Until yesterday, a GPU was just an expensive piece of electronics. You bought it, installed it in a rack, used it for two or three years, and then replaced it. Today, six of the biggest players on Wall Street are saying that GPUs are an investable asset class.
And they are willing to mobilize more than $500 billion around that idea.
Yes, you read that correctly.
Nvidia signed memorandums of understanding with six firms that need no introduction: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR.
The goal? To create the world's first large-scale financing platforms for computing power and mobilize more than $500 billion in third-party capital to build data centers and purchase Nvidia equipment.
And here's the key detail. The money is not coming out of Nvidia's pocket. Nor is it coming from the balance sheets of its customers. It is coming from institutional investors, insurance funds, and private capital.
Each of the six firms will make its own lending decisions. Nvidia simply connects its customers with the financing providers. However, Nvidia will also have the option to guarantee up to 25% of each loan, which could lower borrowing costs for companies that previously had to rely solely on their own creditworthiness.
There is another condition that shows how serious they are. Anyone receiving financing must use Nvidia-approved architecture. Why? So that if something goes wrong, another operator can step in and run the system.
Now pay attention to this. This was not just a press release.
All of them appeared together live on CNBC in a rare joint interview. Jensen Huang standing alongside Larry Fink, David Solomon, and the others.
"This is the first time chips are becoming an investable asset class," Huang said. "They are now revenue-generating assets. They are productive, durable, transferable, and flexible."
Larry Fink of BlackRock went even further. He compared it to the rise of the mortgage-backed securities market in the 1970s when he was just starting his career.
"I see this as the next frontier of financial engineering," he said, adding that capital needs to be raised as quickly as possible.
Jon Gray of Blackstone explained it more simply. We will begin looking at computing power the way a banker looks at a house backing a mortgage. He also noted that AI usage across Blackstone portfolio companies increased sevenfold over the past year.
Waldemar Szlezak of KKR openly discussed securitization.
You take a stream of revenue, divide it into pieces, and sell it to investors.
Bruce Flatt of Brookfield summed it up with a single statement:
"There are hundreds of trillions of dollars in the world."
Now you might ask: did they provide details?
No.
No interest rates, no names of borrowers, no locations, no timeline.
This is where things become even more interesting.
The biggest problem facing AI right now is not technology.
It's money.
A one-gigawatt data center costs more than $50 billion, and McKinsey estimates that global spending on AI infrastructure could reach $7 trillion by the end of the decade.
To understand the scale, this year alone Alphabet, Amazon, Meta, Microsoft, and Oracle have raised more than $150 billion through debt and equity offerings.
Intel recently announced a $15 billion stock offering and later increased it to $20 billion.
In other words, traditional funding sources are beginning to feel strained. This agreement unlocks another half-trillion dollars from elsewhere.
At the same time, Nvidia removes much of the financing risk from its own balance sheet. Until now, there were concerns that the company might eventually have to finance customer purchases directly.
Now the debt sits on Wall Street's books, while Nvidia's cash flow remains free.
And consider the message being sent.
When leaders like Fink and Solomon publicly support something, the market pays attention.
They do not view AI as a bubble. They view it as infrastructure.
However, there are significant risks.
The first is cyclicality.
Funds lend money to companies so they can buy Nvidia chips, allowing Nvidia to report record revenues.
But what happens if AI models fail to generate meaningful revenue from real customers? Then the loans may not be repaid.
The second risk, and perhaps the most serious, is obsolescence.
A highway can last for decades. A chip may be surpassed within two or three years.
How do you repay a ten-year loan with equipment that has a useful life of only three years?
This concern has also been raised by Michael Burry, who argued that major technology companies are overestimating the useful life of their chips while underestimating depreciation expenses.
The third risk is that easy money can create oversupply.
Too many data centers get built, future demand is pulled forward into the present, and eventually the market crashes.
And it is worth noting that these memorandums may never fully materialize.
A few months ago, Nvidia announced plans for an investment of up to $100 billion in OpenAI, but the final commitment ended up being closer to $30 billion.
Even the financiers themselves acknowledged the risks.
"There will be excesses, and there will be setbacks," said Jim Zelter of Apollo.
David Solomon added:
"There will be major companies that succeed, and major companies that turn out to be something very different from what people expected."
Despite all of that, the overall picture remains overwhelmingly positive.
What we are witnessing is a fundamental shift.
Artificial intelligence is no longer just software. It is becoming core infrastructure.
Just as electricity became infrastructure. Just as the internet became infrastructure.
And for that transformation to happen, traditional financing models are not enough.
The $500 billion acts as a liquidity bridge.
It keeps the ecosystem alive during the build-out and training phase, until it reaches the stage where it can generate meaningful and sustainable profits.
