Nvidia: A Fugazi?
Of course, the bearish investor is "shorting" the chips.
At ReadOn, we don’t just report the markets. We help you understand what truly drives them, so your next decision isn’t just informed, it’s intelligent.
Nvidia just made $5.4 billion appear out of thin air. The only problem is that Michael Burry thinks it disappeared just as quickly.
In January 2026, Nvidia hit one hell of a mega-deal. The chipmaker would sell a cluster of its most powerful processors. Elon Musk’s xAI would get the compute power to train its Grok AI models. Apollo Global Management would arrange the financing. Everyone involved gets a nice payday. AI advances. Story over.
Except the story is not as clean as it seems. Not for Michael Burry at least.
Source: X
Burry, the man who saw the 2008 housing crash coming while the rest of Wall Street was still throwing parties, had one word for the whole arrangement: Fugazi. That’s Italian slang for “fake.” Used most memorably in Donnie Brasco, it describes something that looks real, feels real, but is fundamentally an illusion. On X, Burry responded to CNBC’s Jim Cramer defending Nvidia with the following: “It is all Fugazi. How to make tens of $billions worth of NVDA GPUs disappear from balance sheets in 8-12 byzantine steps.”
Now many companies make many deals, and many bear investors will have a negative opinion on them. But what makes this deal and its accusation worthwhile are the potential victims. The old pensioners of the USA.
Let’s see what’s going on.
The Deal Between Nvidia and xAI
The deal took place between famed chipmaker Nvidia, and Elon Musk’s xAI, which is an AI company that operates as an AI division under SpaceX. Now, the companies involved in this deal aren’t the problem. It’s the financing and legal structuring that makes things complicated.
On January 7, 2026, Apollo announced that it had led a $3.5 billion financing package for an entity called Valor Compute Infrastructure L.P. (VCI). VCI is a specialized investment fund managed by operational growth firm Valor Equity Partners. Established in 2026, VCI was created to acquire, own, and lease massive-scale AI hardware. VCI, in turn, used that capital, plus $1.9 billion of equity anchored by Nvidia itself, to acquire over 100,000 Nvidia GB200 GPUs for a combined $5.4 billion. Those GPUs were then leased to a subsidiary of xAI Corp under a five-year triple net lease.
The purpose: to power xAI’s second data center and train Grok, which Nvidia noted has shown “strong performance across benchmarks.” Nvidia booked the full $5.4 billion as completed-sale revenue. xAI got access to one of the most powerful compute clusters anywhere in the world.
Apollo, which manages approximately $908 billion in assets, described the transaction as a “hallmark, downside-protected investment in the AI infrastructure space.” Valor’s CEO Antonio Gracias called VCI an opportunity to invest in “critical AI compute infrastructure with quarterly cash distributions and upside through ownership of the compute assets.”
Big number. Big deal. Big future for AI. And yet.
This is also the same Nvidia whose stock trades at a price-to-earnings multiple that assumes it will dominate global AI infrastructure for the next decade without a single serious challenger materializing. It is not a stock priced for a Cisco moment. Which is exactly what Burry thinks is coming.
The Ownership Problem
Here is what nobody put in the headline: xAI doesn’t own any of those chips.
The legal title to all 100,000-plus GB200 GPUs sits with VCI, a shell special purpose vehicle with no operating business, no independent revenue, and no reason to exist except to hold the chips on paper. xAI only leases them. None of the GPUs, and none of the debt used to buy them, appear anywhere on xAI’s balance sheet.
This matters more than it sounds. xAI folded into SpaceX in February 2026, and SpaceX subsequently filed its IPO prospectus with the SEC in May 2026, targeting a valuation between $1.75 trillion and $2 trillion, which could be the largest public debut in history. The world’s most anticipated IPO is of a company whose entire AI division leases, rather than owns, its core AI computing infrastructure.
Now comes the financing layers. Apollo arranged $3.5 billion in debt, secured against those GPUs. It then packaged that debt into securities. And here is the part Burry flagged most loudly: Apollo sold those securities to Athene, its own insurance subsidiary. Apollo structured the deal, and then sold the risk to itself. Or more accurately, to the balance sheet that American retirees fill with their annuity premiums.
