For two years, the artificial intelligence boom was a story about the stock market — soaring valuations, “Magnificent Seven” earnings, the race for GPUs. Two weeks ago, this newsletter read the August equity shakeout through Carlota Perez. This week, the story moved from the equity markets to the credit markets. That is a much more dangerous neighborhood.
Nearly 70 percent of the $456 billion raised for AI from public markets in 2026 has come from investment-grade debt, according to Bank of America Global Research — $309 billion of bonds, more than double the $136 billion issued in all of 2025. Tech giants and data center developers are no longer just spending their own cash. They are borrowing hundreds of billions to fund the build-out, and that figure does not even count the private credit and off-balance-sheet layer, where Nikkei estimates the five biggest players carry another $1.65 trillion in obligations.
This week the credit market started talking back. On Thursday, news broke that Broadcom is seeking roughly $100 billion through an off-balance-sheet vehicle — the largest deal of its kind ever attempted — and the cost of insuring Broadcom’s debt jumped on the announcement. Oracle’s credit default swaps now trade near a record 215 basis points, up from about 40 a year ago. Alphabet’s hit a record after it reported its first negative free cash flow quarter since going public in 2004. Nvidia’s passed its July peak on Tuesday. And on Friday, Bloomberg reported that data-center bonds rated investment grade are being sold to junk-bond buyers at yields above 7 percent — high-grade paper, priced like junk.
The market is asking a question it has not asked since 2008: what happens if the revenue from these assets never arrives to pay back the debt? To understand why this shift matters so much, you need the economist who spent his life studying why stable markets destroy themselves.
The book that explains the danger
Hyman Minsky was a relatively obscure economist for most of his career. His 1986 book, Stabilizing an Unstable Economy, became the bible of the 2008 financial crisis, reissued as the phrase “Minsky moment” spread across Wall Street. His central contribution, the Financial Instability Hypothesis, is summed up in three words: stability is destabilizing.
Minsky argued that long periods of prosperity create the conditions for collapse. When things go well, investors grow overconfident. They take more risk, borrow more money, and gradually shift the whole system from safety to fragility. He identified three stages of debt, and they map the AI boom with uncomfortable precision.
The three stages of the AI boom
The first stage is hedge finance. Income covers both interest and principal. This was the early AI era: Microsoft and Google funding experiments from enormous cash piles generated by search and software. The debt, where it existed, was backed by proven revenue.
The second stage is speculative finance. Income covers the interest but not the principal, so the borrower must roll the debt over — borrow again to repay the old loan. The hyperscalers are migrating into this stage now. Their capital spending is on pace to consume close to 100 percent of operating cash flow in 2026, against a historical average of about 40 percent. Free cash flow across the five biggest names fell roughly 24 percent in 2025. They are borrowing against the future profits of AI to pay for the present build-out, betting that by the time the bonds mature, the profits will exist.
The third stage is Ponzi finance. Income cannot even cover the interest; the borrower survives only if asset values keep rising. The core hyperscalers are not there — their leverage remains low. But the outer ring of the boom already rhymes with it. CoreWeave, the debt-funded data center operator, has never generated positive free cash flow, and its credit default swaps trade above 850 basis points — a price implying roughly a 50 percent chance of default within five years. The deeper the financing migrates into private credit and special-purpose vehicles, the harder the fragility becomes to see. That opacity is exactly where Minsky said it accumulates.
The warning lights
The “Minsky moment” — the term Paul McCulley coined in 1998 — arrives when lenders stop believing the income will ever cover the debt. They stop rolling loans over. Borrowers sell assets to raise cash, prices fall, and the crisis feeds itself.
The early signals are visible in the plumbing. When Amazon sold a surprise $25 billion bond in July, investor orders covered the deal 2.5 times, down from 3.2 times in March, and Amazon had to pay up to 21 extra basis points to get it done. Honesty requires a scale check, too: AI-linked bonds are still only about 2 to 3 percent of the public bond benchmarks — nothing like subprime in 2008. But the telecom sector went from 1 percent of the bond index in 1995 to 20 percent in 1999, just before the dot-com bust. Sectors migrate fast.
The macro backdrop raises the stakes. The US national debt crossed $40 trillion on Tuesday, and the average interest rate on it has more than doubled in five years, from about 1.5 percent to 3.4 percent. Interest is now Washington’s second-largest expense after Social Security. In a world of expensive money, the cost of being wrong about a half-trillion-dollar bet is the highest it has been in decades.
What to watch next
The test of Minsky’s map comes over the next two quarters. If AI revenue turns real and cash-flow positive, the sector stays in the speculative stage and the debt gets serviced. Three signals will tell you before the earnings do. Watch order coverage on the next jumbo bond deals — falling coverage means lenders are tiring. Watch the credit default swaps — Oracle above 215, Nvidia near 80, Alphabet at 67; further widening means doubt is compounding. And watch how much new financing slides off balance sheet — because in Minsky’s world, risk does not disappear when it becomes invisible. It matures.
As Minsky taught, the time to worry is not when the market is crashing. It is when everything still looks stable.
This week’s book: Hyman Minsky, Stabilizing an Unstable Economy (Yale University Press, 1986; reissued by McGraw-Hill, 2008).
Sources: Bank of America Global Research via Barron’s (Aug 20, 2026); Bloomberg (Aug 21–22, 2026); Fortune (hyperscaler debt and cash flow, July–Aug 2026); Axios (Oracle CDS, hidden debt); Seeking Alpha (Nvidia CDS, Aug 18, 2026); US Treasury / Joint Economic Committee (national debt and interest rates, Aug 2026); Penn Mutual Asset Management (AI bond benchmark share, Aug 2026). Book: Hyman Minsky, Stabilizing an Unstable Economy (1986).