Nvidia wants Wall Street to look at AI compute in a new way: not simply as chips inside data centers, but as an investable asset class. The company is working with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR on $500 billion in financing tied to that idea.
The argument is simple on the surface. If chips can generate rental revenue for years, then financiers may be willing to fund them much like other productive assets. The harder question is whether that framing holds up when demand, depreciation, and AI infrastructure costs are all still moving targets.
The $500 billion compute pitch
Nvidia CEO Jensen Huang described the shift in sweeping terms. Speaking to CNBC, he said, “This is really the first time that technology chips have become an investable asset class.” He added: “These are revenue-generating assets now. They’re productive, they’re long-lived, they’re fungible, they’re flexible.”
BlackRock CEO Larry Fink also compared the moment to an earlier era of financial engineering. He told CNBC, “This is the very beginning, like what it was when I started in the mortgage-backed securities market in the 1970s, and I look upon this as a next future for financial engineering.”
That comparison is doing a lot of work. The source article points out that mortgage-backed securities eventually ran into trouble when mortgages were overproduced, according to former hedge fund manager Mark Rubinstein. In the compute case, the concern is that AI infrastructure may also face oversupply if data centers become too abundant or if demand does not keep expanding.
For now, Nvidia can point to market signals that support its case. The price to rent older chips has been rising, and Silicon Data projects that the trend will continue through 2028. One cloud service provider nearly doubled its prices on Nvidia Blackwell B200 chips for one rental customer during a contract renewal.
Why compute is more than a chip, but not a full data center
Huang has argued that “Nvidia compute is not just a chip.” The reason is CUDA, Nvidia’s software layer. In his framing, CUDA helps make Nvidia AI factories different from ordinary silicon because it can keep improving the output of installed hardware.
He described Nvidia’s system as “a complete AI factory platform including accelerated computing, networking, systems software, AI frameworks and a global developer ecosystem.” That list is important, but so is what it leaves out: physical data center facilities.
Chips and software do not create usable compute by themselves. They need to sit inside large data center infrastructure, including warehouses and power supplies. The source article’s core criticism is that the financing pitch appears to focus on the Nvidia-controlled part of the stack, while much of the previous infrastructure financing has been tied to real estate and facilities.
Rubinstein wrote that Blackstone has built a platform valued at $185 billion including facilities under construction, and that it sees the market for long-term ownership of stabilized data centers potentially growing to $1 trillion over time. For Nvidia, however, the more direct incentive is chip demand.
The depreciation problem
The compute-as-asset argument depends on useful life. A building can last a long time. A GPU is usually understood differently, with estimates ranging from somewhere between two and five years.
That is why Huang’s current language stands out. Last year, while promoting Nvidia’s newer Blackwell architecture, he said of Hopper chips: “When Blackwell starts shipping in volume, you couldn’t give Hoppers away.” He also said: “There are circumstances where Hopper is fine. Not many.”
Now Nvidia is making a case that some of its hardware can remain commercially productive for far longer. Huang has pointed to the pre-Hopper A100 chip, introduced in 2020, as a “powerful example.” He said it “remains in active commercial use” and that “Customers continue to commit capacity for multi-year deployments, extending A100’s economic life toward a decade.”
Those statements are not impossible to reconcile, but they create tension. Nvidia benefits when customers upgrade to the newest architecture. It also benefits if financiers believe older Nvidia systems retain enough value to support long-term lending.
Why the deal structure matters
The source article stresses that Nvidia’s $500 billion financing push is not a completed transaction. It is based on memorandums of understanding. That matters because a memorandum can create a headline without guaranteeing that the announced plan will fully materialize.
The article points to Nvidia’s earlier $100 billion memorandum of understanding to invest in OpenAI, which did not happen. That example does not prove the new compute plan will fail, but it does show why the distinction between an announcement and a closed deal is important.
A similar model has already appeared elsewhere. Earlier this summer, Broadcom put together a $35 billion package with Apollo and Blackstone to fund what is now being called compute, using about a million chips as collateral. Apollo and Blackstone would earn interest, while Broadcom provided a guarantee for two senior notes issued by the special purpose vehicle where the chips are held.
That arrangement was designed to increase demand for Broadcom chips. The source article argues that Nvidia appears to be pursuing a similar path for similar reasons.
The big risk is demand
The compute asset pitch works best if chip demand keeps rising and if the assets remain useful long enough to support financing. Several risks in the source article complicate that picture:
- The AI industry is becoming saturated with data centers.
- Chinese open-source models require less compute despite being fairly powerful.
- Frontier labs such as Anthropic and OpenAI are driving much of the current demand, but the article raises the question of whether they can make money.
- Older chips may be valuable today, but Nvidia’s own product cycle can make last-generation hardware look less attractive when new architecture ships.
None of this means the compute financing strategy cannot work. It does mean the label “asset class” is carrying a heavy promise. Investors are being asked to treat AI compute as durable, productive, and financeable at scale, even as the market is still testing how much compute buyers will need and how long each generation of hardware will stay economically useful.
That is the central tension in Nvidia’s new financial story. Compute may be generating revenue now. But turning it into a major asset class requires confidence not just in Nvidia’s chips, but in the durability of AI demand itself.