For a little under a year, everyone — myself included — has compared NVIDIA to Enron, largely because NVIDIA insisted, in detail, that it was nothing like Enron, WorldCom, or Lucent, a potent example of the Streisand Effect that would be much funnier if NVIDIA wasn’t holding up more than 7% of the value of the NASDAQ.
And as I covered in the first part of the Hater’s Guide To NVIDIA last year, there are material concerns about how the company makes money today and will continue to do so in the future.
I will concede that NVIDIA isn’t exactly like Enron in the sense that it isn’t, to my knowledge, doing anything outright fraudulent, like attempting to hide massive amounts of debt inside SPVs as Enron did with its “Raptors,” which I must be clear are distinct from the SPVs used in AI data center debt, though I’ll add that something being legal doesn’t make it a good idea or ethical.
That being said, NVIDIA CEO Jensen Huang has employed many of the same tactics used by Lucent, Nortel, and many of the big dot-com busts, but has been smart enough to make everybody else carry the risk.
Instead of doing direct vendor financing like Lucent did with Winstar (where it effectively loaned its customers money to pay it with), NVIDIA funded neoclouds like CoreWeave, Nebius, and IREN, operating as an early stage investor, IPO anchor, post-IPO investor, $6.3 billion customer and data center lease backstop, allowing them to raise tens of billions of dollars’ worth of debt from overly-eager asset managers and banks, allowing it to do basically the same thing as vendor financing without having to take on any of that messy risk.
These deeply-unprofitable, cash-intensive, debt-riddled companies exist for one purpose — to raise debt to buy NVIDIA GPUs — and would have fallen apart without the AI hype cycle and NVIDIA’s continued backing. Per Kakashii:
In an April 2026 interview with Dwarkesh Patel, Jensen Huang acknowledged my thesis and said it out loud: “In the case of clouds, if we didn’t support CoreWeave to exist, these neoclouds, these AI clouds, wouldn’t exist. If we didn’t help CoreWeave exist, they would not exist. If we didn’t support Nscale, they wouldn’t be where they are today. If we didn’t support Nebius, they wouldn’t be what they are today.” And, confirming that Nvidia wants lots of neoclouds, not just one: “Don’t pick winners. Either let them all take care of themselves, or take care of all of them.”
That is, in Nvidia’s own CEO’s words, the operating logic behind this sector. Rather than build a cloud business on its own balance sheet, which would put Nvidia’s own results directly at risk if utilization or pricing disappointed, Nvidia has helped create and sustain an entire class of nominally independent companies that take on the capital outlay, the construction risk, and the debt of building AI data centers, while Nvidia supplies the chips, frequently invests equity alongside the debt, and in a growing number of cases finances the build out directly.
In other words, NVIDIA has managed to find a way to do vendor financing without ever having to provide any, finding willing supplicants in the various backers of CoreWeave and other neoclouds that would be willing to front the money, all under the mistaken belief that they were funding the next industrial revolution.
To explain exactly how it works, I’ll return to my imaginary scenario from the Big Short 2:
HUANG: So there’re these companies I invest in that, at least in theory, build data centers using my AI GPUs, but I need them to buy more GPUs, so I sign a contract saying that I’ll rent the GPUs back from them in the future. Because NVIDIA has such a strong balance sheet, these companies can raise billions of dollars to buy my GPUs just because I promised to rent them in the future, and the best part is all the risk is held by the companies and the investors. When I need more money, I just sign another contract, they raise more debt, I sell more GPUs.
BAUM: So — just so I have this clearly — you, the guy who makes the GPUs, invest in companies that exist pretty much to buy GPUs from you and rent them to customers. Except you’re the customer too, and a big one.
HUANG: That’s right. We call them neoclouds. S&P just revised CoreWeave’s outlook to positive.
BAUM: That’s fucking crazy.
HUANG: It’s not crazy — it’s awesome.
It’s a win-win-win for NVIDIA, its customers, and the bankers involved. CoreWeave gets to raise more debt and keep its investors strung along on the still-theoretical, ever-expanding timeline of a return on invested capital, bankers get a slew of fees for pulling together the deal, and NVIDIA guarantees itself billions of dollars of business.
And this approach is something where any investment by NVIDIA has a habit of being amplified by others — like Australian startup Firmus, which just raised $2bn from a bevy of investors (including NVIDIA, which had also backed an earlier round), Jane Street, and Blackrock, with a significant chunk of that money guaranteed to go towards NVIDIA GPUs. NVIDIA also participated in Firmus’s previous $300m round, although was not listed as a “cornerstone investor.”
Earlier this year, Firmus secured a $10bn debt facility, led by Blackstone. NVIDIA will be a net beneficiary of that debt raise, and I would argue that its participation in the company’s fundraising — as well as the various announcements of partnerships between the two — has been instrumental in both the company’s fundraising and its ability to secure debt.
