> ## Content Index > Fetch the complete content index at: https://www.sweatyspice.com/llms.txt > Use this file to discover other available public pages before exploring further. # Revenue from Nowhere (AI Economics) - URL: https://www.sweatyspice.com/revenue-from-nowhere-ai-economics/ - Published: 2026-08-23T11:52:30.000Z - Updated: 2026-08-23T11:52:35.000Z - Description: A meaningful share of the debt financing the AI buildout is collateralized by one company's promise to buy things from another company that needs that promise to be believed. Oracle's credit default swaps just hit their highest level since 2008. Coincidence is a strong word. - Author: Geoff Anderson - Tags: Tech ## Table of Contents - [The Boring, Legitimate Version](https://www.sweatyspice.com/revenue-from-nowhere-ai-economics/#the-boring-legitimate-version) - [The Mechanism, and the Ghost of Enron](https://www.sweatyspice.com/revenue-from-nowhere-ai-economics/#the-mechanism-and-the-ghost-of-enron) - [Whose Money This Actually Is](https://www.sweatyspice.com/revenue-from-nowhere-ai-economics/#whose-money-this-actually-is) Here's a fun one. As of this year, Microsoft has issued zero public bonds. Its total reported debt actually went down, from $44.9 billion to $40.3 billion, the kind of number that makes a balance sheet look like it's on a health kick (MAHA anyone?) Over that same stretch, its finance lease liabilities exploded. Same company, same buildout, same money going out the door. One number goes down because you asked the accountants politely to put it somewhere else. That's not a scandal, technically. It's just the trick, performed in the open, by the cleanest-looking company in the group. Zoom out and the trick has a price tag. Amazon, Microsoft, Google, Meta, and Oracle are sitting on a combined [$2.13 trillion in "signed but off-balance-sheet" leases and commitments](https://finance.biggo.com/news/0713e74c721cfa2a?ref=sweatyspice.com), against a fraction of that in bonds anyone can actually see on a 10-K. And that number is already out of date. Five days after it ran, the [Wall Street Journal put the figure at $3.1 trillion across nine companies](https://www.forbes.com/sites/petercohan/2026/08/19/the-hidden-3-trillion-pipeline-funding-big-techs-ai-expansion/?ref=sweatyspice.com), because Alphabet alone added another few hundred billion in disclosed commitments between one earnings call and the next. > The number you're reading right now is wrong. It was wrong before I finished the sentence. That's not a bug, it's the whole architecture working as **designed**. ## The Boring, Legitimate Version *The stated argument, and it's not idiotic, which is exactly why it's worth taking seriously before we take it apart.* Nobody *invented* a new kind of financial instrument to build a data center. Special purpose vehicles (SPVs) are the same tool that's financed toll roads, power plants, and pipelines for half a century. You want a pension fund to lend money against a twenty-year, cash-flow-generating asset without dragging the parent company's own credit rating into it, an SPV is the boring, legitimate way to do that. The take-or-pay contracts, OpenAI promising to buy compute whether or not it needs it that quarter, are the same structure that lets a power plant get built against a utility's promise to buy the electricity. Private credit funds doing the lending, Pimco, Apollo, Blue Owl, are not rubes. They employ people whose entire job is reading footnotes for a living, and they've decided this pencils out. There's a real historical counter-argument too, and it deserves better than a shrug: the late-1990s fiber buildout looked catastrophically overbuilt and debt-financed right up until it quietly became the backbone the entire internet economy runs on today. Bandwidth nobody could justify in 1999 got dirt cheap and infinitely useful by 2005\. Maybe GPUs are fiber. Maybe today's "who's going to use all this compute" panic reads, in ten years, like an old man yelling about the price of a T1 line. Perhaps. Here's the mechanism, and you can decide for yourself whether it smells like fiber or something older and worse. Sweaty's Corner doesn't have a marketing budget or an algorithm pushing it into anyone's feed. It has readers who share it. If this one landed, sending it to someone who'd get something out of it — or subscribing so the next one finds you directly — genuinely moves the needle. [ Subscribe ](#/portal/signup) Share this post --- ## The Mechanism, and the Ghost of Enron *Every rule governing this exists because of a company that stopped building anything twenty-five years ago.* > The debt didn't disappear. It just moved next door and stopped answering the phone. Strip the euphemism and the structure is dead simple. A parent company spins up a legally separate entity. A private credit fund puts up the debt. The parent keeps a minority stake, small enough that under current accounting rules it doesn't have to fold the SPV's liabilities into its own books. In exchange, the parent signs a long-term lease or take-or-pay contract that gives the SPV "revenue" solid enough on paper to justify the loan. Everyone's balance sheet stays clean. The debt didn't disappear. It just moved next door and stopped answering the phone. You've heard this before, or you should have. The rules governing exactly this exist because of Enron, a company that ran more than [3,000 special purpose entities](https://medium.com/@PipelineProdigy/the-enron-scandal-a-data-driven-analysis-of-corporate-deception-and-its-modern-implications-74d5816f68d0?ref=sweatyspice.com) to keep debt off its books, using an accounting quirk that let a sponsor dodge consolidation as long