Nvidia backs a $500 billion AI data center plan
Nvidia is lining up major investors for AI data centers while also supporting the resale value of older GPUs used as collateral in those deals.
Nvidia has unveiled a plan that could channel up to $500 billion into AI data center construction, with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR said to be willing to commit capital. The huge figure grabbed attention, but the more unusual part of the arrangement is Nvidia’s attempt to build a secondary market for aging GPUs. That could help extend demand for its hardware — and at the same time leave the company more exposed if the market weakens.
This is not just a push to raise money for new infrastructure. Nvidia is also trying to ensure that older AI chips keep enough value to serve as collateral in financing deals, according to IT-PUB News. That matters because the AI boom depends not only on sales of the newest chips, but also on what happens to the hardware as it ages.
Nvidia will absorb part of any resale shortfall
To make the financing more attractive to large investors, Nvidia has agreed to support the value of its chips with its own money. If GPUs used as collateral fail to hold their expected price, the company will cover up to 25% of the difference.
In practice, that gives lenders some protection if a data center owner defaults and the chips have to be sold for less than expected. Nvidia is not guaranteeing the full value of the hardware, but it is taking on enough downside risk to make the structure more credible for financial firms.
That is what makes the plan stand out. Nvidia is not simply selling chips and moving on. It is also helping support the later market for those chips, which is not a typical hardware business model.
Wall Street sees both opportunity and risk
The size of the deal and the way it is structured have led many observers to describe it as smart, unusual and dangerous at the same time. Bond markets were unsettled enough that Nvidia CEO Jensen Huang went on X and business television to argue that the company’s risk would be limited.
The concern is straightforward. Nvidia’s obligations would rise if demand for its chips weakens. In that scenario, the same slowdown that hurts resale values could also pressure the company’s revenues. Financiers call that “wrong way” risk — when the danger grows as the underlying market deteriorates.
Huang has framed the initiative as a way to bring “independent, long-term institutional capital” into AI infrastructure. In his view, the goal is to answer concerns about circular financing, not make them worse.
Nvidia wants older GPUs to keep their value
Behind the financing structure is a broader goal that could matter for startups and enterprises using AI tools. Nvidia wants an ecosystem in which used AI hardware still has value even after newer chips arrive.
That would support a market for older GPUs and make it easier for companies to reuse or resell hardware instead of treating it as nearly worthless once it is no longer state of the art. For Nvidia, that could help sustain demand across more of the product lifecycle, not just at the initial sale.
The company has already committed billions toward buyers of its chips, including frontier AI labs OpenAI and Anthropic, as well as neoclouds such as CoreWeave, which originated the idea of using Nvidia chips as collateral. Other companies mentioned in the source include Nebius, Firmus and Lambda. Bloomberg has also calculated that Nvidia has been working on another $750 billion worth of circular deals this summer.
The Lucent comparison still shadows the plan
The structure has revived comparisons with Lucent Technologies, the telecom equipment company that rose and then collapsed after financing customers so they could buy its products. Huang is aware of that comparison, and Nvidia appears to be trying to avoid repeating that history.
The difference, as Nvidia describes it, is that the company is not lending directly in the same way. Instead, it is bringing in large outside investors and agreeing only to protect part of the hardware’s future value. That leaves most of the capital and risk with financial institutions rather than putting the full burden on Nvidia.
Even so, the comparison has not gone away. If the AI market slows, the financial engineering supporting it could turn from a strength into a weakness. That is why the plan is being seen as both clever and risky.
AI data centers are becoming harder to finance
The timing also says something about the wider AI market. Traditional ways of funding data center expansion are becoming harder to rely on. Some hyperscalers have already taken on substantial debt, while others have issued new equity or used large amounts of cash to keep building.
That helps explain why Nvidia is trying to open another financing route. If AI servers can be treated more like long-term infrastructure and less like fast-depreciating computers, it may become easier to attract institutions looking for predictable, asset-backed returns.
Huang has argued that AI factories should be viewed more like railroads or airlines than like PCs that quickly lose value. He says that when needs change, the same infrastructure can be used by another customer, cloud provider or operator, which should help protect resale value.
That idea sits at the center of the whole plan. It depends on the assumption that AI demand will remain broad enough to support a market for used hardware. If that holds, startups, enterprises and researchers could end up with more hardware options over time. If not, the value of older equipment could drop fast.
For now, Nvidia is betting that AI infrastructure will remain an “investable” asset class and that its chips will keep finding buyers even as they age. The company clearly has the market power to try to shape that outcome. The harder part is whether the financing structure can hold if the AI boom cools.
