visible
This is not a hypothetical risk waiting for a downturn. Some of it has already repriced. On 9 July 2026 S&P downgraded Oracle to BBB minus, one notch above high yield, after the company’s capital spending ran to 55.7 billion dollars in its fiscal 2026, up from 21.2 billion, pushing free cash flow negative against a remaining performance obligation of 455 billion. A single unprofitable customer, OpenAI, accounts for a large share of that backlog. The credit market is already pricing the concentration that the equity market has been happy to celebrate.
What it means
The right question about the AI boom has quietly changed. For two years the debate was whether the equity was overvalued. The more useful question now is who holds the credit, and how much of it sits off the balance sheets that report it.
That reframing globalizes the risk. Nasdaq drawdowns hit shareholders. But investment-grade bonds, century notes and private-credit data-center vehicles land in pension funds, insurers and the limited partners of private-credit managers, in Europe and Asia as much as in the United States. Vanguard is measured about it, and rightly: these are exceptionally strong credits, and nothing in the current data is a solvency warning. Its caution is narrower and more precise. Spreads are thin, near historically tight levels, and concentration, long an equity-market phenomenon, is migrating into the asset class.
The signal from NVIDIA’s quarter is not the 96 billion. It is that the company least in need of debt chose to issue it, and that the sector is routing an ever larger share of the buildout through bonds and off-balance-sheet vehicles rather than retained earnings. When capital is that abundant and that cheap, the constraint stops being whether the money exists. It becomes how far the risk can be spread before someone has to hold it. The next repricing of the AI trade may not show up first in a stock chart. Watch the credit.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.