as some outside sucker held a mere 3 percent stake in the entity. It inflated cash flow by $9.1 billion in a single year before the whole thing came apart and vaporized something like $63 billion in shareholder value. Afterward, FASB rewrote the rule: the outside stake requirement went from 3 percent to 10, and the test for whether you have to consolidate an entity switched to whoever actually eats the majority of its gains or losses, not just who holds the votes. That rule is not a dusty artifact. It is the exact rule today's AI financing is built to satisfy, on purpose, with better lawyers, at a scale that makes Enron's entire 3,000-entity empire look like a rounding error. Forbes reached for the same comparison independently a few days ago, unprompted, in a piece about the exact $3 trillion number cited above. When the business press starts pattern-matching to Enron on its own, you're allowed to notice. The part that should actually worry you, more than the size, is the circularity. Nobody's hiding a fraud here. What they're hiding is that a meaningful slice of the "demand" propping up this entire debt pile is companies promising to buy things from each other. CoreWeave's own SEC filing shows [its OpenAI commitments alone reaching approximately $22.4 billion](https://www.sec.gov/Archives/edgar/data/1769628/000176962825000059/coreweave3q25earningspress.htm?ref=sweatyspice.com), on top of similarly enormous promises to Oracle and Amazon, and S&P has estimated that roughly [half of Oracle's future revenue depends on OpenAI](https://www.forbes.com/sites/petercohan/2026/08/19/the-hidden-3-trillion-pipeline-funding-big-techs-ai-expansion/?ref=sweatyspice.com) specifically, a single customer that lost billions last year and has never turned a profit. > That's the collateral. Not a factory. Not a mortgage on a house that exists whether or not you make your payment. A promise from a company that needs the promise to be believed in order to raise the next round to keep the promise. Oracle's own credit default swaps have [climbed to their highest level since 2008](https://finance.biggo.com/news/0713e74c721cfa2a?ref=sweatyspice.com), which is either an incredible coincidence of phrasing or the market quietly agreeing with the thesis of this entire post. --- ## Whose Money This *Actually* Is *Not Silicon Valley's, mostly. Yours, filtered through several layers of people who took a fee.* Private credit funds don't conjure capital from the ether, whatever the branding implies. They raise it from pension funds and insurance reserves, the same institutions that are supposed to be the boring, safe custodians of other people's retirements. AI-related private credit lending went from essentially zero three years ago to well over a couple hundred billion now, with hundreds of billions more projected over the next few years, and every dollar of it traces back to somebody's actuarial table. Litigation has already started, not hypothetically, right now: Oracle bondholders and CoreWeave shareholders have both filed suit, which one legal analysis described as the opening shots in a wave, not an isolated dustup. There's a correction owed here, to an earlier piece in this series. Post 1[\[1\]](https://www.sweatyspice.com/revenue-from-nowhere-ai-economics/#fn1) cited roughly $500 billion in AI-related corporate bond issuance as the amount competing with the Treasury for capital, a real and useful number, sourced straight from the New York Times. That number only counts what's publicly, transparently issued, the part of the iceberg somebody bothered to put a flag on. It doesn't touch the $2 to $3 trillion sitting in the SPVs described above, which means the actual scale of AI's claim on the world's lending capacity, its real competition with the Treasury for the same pool of pension and insurance money, is understated in that earlier piece by an order of magnitude. Worth saying plainly rather than quietly hoping nobody notices: I was too conservative the first time, and the real number was always bigger than what showed up on anybody's public filing. --- Here's where the 2008 comparison earns its keep, and where it also breaks, and both halves matter. Mortgage debt in 2008 was backing an asset, a house, whose value didn't go up no matter how the crisis resolved. AI infrastructure is a different animal. Even in the ugliest bust scenario, a data center full of GPUs retains some real value independent of which specific over-leveraged company built it, because the compute itself is useful to somebody, eventually, at some price. That's a genuine disanalogy, not a rhetorical hedge. What isn't different, what's identical down to the accounting citation, is the mechanism: leverage engineered specifically to stay invisible, revenue assumptions that only work if the counterparty making the promise is also solvent enough to keep it, and a regulator, the Bank for International Settlements this time instead of the SEC, naming the pattern out loud while it's still inflating instead of writing the postmortem after. History apparently doesn't repeat, it just gets better lawyers and bigger numbers. And when this particular bill comes due, don't expect much help from the people currently in charge of paying the country's other bills. They already told you, a couple pieces back, that the toolbox is empty. Like what you've read? Why not buy me a coffee! All money goes to help defray the expense of running the site. [Buy me a coffee!](#/portal/support) --- [\[1\]](https://www.sweatyspice.com/revenue-from-nowhere-ai-economics/#end1) \- This short series began with the insanity of the Treasury interventions, and commented on why this was necessary. Read it here: