Michael Burry, the hedge fund manager who became a household name by shorting the housing market before the 2008 crash, has a new target. This time it’s not mortgage-backed securities. It’s GPU-backed securities.
On May 29, Burry published a detailed critique on Substack dissecting a $5.4 billion transaction between Nvidia, a special-purpose vehicle called Valor, and Elon Musk’s xAI. His verdict on the deal’s financial engineering: “fugazi.”
How the deal works
Nvidia sold over 100,000 GB200 GPUs to Valor, a newly created special-purpose vehicle established specifically to hold the chips. That sale generated $5.4 billion in revenue for Nvidia.
But Nvidia didn’t just sell and walk away. The company also injected $1.9 billion of its own equity into Valor as a limited partner. In English: Nvidia sold the GPUs to an entity it partially funded.
On the debt side, Apollo arranged roughly $3.5 billion in financing for Valor. That debt was then securitized, packaged up, and sold to Athene, Apollo’s insurance subsidiary. Athene manages annuity assets, which means the ultimate risk holders include retirees whose pension and annuity payments depend on stable returns.
The GPUs are then leased to an xAI subsidiary under a five-year triple-net lease arrangement. The triple-net structure means the lessee, xAI’s subsidiary, covers maintenance, insurance, and taxes on top of rent. More importantly for the accounting, this setup keeps the GPU assets off the balance sheets of both Nvidia and xAI.
Why Burry is sounding the alarm
Burry’s critique centers on several structural risks that he believes auditors and regulators should be examining more carefully.
The first is what he characterizes as round-tripped capital. Nvidia puts $1.9 billion into Valor, then books $5.4 billion in revenue from selling GPUs to that same entity.
The second concern is concentration risk. Valor’s entire asset base consists of Nvidia GPUs leased to a single customer, xAI. If xAI’s subsidiary stops making lease payments for any reason, whether financial distress, a pivot in strategy, or something else entirely, the whole structure unravels.
Then there’s the obsolescence problem. GPUs are not commercial real estate. A five-year lease on technology that could be superseded by next-generation chips within two years creates a mismatch between the financing timeline and the asset’s useful life.
And those debt holders, through Athene, include retirees. Burry explicitly flagged the risk to annuity holders, arguing that the securitized debt essentially transfers AI infrastructure risk to people who almost certainly didn’t sign up for it.
The broader context
Apollo has acknowledged its role in financing Valor’s acquisition and the leasing strategy for Nvidia’s compute infrastructure.
Burry was careful to note one thing that might surprise readers given the players involved: there are no crypto assets or tokens in this deal. His analysis stays firmly in the lane of traditional private credit and AI infrastructure financing.
What this means for investors
The immediate question for Nvidia investors is whether this type of transaction structure could inflate revenue figures in ways that don’t reflect genuine organic demand. If a meaningful chunk of Nvidia’s GPU sales involve entities that Nvidia itself is capitalizing, the quality of that revenue deserves a different kind of scrutiny than a straightforward sale to a customer spending its own money.
For retirees and annuity holders connected to Athene, the $3.5 billion in securitized debt represents real exposure to a single-technology, single-customer asset.