Jim Rickards says AI company debt is a hidden systemic risk.

That’s the thesis of his June 2026 AI Black Paper campaign. Rickards argues that the biggest tech companies in the world are borrowing more money than they can realistically pay back, and when credit markets figure this out, the whole thing unwinds.

If you’re a normal person, you hear that and think: “That’s what the crisis guy says.”

Fair response. Rickards has been warning about financial collapse for years. He’s been right before — 2008 being the big one — and wrong on timing more times than he’d like to admit.

But here’s where this gets interesting.

The biggest bank in America didn’t write a warning about AI debt. It built a product to trade it.

Let me explain why that matters.

What JPMorgan Actually Did

In March 2026, JPMorgan launched a basket of credit default swaps on the debt of five companies: Alphabet, Amazon, Meta, Microsoft, and Oracle.

A credit default swap is insurance on a company’s debt — you pay a premium and if the company defaults you get paid. You don’t have to own the underlying bonds or be exposed to the company at all; you’re betting that the debt goes bad.

JPMorgan structured this in $25 million blocks — $5 million per company. Hedge funds and institutional investors can now take a direct, liquid, and tradeable position against the debt of the five largest AI infrastructure builders on the planet.

This is not a research note. It’s not a cautious analyst warning about “elevated leverage in the tech sector.” It’s a product. JPMorgan built it because there was demand for it.

The Market Signal, Not the Warning

Rickards says AI debt is overextended and vulnerable.

JPMorgan says: “Here’s a way to bet on that exact outcome.”

That distinction matters. JPMorgan’s economists might have a different base case and their public research might be cautiously optimistic about AI, but the derivatives desk operates on a different truth. It prices what the market actually wants to trade — not what the bank’s official forecast looks like.

When a bank builds a basket of CDS on AI company debt, it’s observing a signal: enough sophisticated money wants to bet against this stuff that the product makes economic sense.

That’s independent validation of Rickards’ thesis — not from a newsletter but from the people who price risk for a living.

The Oracle Data Tells the Real Story

The clearest evidence of stress in AI credit markets isn’t a CDS price. It’s a debt deal.

JPMorgan and Mitsubishi UFJ Financial Group spent six months trying to place a $38 billion debt package for Oracle data center projects in Texas and Wisconsin. That is Bloomberg-reported, real-world $38 billion in actual dollars — one of the largest AI infrastructure financings on record, and it took six months to place a single debt deal. That is not a sign of a smoothly functioning market. That’s a sign that syndication desks had to work every angle, call every contact, and potentially price at a discount to get it done. Some lenders reportedly explored selling portions at a discount just to free up balance sheet space, according to The Wall Street Journal.

Oracle is the weakest credit in JPMorgan’s CDS basket, carrying $124 billion in borrowings. Its Baa2 rating is just two rungs above junk. Among the five hyperscalers, it’s the only one where debt-to-enterprise-value sits above the cohort’s sub-5% average.

If the market struggled to absorb $38 billion for Oracle, what happens when the next wave hits?

Morgan Stanley’s Admissions

Morgan Stanley said something in its “Thoughts on the Market” podcast that deserves attention: “If credit markets lock these companies out, the AI supercycle ends.”

Translation: The entire AI buildout depends on bond buyers continuing to lend.

Not on revenue, adoption, or productivity gains — but on the willingness of credit markets to keep financing a buildout that hasn’t proven its return on investment.

Morgan Stanley’s own numbers underscore the scale. The five hyperscalers issued $121 billion in bonds in 2025 — up from a $28 billion annual average between 2020 and 2024. That’s a 4x increase. Tech debt hit 18% of total investment-grade supply, the highest share on record. Bank of America projects the hyperscalers could issue $140 billion to $300 billion annually for the next several years.

The bond market is the engine. If it stalls, everything stops.

The Honest Take

Rickards’ thesis has independent validation from the largest bank in America.

JPMorgan built the product not as a warning but as a trade — and that distinction does not make Rickards right about timing or severity, but it does mean the structural risk he is pointing at is real enough that the professionals who price risk for a living are building dedicated products around it.

JPMorgan didn’t build this product because someone had a clever theory. It built it because hedge funds and institutional investors wanted to put real money behind the thesis that AI debt is mispriced.

The difference between Rickards and JPMorgan is simple: Rickards writes about it while JPMorgan trades it, and one of those tells you more about what the market actually thinks.