Something worth noting here is that Valor Equity Partners is not some fly-by-night operation. It manages roughly $55 billion in assets as of December 2025 and has a long history with Elon Musk’s companies. But VCI, the specific vehicle at the centre of this deal, is a new fund created for this transaction, with no prior operating track record, no diversified revenue base, and a single concentrated bet on xAI’s ability to generate returns from its AI compute. The entire structure rests on one question: does Grok make enough money to service a five-year GPU lease?
As if this one existential question isn’t bad enough, things get more complicated when we look at the origins of Athene.
What The Fugazi?
Athene is the company Apollo sold its debt securities to. Athene is also how ordinary people’s retirement savings end up in extraordinary places.
When Americans buy annuities from Athene, handing over lump sums in exchange for guaranteed retirement income, that capital becomes Athene’s investable float. Athene deploys it into assets. In this case: AI infrastructure debt secured by Nvidia GPUs held inside a Valor shell vehicle, leased to a subsidiary of a company planning the biggest IPO in stock market history.
Source: X
Burry titled his diagram “The Retiree/Apollo/Nvidia/Bermuda/AMAPS/xAI Pipeline.” The name tells you everything about who sits where in this chain. He flagged four numbers inside Athene’s structure that he argues should alarm any serious investor.
First: $74.2 billion in total reserves. Second: $217 billion in assets shifted to a Bermuda-based captive insurer, sitting outside normal US state regulatory oversight. Third: $103 billion, or 34.7% of those assets, classified as Level 3, or illiquid, unobservable, valued by internal model rather than any real market price. And fourth: the entire structure runs at approximately 16x leverage.
Source: X
The pipeline, as Burry maps it, runs like this. Nvidia records a clean $5.4 billion sale. xAI secures $5.4 billion in compute capacity with nothing on its books. Apollo earns fees across origination, structuring, distribution, and servicing. And American retirees, who signed up to receive a fixed monthly cheque in their old age, sit at the very bottom of a 16x leveraged structure of illiquid assets they cannot independently price, unknowingly carrying the credit risk on a GPU lease tied to the success of Elon Musk’s AI model.
That is why Burry calls it fugazi. Not necessarily fraudulent. Not necessarily illegal. But fake, in the sense that Nvidia’s $5.4 billion revenue appears clean precisely because twelve clever financial steps have obscured where the actual risk lives. The chips are real. The revenue is real. The risk, though, has been relocated somewhere quieter.
The Takeaway
Burry’s wider warning is a historical one. He keeps returning to Cisco in 1999, the undisputed infrastructure company of the internet era, with dominant hardware, real revenues, and universal analyst admiration. Its stock fell more than 80% from its peak. He believes Nvidia’s position rhymes with that story: genuinely dominant product, but revenue increasingly dependent on structured finance to keep the headline numbers growing.
Source: Multpl
The S&P 500’s Shiller CAPE (Cyclically Adjusted Price-to-Earnings) ratio sat at roughly 41.6 in May 2026. This metric, developed by Nobel laureate Robert Shiller, divides the index’s current price by the average of its inflation-adjusted earnings over the past 10 years. At over 40, the market is trading at a significant premium compared to its 20th-century benchmarks. It’s also the second-highest reading in over 140 years of US market data, just below the dot-com peak of 44.2 hit in December 1999.
Nvidia has not commented on Burry’s analysis. Apollo has not publicly addressed the specific Athene linkage. VCI is a private fund, with no disclosure obligation.
The deal may work out perfectly. xAI may generate more than enough AI revenue to service every lease payment. Athene’s Bermuda assets may be entirely stable. Apollo’s $40 billion-plus deployed into next-generation infrastructure since 2022 may prove out exactly as planned. Someone, after all, has to finance the trillions of dollars the world’s data centres will need in this decade.
But Burry’s point isn’t really about whether this particular deal fails. It’s about the possibility that when the biggest beneficiary of the AI boom runs its business through enough financing layers that the downside becomes invisible butthe risk remains. It’s just been pushed under enough layers to move legally into the retirement accounts of people who were told they were buying safety. People who don’t really have any way to know that the risk is cleverly hidden, not gone.
That’s the fugazi.
Until the next one, ReadOn!