AI Black Paper Jim Rickards

Here is the idea.

Jim Rickards says roughly $200 billion in AI debt sits inside off-balance-sheet structures. He says those structures look a lot like the SPVs Enron used before it collapsed. And he says July 29 — Meta’s earnings date — might be the day the market wakes up to it.

The ad says an 80% Dow drop. It says 600% gains during the crash. It says July 29 at 6:30 PM.

Let me walk through each number. Not to fact-check it. To understand what those numbers actually mean.

The 80% number

An 80% drop in the Dow takes the index from roughly 33,000 to under 7,000. That is not a correction. That is a generational collapse.

The Dow has never fallen 80% in a single event. Not in 1929 (that took three years). Not in 2008. The closest modern parallel is the Nasdaq from 2000 to 2002 — a 78% drawdown over two and a half years. Tech-heavy companies that looked invincible lost nearly everything. Cisco dropped 90%. Intel lost 80%. Sun Microsystems went from $200 to $4.

Rickards is reaching for that parallel. He is not saying it happens overnight. He is saying the structure of AI debt bears the same fingerprints as the dotcom unwind — massive capital spending, opaque financing, and a narrative that breaks when spending stops.

The number lands harder when you put it in context. An 80% Dow drop would be the deepest in American history. It would erase the gains of the entire 2010s bull market. It would mean 401(k) balances become fractions of what they were.

That is the claim. It is extreme. Rickards knows it is extreme. That is why it is the headline.

The 600% number

This one is harder to put a frame around because Rickards has not disclosed the mechanics.

600% in 12 months during a market crash means one of two things.

Scenario one: you buy puts or inverse ETFs when the thesis starts to break. A deep-in-the-money put on the Dow could 10x or 20x if the index falls 30% or more. If you compound that through a year of volatility, 600% total return is large but not impossible. I have seen 10-baggers on single puts during the 2008 meltdown.

Scenario two: you short specific names that Rickards sees as the most overleveraged in the AI buildout. If Meta, Nvidia, or a handful of data-center REITs drop 60-80% from current levels, a concentrated short book could return multiples. That is harder to execute because short squeezes can wreck the timing.

The 600% number means there is a specific trade behind the free presentation. You do not promise that kind of return on a general-market thesis. You promise it because you have a position in mind. Whatever the paid product is — a premium letter, a model portfolio, an options service — it is built around that trade.

I do not know which trade yet. But the number tells me there is one.

July 29 at 6:30 PM

July 29 is Meta’s Q2 2026 earnings date. 6:30 PM is the after-hours release window.

Rickards chose this date deliberately. Meta is the most leveraged of the AI hyperscalers — $86.8 billion in total debt, $125-145 billion in 2026 CapEx guidance, and roughly $80 billion in cumulative Reality Labs losses since 2020. If any single company carries the AI spending narrative on its shoulders, it is Meta.

The logic: Meta reports. If revenue misses, or if Q3 guidance disappoints, the market revalues the entire AI capex cycle. Stocks that depend on Meta’s data-center orders — Nvidia, AMD, Broadcom — get repriced. The off-balance-sheet debt structures that Rickards has been documenting face their first real test.

The date is real. The catalyst is real. What happens is unknown.

The thesis behind the numbers

The AI buildout since 2023 has been the fastest infrastructure spend in history.

Data centers went up in months, not years. Hyperscalers signed multiyear lease commitments that ran into the billions. Nvidia shipped $18 billion in H100 GPUs in a single quarter. By some estimates, total AI-related debt issuance in 2025 approached $200 billion.

The question Rickards asks is simple: where did the money come from?

His answer: a lot of it came from off-balance-sheet structures — special-purpose vehicles, synthetic leases, vendor-financing arrangements, and securitization deals that bundle data-center cash flows into bonds. These are the same financial engineering tools that Enron used to hide debt in the 1990s. Same tools that Lehman used to hide leverage before 2008.

The difference is size. The AI SPV market is measured in hundreds of billions. If those structures break — if a major hyperscaler misses earnings, or if a data-center REIT defaults on its leases — the unwind could cascade through a system that most investors cannot see.

That is the thesis. Not “AI is a bubble.” Not “the crash is coming.” The thesis is: $200 billion in hidden debt will eventually test the real economy, and July 29 is the first credible test date.

What to make of it

I have spent my career around big macro calls. Some hit. Most miss. The ones that hit usually start with a structural argument — not a timing argument.

Rickards makes a structural argument. The AI debt buildout is real. The off-balance-sheet structures are real. The Enron parallel has historical precedent. Those facts do not depend on July 29.

Does 80% sound extreme? Yes. Does 600% sound hard to deliver? Yes. But the thesis underneath both numbers is the same: a debt structure this large, this opaque, and this concentrated has never unwound quietly.

July 29 is 19 days away. If the thesis holds, the numbers start to matter. If it does not, the thesis waits for the next catalyst.

Either way, the idea is worth sitting with.