Nvidia announced this week that Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR are willing to spend up to $500 billion to build AI data centers. While this eye-popping number garnered a lot of attention, the bigger story is Nvidia’s efforts to create a secondary market for its aging GPUs.
To convince these financial giants, Nvidia agreed to guarantee with its own funds that the chips used as collateral in these transactions would retain their value.
Many people have commented on how unusual, clever, and dangerous this plan is. That’s all. The bond market was so shaken that NVIDIA CEO Jensen Huang went on X and Business TV to explain in more detail how NVIDIA’s risk will be limited.
But behind the financial maneuvering to fund AI data centers (and keep Nvidia profitable) is something perhaps far more interesting to startups and companies. Huang wants to ensure that the used AI hardware ecosystem thrives and maintains demand for aging Nvidia hardware.
Specifically, NVIDIA has promised that if the value of the GPUs used as collateral does not hold up as expected, the company will cover up to 25% of the difference. So even if the data center owner defaults on the loan and the lender has to liquidate, Nvidia will still provide the funding if the chip fails to achieve its on-book price.
The danger for Nvidia is that this creates what financiers call “wrong way” risk. That means NVIDIA’s obligations will increase as demand weakens. If that happens, there is a high possibility that profits will come under pressure.
Still, the plan is intentionally different from the comparisons some are making to Lucent Technologies. Lucent was a communications equipment provider that rose and collapsed with the dot-com bubble after lending customers money to buy products.
Huang knows that comparisons to Lucent will cast a shadow on Nvidia. And it’s not unfair. Nvidia is undoubtedly pumping billions into people buying its chips, including frontier AI labs OpenAI and Anthropic, neo-clouds like CoreWeave (the originator of which uses Nvidia chips as collateral), as well as Nebius, Firmus, and Lambda. And the company is working on another $750 billion worth of circular trades this summer, according to Bloomberg estimates.
“Is this a circular loan?” Huang wrote to X about the new plan. “This initiative is designed to address that concern. We are bringing independent, long-term institutional capital to the AI infrastructure market.”
it’s true. Unlike Lucent, Nvidia only agrees to protect some of the value of its chips in the future, leaving most of the capital and risk to other companies.
If this plan works, Nvidia will find new sources of funding to build AI data centers, as many traditional methods are starting to decline. For example, some hyperscalers are already taking on large amounts of debt (e.g., Oracle), issuing new tranches of equity (Google), and burning through large amounts of cash (Meta).
The situation has become so dangerous that Microsoft CEO Satya Nadella recommended a book called “1873” during a recent earnings call. It’s about the financial engineering of the railroad era that destroyed the country’s economy.
The risk is that today’s AI boom, with demand far outstripping production capacity, won’t last much longer. What happens if businesses and consumers curb their use of AI instead of being in its infancy, or will new technologies emerge that make existing infrastructure more efficient or make all of today’s AI infrastructure obsolete?
And then, like many bug whips in the face of a car (to paraphrase Danny DeVito’s Lawrence Garfield), demand dries up and everything collapses.
But Huang argues that won’t happen, pitching his vision of AI as long-term “investable infrastructure.” As such, AI servers, which he calls “AI factories,” are more like railroads and airlines than rapidly depreciating assets like PCs.
“If needs change, the factory can be used by another customer, another cloud, or another operator. This broad ecosystem gives NVIDIA Computing a deep market of potential users and off-takers and helps protect residual value,” he promised.
In its future, Nvidia is as concerned with aging architecture as it is with new chips. And perhaps, just as startups, businesses, and even researchers are starting to opt for affordable open-weight models alongside cutting-edge options, we’ll likely start to utilize a greater variety of hardware tailored to different AI needs.
As the king of AI, Nvidia has the power and opportunity to make it happen.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
