Michael Burry, famous for foreseeing the 2008 subprime collapse, has called 'fugazi', meaning fake, a financing structure linking the chipmaker Nvidia, the asset manager Apollo and xAI, Elon Musk's artificial intelligence company. At the heart of his warning: a loop in which credit risk ends up resting on the retirement savings of millions of Americans.
1 The circular mechanics
How $5.4 billion of chips turns into an almost closed loop.
The vehicle
Valor raises $5.4 billion to buy Nvidia chips
In January 2026, Valor Equity Partners raised $5.4 billion through a special purpose vehicle, Valor Compute Infrastructure. The vehicle buys more than 100,000 Nvidia GB200 processors, then leases them to an xAI subsidiary to train the Grok model. xAI thus gains considerable computing power without weighing down its own balance sheet.
The three links in the loop
①
Nvidia invests about $1.9 billion of equity in the vehicle that buys its own chips, while booking $5.4 billion of revenue immediately.
②
Apollo provides $3.5 billion of debt financing, which it securitizes and then places with its insurer Athene.
③
Athene sells annuities to American savers, often retirees, whose payments now underpin this computing infrastructure.
The loop
Nvidia capitalizes the vehicle that buys its chips, books that sale as revenue, and the credit risk then migrates toward retirement annuities. It is this circularity, not any illegality, that Burry finds troubling.
2 'Fugazi': Burry's thesis
Burry alleges no fraud, but an organized opacity.
Burry's word
Risk shifted, off balance sheets and away from market prices
On his 'Cassandra Unchained' blog and on X, Burry describes a structure designed, in his view across eight to twelve steps, to move credit risk away from the big players' balance sheets, away from observable market prices, and ultimately onto people who did not choose it.
The risk has not vanished: it has simply changed hands, from the balance sheets of the big players to the annuities of millions of retirees.
3 The Enron parallel, and its limits
The comparison is tempting, but it misleads on one decisive point.
Echoes of 2001
The shadow of Enron's off-balance-sheet vehicles
Some observers liken this structure to the special purpose entities Enron used to hide debt and losses before its spectacular 2001 bankruptcy. The common thread is the opacity and circularity of the flows. The difference, however, is essential.
Enron, 2001
Proven accounting fraud: falsified balance sheets, fictitious or overvalued assets, gains resting on stock manipulation. Enron actively concealed its losses.
The AI structure, 2026
Legal and disclosed operations: the flows are public and registered with regulators, the assets are real (high-value chips in strong demand) and the revenues are genuine. This is not fraud, but an authorized transfer of risk.
The real issue
These structures are legal and increasingly common in large AI infrastructure builds. The issue is therefore not legality, but transparency toward those who bear the risk.
4 The real risks for savings
The absence of fraud does not remove the loop's fragility.
🔥
xAI's cash burn
Close to one billion dollars burned per month, and $6.4 billion of losses in 2025.
🏝️
Athene in Bermuda
About $217 billion of assets moved to a captive insurer in Bermuda, outside ordinary US regulation.
📉
Assets without a price
$103 billion, or 34.7% of the portfolio, classified as 'level 3', with leverage of 16.6 times.
Two major vulnerabilities
①
Dependence on xAI. If AI monetization fails to keep pace with the investments, or if the GB200 chips are made obsolete by a new generation, the leasing chain weakens and leverage amplifies the losses.
②
Exposure of savings. Most future retirees are unaware that their annuity underpins highly volatile technology bets. The question is not legality, but informed consent.
Asymmetric positions
Sophisticated investors, including Burry himself, have taken short positions. In a severe correction, those positions would generate large profits, while ordinary retirement portfolios would bear the losses.