Michael Burry's Apollo chart, fact-checked against the filings

The chain, one hop at a time

The chart traces a single path, and each step is mundane on its own. It is the full chain that raises eyebrows.

It starts with an annuity. You hand an insurance company a chunk of your savings, and it promises you a steady income for years. Safe, predictable, the opposite of risky — that is the whole pitch. But the insurer only profits if it earns more than it promised to pay you. So your savings don’t sit in a vault; they get put to work.

The insurer is Athene, owned by Apollo (ticker APO), a firm managing over a trillion dollars whose specialty is private credit — loans that aren’t traded on any exchange. No exchange means no public price. You can’t simply look up what they are worth; their value is, to a meaningful degree, whatever a model says.

Then it routes through Bermuda. Athene reinsures a large block of these promises to a sister company it owns in Bermuda, where the capital rules let it hold a thinner safety cushion for the same risk. More leverage, same liabilities.

And it ends in GPUs. Where does the money increasingly go? Into AI. Apollo built a roughly $5.4B company called Valor with one job: buy Nvidia’s (NVDA) newest chips and rent them to xAI, Musk’s AI lab. Your safe little annuity is, a few steps down the chain, a stack of GPUs humming in a data center.

The part that worries Burry: circular financing

Here is the wrinkle. Nvidia doesn’t just sell those chips — it also puts its own money into the vehicles buying them. So the chip seller is helping fund its own customer. Cash goes out one door as an investment and comes back through another as revenue. Burry calls it circular financing, and his concern is that it can make end demand look stronger than it is.

Why this is supposed to be your problem

If grandma’s annuity is guaranteed, why should she care where the money goes? Here is the catch the chart leans on: that guarantee is a promise from Athene, and unlike a bank deposit it is not government-insured. It is only as safe as Athene itself. Her money isn’t in a vault; it is backing these exact positions. If the AI-linked loans sour and leverage multiplies the losses, what is on the line isn’t a banker’s bonus — it is the retirement check. That is the emotional core of the chart, and it is a fair worry to raise.

Verdict: mostly true

So is the chart real? Mostly, yes. The deal, the players, the giant Bermuda numbers, even Apollo openly describing parts of the structure as “downside-protected” — all of it checks out against the filings. This is not a secret. It is disclosed, and it is roughly how Apollo and Athene normally operate.

Where the chart cheats

But it does cheat in a couple of places, and this is where having the primary sources matters.

• It glues two different deals into one. The narrative quietly merges separate transactions to make a single, scarier picture.
• Its most alarming numbers aren’t in any filing. The headline figures — roughly $217B supposedly shipped to Bermuda, and 20x leverage — do not appear in the disclosures. They are Burry’s own estimates, behind his paywall.
• It picks the scariest version of every number. The alarming “16.6x leverage” drops to about 11.6x if you simply count all the equity. The figures are real; the framing is turned up to eleven.

The bottom line

This isn’t Enron. It is, in Burry’s own framing, closer to Cisco in 1999 — a great company in a real boom, whose stock still fell about 75% when the cycle turned. The worry underneath is simple and fair: a mountain of borrowed money, a big slice of assets nobody can independently price, and chips that lose most of their value in a few years, all sitting behind everyday retirement savings. Reasonable people can look at that and disagree about how much to worry. What you should not do is take the screenshot’s scariest number as gospel.

How to fact-check a chart like this yourself

The reason we could grade the chart instead of just reacting to it is that every claim was traced back to a primary source — investor materials, SEC filings, the regulator’s own document — and the verified numbers were separated from the paywalled estimates. You can run the same play. With a markets database connected to an AI agent over MCP, the workflow is plain English:

“Pull Apollo’s latest 10-K and 10-Q. What does it disclose about reinsurance to Bermuda affiliates and about Level 3 / hard-to-value assets? Show me the figures and the exact sections, and flag anything the filing does not state.”

The agent retrieves the real sections instead of recalling a plausible summary from memory. When a claim has no filing behind it, the absence is the finding. That is the difference between checking a number and vibing on a number — and it is the same discipline behind every playbook on this blog, from reading 13F filings to grading insider buys.