Last night, NVIDIA made a big splash.
By teaming up with Blackstone, BlackRock, Goldman Sachs, Apollo, KKR, and others—six Wall Street titans—they’re looking to set up a $500 billion AI compute financing platform.
What it means is: you customers who want to buy my GPUs but don’t have the money—don’t worry. I’ve brought in the big daddy of investors to lend you the money to buy my goods.
Huang Renxun personally took charge, and none of the six major powerhouses said no.
So what?
NVIDIA’s stock price dropped 2.86%, and its market value evaporated by $130 billion.
In late trading, the Philadelphia semiconductor index plunged; Coherent fell by more than 14%, and Lumentum fell by more than 8%.
The $500 billion boost to sentiment resulted in $130 billion in value evaporating.
Has the market gone crazy?
The market isn’t crazy. The market is just doing the math.The name of this ledger is called “circular financing.”
Translated into plain human language, it’s: left hand to right hand.
Nvidia sells chips to customers while pulling Wall Street money to lend to those customers, so the customers can keep buying Nvidia chips.
I borrow money to you so you can buy my goods.
You use my goods as collateral, then come to borrow money from me to buy even more goods.
Doesn’t it sound familiar?
Sounds like before the 2008 financial crisis, when banks packaged subprime mortgages into CDS and sold them to investors, and then the investors used those CDS as collateral to borrow even more money, right?
Well-known short seller Jim Chanos directly blasted it on social media, likening Nvidia’s playbook to financial engineering during the 2008 financial crisis.
He implies it very clearly: if the AI bubble bursts, related parties may repeat the fate of Wall Street executives from back then—being held accountable by Congress.
Even more chilling is the second question:
Why doesn’t Nvidia borrow from traditional banks and instead bring in private equity?
Some analysts believe this shows banks no longer have much confidence in the massive investment in data centers, making it harder for companies to get loans from banks.
Think about it—
If AI data centers were truly a no-risk, guaranteed-profit business, banks would have rushed in long ago.
Does Nvidia really need to personally pull in a bunch of private equity firms to set up a financing platform?
Private equity is willing to lend money that banks won’t.
It’s not that private equity is smarter—it’s that private equity money is more expensive, shorter-term, and more urgent.The third fear is hidden even deeper:
Is this $500 billion “new demand,” or “prepaid overconsumption”?
Since this year began, Nvidia has been pushing a series of massive transactions:
With SK Group: cooperation on $500 billion AI infrastructure.
With OpenAI: discussing providing $250 billion in financing guarantees for leasing data centers, while also discussing $350 billion in chip procurement financing.
For CoreWeave and Nebius: invest $2 billion each.
Just these few deals add up to more than $1 trillion.
What is Huang Renxun betting on?
Bet that in 2–3 years, AI applications will explode, and customers will be able to recoup their costs.
But what if, in 2–3 years, AI applications still aren’t profitable?Then this $500 billion isn’t bridge financing—it’s a “debt black hole.”
When that happens—left foot stepping on right foot and you fall—it's not just Nvidia that will be hurt.
What’s the most ironic thing of all?
Even Nvidia knows how big the risks are.
After the news was announced, the 5-year credit default swap (CDS) price used to gauge Nvidia’s credit risk surged to 77.2 basis points, the biggest single-day jump in two weeks.
Since late May, the cost of insuring Nvidia’s debt against default has doubled, rising from 41.6 basis points to 77.5 basis points.
The cost of insuring Nvidia’s debt doubled in two months.
Even insurance companies are scared of it.
Let me say something that hits hard—
When it comes to pricing in capital markets, it’s never about how big the “story” is, but about how tangible the “risk” is.A $500 billion narrative sounds beautiful.
But the $130 billion market cap evaporation shows the market is voting with its feet.
Nvidia wants to turn itself from “selling shovels” into “making loans.”
But the market only cares about one thing: can this money really be collected?
Finally, let’s say something even harsher—
Nvidia is now tied to half of Wall Street’s life, tied to America’s tech hegemony, and tied to pension funds.
If it ever gets to the point of blazing fires on the front lines, the Fed will most likely step in to backstop it.
Isn’t this the same script as that of the past for Fannie Mae?
Win big and you profit enormously; lose and the whole public pays the bill.
Huang Renxun turns AI into infrastructure, and GPUs into “digital real estate.”
But what’s the biggest feature of infrastructure?It’s too important to fail.
So who’s paying the bill? Every single taxpayer.
If, in 2–3 years, AI applications still aren’t profitable, who fills the $500 billion hole?
Let’s chat in the comments.