Thinker lens
Hyman Minsky
A borrower is safe or fragile by when its debts fall due against when its cash comes in. Minsky's test asks which, and why good years push borrowers the wrong way.
The Lens
Hyman Minsky sorted borrowers by one question: when a payment falls due, where does the cash come from? In Stabilizing an Unstable Economy (1986) he gave three answers. A hedge borrower’s operations cover every payment, interest and principal. A speculative borrower covers the interest but must roll over the principal. A Ponzi borrower cannot cover even the interest, so it borrows to pay it and its debt grows. Minsky wrote that the last term does not imply intent to cheat.
A hedge borrower is exposed to its own sales and costs. The other two are exposed to their lenders as well:
“Debtors and bankers engaged in speculative and Ponzi finance expect payment commitments on debts to be met by refinancing, increasing debts, or running down superfluous stocks of financial assets.” — Minsky 1986, Stabilizing an Unstable Economy (ch. 9, “Hedge, Speculative, and Ponzi Finance”)
A shortfall in income or a rise in rates moves a borrower down the scale. Minsky’s central claim is that good years do too, because they shrink the cushion lenders ask for:
“…success breeds a disregard of the possibility of failure; the absence of serious financial difficulties over a substantial period leads to the development of a euphoric economy in which increasing short-term financing of long positions becomes a normal way of life.” — Minsky 1986 (ch. 9, “The Thrust toward Speculative and Hence Ponzi Finance”)
On chips, Minsky split assets in two: land and inventories, which lenders can lend against, and plant built for one job:
“The other class of capital assets consists of plants and equipment with no significant value outside of particular uses. … These capital assets are not suitable for to-the-asset financing.” — Minsky 1986 (ch. 13, “Ordinary Corporations as Financial Institutions”)
Why It Matters Now
The Credit Wall sets out the record: the largest AI borrowers sold bonds into deep order books while banks struggled to place loans on sites still being built. Minsky’s scheme predicts that split. A hedge borrower barely meets the credit market; a speculative one meets it at every maturity. So the lens says where to look: not at an index of bond spreads, but at who must refinance, when, and what their lenders take as security.
Ray Dalio asks whether borrowed money earns enough to repay. Minsky asks whether it earns in time. A data centre that earns over many years, funded by a loan due when it is finished, can pass the first test and fail the second.
Applied to Tech Markets
CoreWeave shows the move in dated steps. In August 2023 it borrowed $2.3 billion against Nvidia chips. In March 2025 the Financial Times reported nearly $7.5 billion of repayments due by the end of 2026, far above its cash flow, and that it had breached terms of a Blackstone loan in 2024. In May 2025 it planned a $1.5 billion junk bond to refinance part of what it owed. That is Minsky’s speculative posture: principal met by new borrowing. In FY2025 its operating cash flow was $3.1 billion against $10.3 billion of capital spending, and its debt before leases rose to $21.4 billion from $7.9 billion. By 14 October 2026 its bonds yielded close to 15%.
TEXXR’s graph holds 32 dated CoreWeave edges that name debt, a loan, a bond or collateral. The 7 from 2023 and 2024 all record money raised. The first to record money owed is dated 22 March 2025, and 9 of the 25 since then concern debt already on the books: the repayments, the breach, the refinancing, the yield. These are edges, not events; four of the nine record one week’s yield.
The record does not show the Ponzi posture. That test needs CoreWeave’s interest bill against its income, and no source here gives it. Ponzi finance is interest paid with new debt; the record holds no report of it.
The security shifted. The $8.5 billion loan CoreWeave raised in March 2026 was called chip-backed, but Bloomberg reported it was backed by a Meta contract worth up to $14.2 billion. In February AMD agreed to backstop a $300 million loan so Crusoe could buy its chips. In both, a party with income stood behind plant built for one job, as Minsky’s split predicts.
Oracle is a profitable company that moved from paying for its plant to borrowing for it. In the year to May 2024 it produced $11.8 billion of free cash flow. In the year to May 2026 it spent $55.7 billion on plant against $32.0 billion of operating cash, and its borrowings reached $129.5 billion, from $92.6 billion a year earlier. That is the “external financing of investment” Minsky expected a boom to bring. In April 2026 banks struggled to spread the risk of loans on Oracle-leased sites; in October the New York Times reported some Oracle debt trading near 80 to 90 cents on the dollar.
Alphabet and Nvidia sit at the hedge end. Alphabet’s FY2025 operating cash flow of $164.7 billion covered $91.4 billion of capital spending. Nvidia produced $96.7 billion of free cash flow in FY2026 against $11.0 billion of debt and leases. Both borrow. Neither needs one lender to pay another.
None of this dates a crisis or rates a security. It sorts four borrowers by where their next payment comes from: Alphabet and Nvidia from sales, Oracle more and more from lenders, CoreWeave from lenders at each maturity.
The Library
Stabilizing an Unstable Economy (1986) is the source text. Its ninth chapter defines the three postures and the drift between them; its thirteenth argues that the corporation, built to hold long-lived plant on long debt, can as easily borrow short against it. Minsky took his investment theory of the cycle from John Maynard Keynes, whose General Theory is on the shelf.
The library has no reading notes for Minsky’s book yet; Big Debt Crises is the nearest neighbour. Minsky died in 1996. This page applies his framework to a record he never saw.