Wall Street Wants to Turn Nvidia Chips Into a New Asset Class. Here's Why That's Complicated.
Six of the world's biggest financial firms are joining Nvidia to bundle $500 billion in chip financing. The pitch is that GPUs are now investable assets like buildings or bonds. The catch: Nvidia's own CEO said something very different about chip lifespans just last year.
Key points
- Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR are working with Nvidia on a $500 billion financing package to treat AI chips as investable assets.
- Nvidia CEO Jensen Huang now says one of the company's chips, the A100 from 2020, has an economic life of up to ten years, contrasting sharply with comments he made last year dismissing older chips as nearly worthless.
- The deal rests on memorandums of understanding, which are statements of intent rather than binding contracts, and a similar $100 billion Nvidia MOU with OpenAI announced in 2024 did not go ahead.
- Short seller Michael Burry puts the useful lifespan of a chip at two to three years; IBM's CEO says five years; Huang now claims ten.
- A cloud provider nearly doubled its rental price for Nvidia Blackwell B200 chips during a contract renewal, according to data cited by The Verge AI.
Six of the most powerful names in global finance announced this month that they are teaming up with Nvidia on a $500 billion plan to make AI chips something ordinary investors and lenders can treat like real estate or bonds. The group includes Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR.
"This is really the first time that technology chips have become an investable asset class," Nvidia CEO Jensen Huang told CNBC. He called his chips "revenue-generating assets" that are "productive, long-lived, and fungible," meaning they can be swapped between users and purposes.
BlackRock CEO Larry Fink compared the moment to the birth of mortgage-backed securities, the bundled home loans that became a cornerstone of 1970s financial engineering and, later, the centrepiece of the 2008 financial crisis.
Why are chips suddenly being treated like property?
Think of it like a GPU-backed loan. A GPU, or graphics processing unit, is the specialised chip that does the heavy number-crunching AI needs. Firms buy huge quantities of them, and now banks want to lend money against them the way a mortgage lender lends against a house.
The value of that loan depends heavily on how long the chip stays useful. That is where Huang's current pitch runs into his own past statements.
Last year, Huang told attendees at an Nvidia conference that when the company's newest chips started shipping in volume, "you couldn't give Hoppers away," referring to the previous generation of Nvidia chip. "There are circumstances where Hopper is fine. Not many," he said.
Now, to support the financing pitch, Huang points to the A100 chip Nvidia launched in 2020 and says it "remains in active commercial use" with customers committing to multi-year contracts. He claims the economic life of that chip is approaching a decade.
Does the maths actually add up?
That claim matters enormously to lenders, because the longer a chip holds its value, the more money they can lend against it and the longer the loan can run.
The trouble is that nobody agrees. Short seller Michael Burry, known for predicting the 2008 mortgage collapse, argues chips depreciate, or lose value, over two to three years. IBM CEO Arvind Krishna says five years. Huang now says ten.
For reference, Broadcom signed a similar $35 billion deal with Apollo and Blackstone earlier, using roughly a million chips as loan collateral. That deal came with a guarantee from Broadcom itself backing the loans. No equivalent guarantee from Nvidia has been announced here.
Should ordinary people care about this?
If you rent cloud computing for a business, yes. The price to rent Nvidia chips has already been rising and one market analyst, Brendan Burke, says the cause is a shortage of chips used for inference, which is the process where an AI model actually answers your questions or processes your data after it has been trained. A cloud provider nearly doubled its per-chip rental rate at contract renewal.
More broadly, this is not a done deal. The announcement covers memorandums of understanding, which are formal letters of intent rather than signed contracts. A $100 billion Nvidia MOU with OpenAI announced in 2024 never produced a deal.
Watch for two things. First, whether any of these firms publish binding loan terms and what depreciation schedule they accept. Second, whether chip rental prices keep climbing or level off as new data centres come online and Chinese open-source AI models cut demand for raw compute.
