The Circular Trillion, Confirmed: What the Actual Filings Show

Five customers now account for 70% of Nvidia's accounts receivable. Oracle was downgraded over its dependence on one of them. Alphabet just raised $85 billion partly to keep funding the same circle. The filings now confirm, line by line, the circular financing structure described on this page a month ago.

A month ago, a piece on this page described a circular financing structure running between SoftBank, OpenAI, Nvidia and the hyperscalers, the same handful of companies buying from, lending to, and marking up each other's valuations. That was reasonable inference from public reporting at the time. The filings that have landed since are no longer inference. They are the companies' own numbers, and they describe the same structure with more precision than the original piece could.

What Nvidia's Own Filing Shows

Nvidia's 10-Q for the quarter ended July 26, 2026 discloses, in its own customer concentration note, that five direct customers accounted for 70% of accounts receivable, individually 22%, 14%, 13%, 11% and 10%. A year earlier, the same disclosure named three direct customers accounting for 56%. The comparison is not perfectly like for like, since the number of named customers changed too, but the direction is unambiguous: a shrinking set of counterparties owes Nvidia a growing share of what it is owed. Net accounts receivable itself rose roughly 63% over the same stretch, from $38.5 billion to $63.1 billion, partly a function of Nvidia extending longer payment terms, in some cases up to a year, to help investment-grade customers finance their own data center buildouts. Nvidia is not just selling chips to these companies anymore. It is financing the purchase.

Oracle's Downgrade Made the Dependency a Credit Rating

On July 9, S&P downgraded Oracle to BBB-, one notch above speculative grade, citing its dependence on a single customer: OpenAI. Oracle's remaining performance obligations, its contracted backlog, stand at $638 billion, and roughly half of that, a more than $300 billion, five-year cloud agreement including Stargate capacity, is tied to OpenAI, which is contractually committed to paying Oracle roughly $60 billion a year. S&P separately forecasts Oracle running a free operating cash flow deficit of nearly $42 billion in fiscal 2027. A rating agency is now pricing a bond issuer's creditworthiness primarily on the financial health of a company that is not itself investment grade and is not yet profitable. That is not a hypothetical concentration risk. It is a credit rating with OpenAI's name in the rationale.

Alphabet Just Raised $85 Billion to Keep Funding Its Own Customers

In June, Alphabet raised $84.75 billion in equity, its first equity raise in more than two decades and reportedly the largest equity capital transaction in corporate history, to fund AI infrastructure and compute against 2026 capex guidance of $180 to $190 billion. The raise included a $10 billion private placement to Berkshire Hathaway. Alphabet is simultaneously a major investor in Anthropic, a major cloud and compute supplier to the broader AI buildout, and now a company that has gone to public equity markets at a scale it has not used in twenty years to keep funding that same buildout. The capital raised to build the infrastructure and the capital invested in the labs renting that infrastructure increasingly trace back to the same balance sheet.

The Two Companies at the Center: OpenAI and Anthropic's Diverging Q2

The two labs most of this financing ultimately serves reported very different second quarters, per the Wall Street Journal. OpenAI's revenue grew to $6.7 billion, up 18% from $5.7 billion in the first quarter, while its operating loss, including stock-based compensation, widened to $12.3 billion from $9.3 billion. Anthropic's revenue more than doubled to $11.6 billion, overtaking OpenAI's for the first time, and the company posted a $559 million adjusted operating profit, its first, though that figure excludes stock-based compensation and the full methodology has not been disclosed. One of the two companies at the center of the circular financing structure is now larger by revenue and nominally profitable on an adjusted basis. The other is still losing more every quarter. Both are still substantially funded by the same small set of counterparties that also supply, insure and revalue them.

Eisman's Comparison, and Why the Filings Back It Up

Steve Eisman, the investor known for his bets against subprime mortgage debt before the 2008 crisis, has described this structure as resembling CDO-squared securitization: a small set of counterparties repackaging and revaluing exposure to each other until the underlying risk is hard to see from any single balance sheet. That framing is his own interpretation, not a fact in itself, and the comparison to a specific 2008-era instrument is a rhetorical one rather than a structural claim that the instruments are identical. But the filings above are not interpretation. Nvidia's own disclosure shows receivables concentrating in a shrinking set of customers. Oracle's own credit rating now explicitly prices dependence on one of them. Alphabet's own capital raise funds the buildout those customers are renting. None of that requires taking Eisman's word for the comparison. The companies filed it themselves.

What This Confirms for a Portfolio Already Warned Once

The original piece on this page argued that AI-lab exposure, AI infrastructure financing and the chipmaker supplying both were not independent risks, just dressed in different wrappers. The filings since then have not complicated that argument. They have itemized it: a receivables note, a credit rating action, an equity prospectus, and two quarterly income statements, each describing a different piece of the same structure from a different company's own disclosure. A portfolio that treated this as speculative a month ago has had a month to see it confirmed in filings rather than commentary. The question is no longer whether the dependency is real. It is how much of a book is quietly priced on the assumption that it will keep resolving in everyone's favor at once.