The vastly expanding world of artificial intelligence-vectored “data centers,” upon which most measures of “economic growth” in the United States now depend, has entered a period of chaos, called by stock analysts and the like a “debt shakeout.” Data center builder/leaseholder firms like Coreweave, and even the hyperscaler Oracle and the lender Softbank, are suddenly borrowing at 10%-and-higher Junk-bond interest rates, which lenders characterize as “high risk” or subprime debt. The biggest data center plungers, like Meta, Amazon, and Oracle, have been doing “circular” debt deals for some time; they lend $100 billion-multiples of (borrowed) money to each other, to buy data center cybernetics capacities (vulgarized in slang as “compute"), advanced semiconductor chips, or promises of electric power from each other.

Articles in the New York Times, Wall Street Journal, and others are cataloguing the tremendous and chaotically duplicative “overbuild” of the data centers, relative to the U.S. economy’s physical capacity for generating and delivering the amounts of electric power which can feed them. And from below, popular opposition to the centers keeps rising, becoming a major issue in some upcoming Congressional elections.

In strict terms, the data center “buildout” is the biggest infrastructure investment in history (no Trump exaggeration!), estimated equal to the national highway program of the 20th Century and the railroad expansion of the 19th Century combined—and virtually all done by private capital. The “hyperscalers” of A.I., Oracle, Nvidia, Microsoft, Amazon, Meta, Google, etc., are spending 7-8% of GDP on a single infrastructure purpose—information—over the past two years, in a nation which has never before spent more than 3% of GDP for all purposes of infrastructure, public and private, put together. All of it is borrowed, and large portions of it now are junk debt.

Here is the U.S. financial landscape:

— AI debt: the “Mag7,” as they are called on Wall Street,—lost more than $5 trillion in stock value the last several weeks alone. Despite their very profitable Silicon Valley “legacy” tech businesses, these companies now suffer negative free cash flow: Meta -$2.12 billion third quarter; Amazon -$42.4 billion; Alphabet -$9.07 billion, for examples. In other words, they have to borrow all of the nearly $1 trillion per firm of these companies’ committed data center spending, from the bond markets, at high interest rates.

— U.S. defaults in the category of debt called “private credit,” meaning what used to be called “non-bank lending,” have reached a record: 6.3% of all $2.7 trillion private debt is in default. This category is now completely overlapped with junk debt (or “leveraged lending,” debt borrowed by already overindebted companies).

— Crypto: the largest crypto exchange, Binance, lost $6.3 billion to a hack, which it covered with its own money, but upon reopening, it has lost about $6 billion more to withdrawals. The hack was reportedly carried out using AI agents: The largest hack in U.S. history, done by AI robots. What could they not hack?

— Interest rates stopped their steady climb two weeks ago and have, relatively speaking, gone into a slow dive, with the reality of Federal Reserve quantitative easing being masked by what is being called a “flight to safety out of AI stocks.”

If the A.I. “bubble” of debt soon bursts, what will remain of the new “technology” of artificial intelligence, and of the United States’ economy? According to Elon Musk in thestreet.com, podcasting with biographer Walter Isaacson, AI capability will continue to grow, regardless of a financial crash—which Musk admits is inevitable. “AI and robots,” he says, will then produce far more growth in goods and services than money-supply growth, causing general deflation, only to be solved by giving away bitcoin and/or other cryptocurrency to the population at large. This, Musk claims, will be the only way to pay U.S. sovereign debt—by aggressively inflating it away with bitcoin. It will otherwise implode soon.

A more telling sign was that AI, in the second quarter of 2026, did not produce any multifactor productivity growth at all—zero—in the U.S. economy.