The $500 Billion Question Causing Waves Across Every AI Stock: How Much Are Nvidia GPUs Actually Worth?

Back in August, Nvidia (NASDAQ: NVDA) announced a partnership with six Wall Street asset managers to provide financing for AI infrastructure build-outs. The effort aims to provide up to $500 billion to companies that build data centers and install Nvidia's GPUs in them.

Here's the key to those deals, though: The GPUs themselves act as collateral on the loans. That means the value of the GPUs is one of the main considerations for lending decisions.

An extra $500 billion in easy-to-access capital would be a boon for companies that are heavily reliant on debt to build new data centers. Neocloud companies like CoreWeave (NASDAQ: CRWV) and Nebius (NASDAQ: NBIS) are highly leveraged, and making huge bets that the AI compute shortage will continue for some time. Nvidia, meanwhile, benefits from its customers having more capital in their hands to put toward the data center build-out.

But Nvidia is facing some major pushback from its financing partners on Wall Street, and it all centers around how much of their value its GPUs retain over time. The answer to that question could reverberate throughout the entire industry.

Image source: Nvidia.

A big mismatch between Nvidia and the banks

At the center of the controversy is just how long an Nvidia GPU's useful life is. Nvidia says it could be a decade; Wall Street bankers say it's more like three to four years, according to a Reuters report.

Nvidia published a blog post in response to the Reuters article. At the center of its argument was the question of how hyperscalers have changed their depreciation schedules for server and networking equipment over the last few years. It showed that Alphabet, Microsoft, Amazon, and Meta Platforms have extended the depreciation life of their hardware from three or four years to five or six years.

Also working in Nvidia's favor are recent pricing changes from CoreWeave and Nebius. CoreWeave announced that it "continued to contract new compute capacity at higher prices," signing new customer contracts at an annualized rate of $40 million per megawatt in September. Nebius, meanwhile, just raised prices across the board for hourly compute rentals by 17% to 21% for various Nvidia GPUs, including four-year-old chips.

Nonetheless, at the end of the day, the banks are the ones putting up the capital, so they hold all the cards. The challenge is that it's truly impossible to know yet whether the way AI data centers function has created a structural change in the useful life of an Nvidia GPU. It's just too early to tell. Whether the banks are right or not will have a huge impact on some of the most popular AI stocks in the market.

How it could impact the major AI stocks

If the banks want better guarantees, Nvidia or its customers will have to provide them. The Reuters article indicates there are already tens of billions of dollars in loans in the pipeline with stronger guarantees than were originally outlined in Nvidia's August announcement. Those guarantees could come from Nvidia itself, which would require it to maintain capital reserves, or from the revenue streams of investment-grade technology companies such as Microsoft, Alphabet, Amazon, or Meta Platforms.

Nvidia is reportedly in talks with insurance companies to hand off some of the financing risk tied to these loans in case a borrower defaults and the value of the GPUs fails to cover the debt. That could add to the cost of Nvidia's financing plans, however. The coverage is specifically aimed at neocloud companies like CoreWeave and Nebius.

It's important to note that if a neocloud defaults, it will likely be in part because the Nvidia GPUs powering its data centers aren't worth as much as originally anticipated. The counterargument to the fact that neoclouds are raising prices is that we're currently in the midst of a severe shortage of AI compute. That has pushed prices higher for what's essentially a commodity product. When the market reaches equilibrium, prices for compute will begin to decline.

In other words, today's higher prices aren't a reflection of the increased durability of Nvidia GPUs, but rather a reflection of AI developers' desperation to get their hands on any available compute.

The hyperscalers don't carry nearly as much risk, but if AI chips' useful lives prove shorter than expected, it will negatively impact their return on invested capital. The good news is that Amazon reports it will break even on its AI accelerator chips within three years, and Alphabet says its breakeven time is just two years.

These days, both hyperscalers are installing more of their own custom AI accelerators than Nvidia GPUs, and those application-specific chips have even shorter payback periods. While GPUs can still generate positive returns on investment, they might not be the massive returns currently projected.

Adam Levy has positions in Alphabet, Amazon, Meta Platforms, and Microsoft. The Motley Fool has positions in and recommends Alphabet, Amazon, Meta Platforms, Microsoft, and Nvidia. The Motley Fool has a disclosure policy.