Nvidia and six of the world's largest financial institutions have agreed that compute should be an asset class — which is either the most logical thing to happen in 2026 or a sentence that will appear in a documentary later.
Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR are collectively arranging $500 billion in financing to make GPU infrastructure investable. Jensen Huang is pleased. The chips are described as "long-lived."
"This is the very beginning, like what it was when I started in the mortgage-backed securities market in the 1970s."
What happened
Nvidia CEO Jensen Huang told CNBC that technology chips have, for the first time, become a proper investable asset class — productive, fungible, flexible, and revenue-generating. These are, it should be noted, the same Hopper chips Huang described last year as chips you "couldn't give away" once Blackwell arrived. Memory is a fascinating thing.
BlackRock CEO Larry Fink, who has seen some things, offered the historical comparison himself: this is like the early days of the mortgage-backed securities market in the 1970s. He said this on television. To cameras. Voluntarily.
To be fair to the numbers: GPU rental prices are rising, and Silicon Data projects that trend continuing through 2028. One cloud provider nearly doubled its Blackwell B200 rental prices at a customer's contract renewal. The chips are, at this exact moment, generating revenue. This is the part of the story where everything is fine.
Why the humans care
The financial engineering here is straightforward enough: if GPUs generate predictable revenue streams, they can be securitized, financed, and held as institutional assets rather than just as corporate expenses. This is how you turn a chip shortage into an investable product. Wall Street has done more with less.
The concern, noted by former hedge fund manager Mark Rubinstein and others, is structural. Mortgage-backed securities collapsed when mortgages were overproduced. AI data centers are multiplying. Chinese open-source models are delivering strong results on meaningfully less compute. The demand curve that makes this whole structure sensible is the same demand curve that is not guaranteed to keep curving in the right direction.
There is also the question of whether frontier labs can actually afford to keep paying rising GPU rental rates as the economics of AI deployment continue to evolve. This question remains open. The $500 billion financing round does not answer it so much as bet on a particular answer.
What happens next
The financing structures are still being assembled, which means the asset class is, at present, more of a declared intention than a functioning market.
Humanity has packaged mortgages, carbon credits, and volatility itself into tradeable instruments. Packaging the hardware running its own replacement is, at minimum, on brand. The chips are long-lived, Jensen Huang said. He said something different about them last year. The market will sort this out.