If you’re valuing upstream AI companies on “order books” and “contracted capacity” — this piece will make you re-read those numbers. Because in a circular market, orders stop being evidence of demand.

Connect the biggest deals on the 2026 AI value chain and you get a strange picture: Nvidia invests in OpenAI; OpenAI reportedly commits $300 billion over five years to Oracle for cloud; Oracle buys tens of billions of dollars of Nvidia chips to fulfill the contract; Nvidia’s revenue and stock rise, funding its next customer investment.

The bluntest specimen sits in official filings: in October 2025, AMD and OpenAI announced a 6-gigawatt GPU agreement — with a warrant for up to 160 million AMD shares, vesting as purchase milestones are hit. A structure written into an SEC exhibit, in black and white: the more the customer buys, the more it earns on its supplier.

Add Microsoft’s OpenAI investment consumed as Azure compute, and Amazon’s and Google’s Anthropic investments returning as AWS and GCP commitments — Bloomberg’s 2026 tally of such arrangements: over $800 billion.

Lucent’s Ghost

This isn’t a new invention. In the late nineties, Lucent and Nortel lent money to telecom carriers to buy their own equipment. Carriers ordered with borrowed money; vendors booked the orders as revenue; revenue propped up stock prices; stock prices funded more lending. Lucent’s 1999 financials looked flawless — until 2001, when carriers defaulted en masse and America’s most valuable telecom equipment maker became a restructuring case within a year.

The textbook summary of that era is one sentence: when a supplier finances its own customers, revenue stops being evidence of demand and becomes evidence of lending.

The 2026 version is safer in two ways: the loop’s core (Big Tech, Nvidia) sits on the strongest cash flows in history, while Lucent lent to dying carriers; and this round’s compute has real, fast-growing end-user payment, while 2000’s fiber demand was almost pure imagination.

But it is more dangerous in one way: this loop is multilateral and nested. Lucent-to-carrier was visible, computable, bilateral lending. Today it’s a multi-layer web among chipmakers, clouds, labs, neoclouds, and private credit funds — equity, purchase commitments, warrants, and contract-collateralized loans wrapped around each other. The risk hasn’t disappeared; it has been structured until no single participant can see the whole picture. That is the topology of the pre-2008 housing credit market.

The Real Cost of the Loop: Price Discovery Is Dead

Every defense of circular financing is partly true — ecosystem investment, supply-chain alignment, a rational bet on future demand. But the loop carries one indefensible systemic cost: it kills “demand” as a signal.

In a normal market, an order is a customer voting with its own money. In a circular market, an order is a customer voting with the supplier’s money. The contracts are real and signed — but they no longer answer the question that matters most: without these capital arrangements, how big is organic demand?

Nobody knows. Not the bulls, not the bears, not even the companies inside the loop. The industry is making history’s largest capital-allocation decisions with price discovery switched off.

Where the Chain Breaks

Loops don’t collapse from being “found out” — they break at the weakest link when cash flow stops. By fragility: refinancing-dependent neoclouds (borrowing to buy GPUs, principal rolling due — CoreWeave alone has $4.2 billion due within 2026; widening credit spreads are the system’s earliest smoke alarm) → second-tier labs (first to die when the funding window tightens; their death shows up upstream as evaporating orders) → Oracle-style intermediaries (who converted labs’ credit risk into their own) → the core vendors (who won’t fail, but whose valuations sit on order books that shrink).

Three sentences to close. Split every upstream company’s revenue into “organic demand” and “loop demand” before valuing it — the more its counterparties depend on financing, the lower the revenue quality. Watch two gauges: neocloud credit spreads, and counterparty concentration in mega-contracts. And remember Lucent’s lesson in its modern form: the fastest-growing supplier may simply be the most aggressive lender.

The upstream names in your portfolio — how much of their order book comes from customers they themselves funded? Comments open.

Data sources (verified, June 2026): circular financing arrangements >$800B (Bloomberg, 2026); AMD–OpenAI 6GW agreement and 160M-share warrant (AMD press release + SEC exhibit, Oct 2025); Oracle $300B contract (media reports); CoreWeave debt (SEC 8-K).

— From Chapter 3 of a book in progress, working title The Deflation Sandwich

#AI #CircularFinancing #AIIndustry