Hidden Debt Returns: Billions Slip Off the Balance Sheet in the AI Race

Eisman opens the week with two numbers: oil prices and the 10-year US Treasury yield. Both are unpredictable in his view, yet either can move the market on its own. Early in the week the 10-year yield dipped briefly below 5 percent, then strong data pushed it above 5.10 percent and stocks sold off. By Thursday, reports of a possible understanding around the Strait of Hormuz lifted the market again. Reuters' bond-market analysis that week argued 5 percent is no longer a shock level but a line investors watch with unusual care. Eisman's verdict is blunt: while the yield stays above that line, a correction remains on the table.

Two numbers run the market

The most unsettling corporate item came from Oracle and Blue Owl. Blue Owl is building an AI data-center campus codenamed Jupiter for Oracle, and Oracle triggered a force majeure notice that lets it defer payments if the project runs late. Data-center outlet DataCenterDynamics reports that Jupiter-linked debt was already changing hands near 90 cents on the dollar, with pressure building further. The picture speaks of Oracle's anxiety over thin cash buffers and a swelling debt load. For the speaker, this is a project-finance signal that talks about balance-sheet health faster than any rating.

Oracle's rating at the junk edge

Oracle's credit rating had already cracked. The Wall Street Journal (WSJ) reported that S&P cut the company from BBB to BBB- in July, a single notch above speculative, or junk, territory. Eisman reads this as the inevitable bill for leverage-led growth: debt rises, the rating falls, and a lower rating makes fresh debt more expensive. He argues that agencies mostly stare at debt that sits on the balance sheet while treating off-balance-sheet obligations as if they vanished. That blind spot takes center stage in the Meta chapter of the story.

Paramount settlement and the streaming wars

On the Paramount front, a settlement closed the week: the company ended the antitrust case that 11 state attorneys general had filed against its Warner Bros. combination. Eisman never took the case seriously, and his numbers explain why: Netflix is worth about $300 billion, Disney about $180 billion, and the combined Paramount-Warner entity only about $88 billion. In the streaming wars scale is everything, so the merger arguably sharpens competition rather than dulling it. CNN reports the settlement includes a board overseeing CNN and a pledge of 30 theatrical films a year from Paramount, neither of which has anything to do with textbook antitrust. In his reading, this was the California attorney general imposing political oversight.

None of the entertainment names looks like a good investment to him. The thesis is simple: Netflix has hit a growth wall because the streaming wars no longer expand the market, they only fight over its shares. Raising prices looks like the only lever for revenue growth, yet prices are already so high that another increase risks enraging customers. This is the mature phase of the subscriber economy : once the pool of new customers dries up, companies poach each other's wallets. Even Disney's $180 billion scale does not change the equation, because the problem is not size but the source of growth. The result is an expensive wait behind thin defensive margins.

SoftBank's leveraged gamble

The week's boldest AI move came from SoftBank. Chair Masayoshi Son had taken a $10 billion bridge loan from a bank group over the summer to fund a $10 billion OpenAI investment, and this week he moved to refinance it with an $11 billion junk-bond sale. CNBC describes the issue as a giant high-yield operation built to bankroll the OpenAI bet. Eisman's point is about temperament as much as accounting: risking your own money is easy, borrowing to buy someone else's equity is hard. Buying stock with leverage means the debt and the position can melt at the same time. This opens the week's main theme: invisible debt deserves as much attention as visible leverage .

Circular money and invisible debt

Eisman first clears away a common objection: the circular financing in which Nvidia funds a customer that then buys Nvidia chips is not what worries him. That structure sits in daylight, everyone can see it, and Nvidia's enormous cash flow can carry it. The real problem is exploding cost: Nvidia recently raised chip prices by about 15 percent, and CIO reports the memory squeeze is pushing prices higher still. Fortune estimates that hyperscalers such as Google, Amazon, Microsoft, Oracle and Meta will bury roughly $700 billion in AI capital expenditure this year; companies that once barely needed capital have seen their cash flow evaporate and are raising outside money for the first time in decades. Off-balance-sheet appetite swells exactly here.

What follows is a history lesson. Enron set up special-purpose vehicles to flatter its debt load, shifting billions into entities controlled by its own finance chief, Andrew Fastow. Eisman recalls reading the company's 10-K, seeing everything disclosed in plain sight, and still wondering how the auditors signed off. Analysts shrugged that disclosure made it fine, which he takes as the era's naivety in a sentence. The rest is record: the US government sued auditor Arthur Andersen, the firm collapsed, Fastow went to prison and his gains were wiped out. The lesson holds: removing debt from the balance sheet does not destroy the debt; it only distracts the rating and the audit for a while.

The global crisis staged the same play in different costume: structured investment vehicles , the shadow banks of the era. Big banks parked assets in these conduits to dodge leverage and capital rules; the vehicles bought long-dated, high-yield mortgage paper with cheap short-term borrowings. When the housing bubble burst, the short-term debt could not be rolled and the whole load snapped back onto bank balance sheets; the magic wand worked in reverse. Eisman says he still cannot grasp how the agencies permitted it. The surviving concept is too big to fail , and his question carries into today: is the same film being remade?

Meta's Louisiana equation

The AI era's first big specimen is rising in Louisiana. Meta financed its $27 billion data-center campus not on its own books but through a variable-interest vehicle called Bayou Investor; Meta owns only 20 percent, with Blue Owl and PIMCO holding the rest. Yet Meta is building the campus, commits to leasing it for twenty years, and absorbs delays and overruns. PitchBook's analysis notes that investor protections in such joint ventures are under the microscope for exactly this reason: the debt never touches Meta's balance sheet, but the shadow of the risk stands right beside it. Eisman calls it a cunning maneuver that recalls painful days.

The week's second half belongs to Muse, the social assistant Meta released on September 8. Eisman's wife found hard-to-get theater tickets and organized a work trip with it, then started complaining about its limits a day later. The speaker's behavioral point: the assistant scans every site for the cheapest ticket, so spending does not stop, it redirects; he calls it directed spending . TechCrunch's late-September analysis found Muse gaining serious momentum after the Connect conference, with Meta throwing its full weight behind it. The real questions are economic: does this convenience lower prices, how does it reshape consumption, and where does profit land? The travel-agent example shows the limits; the investor questioning his Nike loyalty suggests even ad revenue is no longer safe.