Sidenote: According to an AI industry insider interviewed by AlphaSense, Nebius was experiencing financial difficulty — the existential kind — and was saved by NVIDIA, which swept in and offered it access to hardware under a revenue sharing model. The insider also notes that neoclouds hate said revenue sharing agreements, because they tend to stack the deck in NVIDIA’s favor.
You’ll notice I haven’t mentioned “AI” or “LLMs” up until this point, and that’s because technology has, for the most part, very little to do with these transactions. As I discussed in this week’s free newsletter, 70% or more of hyperscaler revenues are from OpenAI and Anthropic, and CoreWeave’s largest customers are Microsoft (for OpenAI), Google (for OpenAI), Anthropic, NVIDIA itself, and Meta. Customers are not coming to it for any particular technological moat or unique offering outside of its ability to sling more NVIDIA GPUs to the same customers that everybody else has.
While GPUs technically are used for AI training and inference, their relationship to NVIDIA is only as good as their ability to create more hype. As I discussed a few weeks ago, it has promised somewhere between 10x and 25x “operating cost savings” with every successive generation of GPUs, though it’s never really clear how that manifests or what it actually means, or whether any of that even matters to OpenAI and Anthropic, its largest customers by proxy.
Nevertheless, it’s pretty difficult to work out what each generation really changes. SemiAnalysis claims it “delivers 5.4x performance per MW and 5x performance per dollar against [the previous generation] GB200 NVL72,” but that’s for DeepSeek R1, a year-and-a-half old open source model that’s vastly smaller and less-powerful.
But that’s not really a problem, because all NVIDIA needs to do is keep up the appearance of innovation in as precise or imprecise a way to justify increasing prices with each new generation, and to convince people that they’re building “AI factories” as they fund data centers for customers that don’t really exist outside of the big AI labs. While NVIDIA has thousands of talented engineers building its GPUs and the associated software, the only real purpose is to create a vague sense of “more” and “bigger” and “more powerful” to justify racks of 72 GPUs that are more than twice the price of their predecessors.
That’s because NVIDIA is no longer a technology company so much as it is an asset management and marketing firm that happens to sell semiconductors. To that point, I believe that the comparisons to Enron, Lucent, and other dot-com flameouts are on the right path, but misses one very, very obvious comparison: GE Capital, the financial services of General Electric, specifically in the Jack Welch years that I covered two years ago in the Shareholder Supremacy.
Welch’s GE did whatever it needed to to survive, buying and selling companies to help boost GE’s earnings every quarter, and eventually grew into what David Gelles would call a “large, unregulated bank,” to the point that GE Capital was bringing in $425 billion in revenue in 2001 (about 50% of GE’s revenue), providing everything from direct leases of equipment to assuming its customers debts to investing directly in its customers, all to make sure that, well, said customers continued being able to buy GE gear.
Unlike GE Capital, NVIDIA has the advantage of a much, much simpler business model and far fewer products to sell, but said advantage is a problem for two brutal reasons: its customers are driven by desperation and a fear of missing out, and its remarkable revenue growth means that it must in turn grow by ridiculous amounts every single quarter from here to eternity.
Yet this problem is driving it to take increasingly-Welchian measures to make sure that demand keeps up with investor expectations. It (per the FT) just signed leases worth as much as $50 billion for a Texas-based data center built by Hut 8, which makes it likely that this capacity is being built for Anthropic, with which it already has multiple deals. In the same piece, the FT mentions that NVIDIA is in talks to backstop $250 billion in compute costs for a still-theoretical 10GW data center in Ohio.
And again, much like GE, NVIDIA uses its stellar credit rating (AA- - two rungs lower than GE at its height) to secure these deals, per the FT:
“They have the balance sheet to acquire power, and in doing so, ensure their product is deployed,” the person said, asking not to be named.
In the end, GE’s greater collapse led to lawsuits, SEC fines and revenue revisions, all as a result of its “aggressive” accounting practices. For example, it was forced to restate its 2016 and 2017 earnings as a result of “new accounting standards” it instituted as a result of an SEC investigation into its insurance and power divisions that eventually cost it a $200 million fine, cutting a remarkable $4.24 billion off of earnings in the period.
While I’m not accusing NVIDIA of anything untoward, it’s impossible to ignore the sheer aggression of its circular financing and willingness to do whatever it takes to keep selling further GPUs. NVIDIA is now a semiconductor manufacturer, a venture capitalist, a lender of last resort,
Today’s premium newsletter is the story of NVIDIA’s descent into circular madness, and how Jensen Huang is increasingly becoming the Jack Welch of AI.
This is Part 2 of The Hater’s Guide To NVIDIA, or WUDA CUDA SHUDA