It’s likely that most investors don’t give depreciation schedules much thought. However, investors in Nvidia, technology funds, or even S&P 500 index funds may have a stake in the longevity of the costly chips that underpin artificial intelligence.
Two investors who gained notoriety for betting against the U.S. housing market are now on opposing sides of a recent Wall Street controversy because of that question.
According to Michael Burry, big tech firms might be delaying the purchase of AI hardware for too long. If those chips lose economic value more quickly than businesses anticipate, reported profits may appear higher than the underlying economics support.
Although Steve Eisman believes the concern is legitimate, he thinks it might be focused on the wrong issue.
On the “New Money” podcast, Eisman described Burry’s argument as “too academic,” according to Business Insider. He cited ongoing demand for older Nvidia (NVDA) chips as proof that AI hardware might remain economically beneficial for longer than detractors anticipate.
But Eisman had a warning. He views the industry’s dependence on rapidly expanding clients like OpenAI and Anthropic as a greater risk than chip depreciation.
The main point of contention among investors is whether the massive sums of money being spent on AI will generate enough revenue to cover the costs.
Michael Burry sees a hidden cost in the AI boom
Although Burry’s argument appears technical, the fundamental idea is simple.
Let’s say a company purchases a costly machine and plans to use it for six years. The business spreads the cost over the machine’s anticipated useful life rather than immediately recording the full purchase as an expense.
If, after just three or four years, it turns out that the equipment needs to be replaced, its yearly economic cost will exceed the initial estimate.
Burry believes investors should ask whether AI infrastructure represents a similar situation.
Major tech companies are spending tens of billions of dollars on data centers that are stocked with graphics processing units, or GPUs, as faster and more advanced chip generations emerge.
According to Burry, prolonging the useful lives of computer equipment can lower annual depreciation costs, supporting higher reported earnings.
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
As Burry emphasized in a post on his Cassandra Unchained Substack, he’s not claiming that older GPUs abruptly stop functioning. His worry is whether its earning potential declines quickly enough for businesses to recoup their initial investment in a shorter amount of time.
That distinction is important for investors. An older chip can still earn revenue without producing the same economic return it did when it was new.
Burry has estimated that hyperscalers could understate depreciation by more than $175 billion between 2026 and 2028, according to Business Insider.
“Depreciation is not a bet on when a chip stops working,” he said.
Thus, his argument goes beyond a technicality in accounting. Investors may be overestimating the return on one of the biggest corporate spending booms in recent memory if businesses believe their AI equipment will remain economically valuable for longer than it actually does.
Steve Eisman points to old Nvidia chips still making money
Eisman observes real-world evidence that complicates Burry’s case.
In an investor presentation in September, CoreWeave revealed that a customer had extended Nvidia A100 infrastructure through 2029. In 2020, Nvidia unveiled the A100 architecture.
CoreWeave cited switching costs and workloads already built around the older architecture as reasons why the three-year renewal was priced similarly to standard one-year terms.
That example is significant because it demonstrates that when Nvidia releases a faster GPU, an older one does not necessarily become commercially irrelevant.
Different clients require varying degrees of processing power. A company that trains the newest frontier model may want the newest processors. If older hardware is available at a reasonable price, another company that runs scientific workloads, fine-tunes smaller models, or performs inference might be quite content.
CoreWeave’s presentation also demonstrated a three-year renewal of Nvidia H200 infrastructure at a premium to the initial contract. Customers who are currently operating significant workloads on well-established architectures may incur significant switching costs, according to the company.
That supports Eisman’s argument that technological progress does not automatically destroy the earning power of previous-generation chips.
It does not entirely refute Burry’s thesis, however.
Investors know the chip has economic value from the fact that an A100 can still be rented in 2029. What Burry wants to know is whether the observed demand offers a sufficient return on investment.
Eisman sees a different danger for AI investors
Eisman acknowledges there is risk involved in the AI trade.
He is further down the supply chain.
Nvidia sells the processors. Cloud providers and businesses such as CoreWeave (CRWV) build infrastructure around those chips. Alphabet (GOOGL), Microsoft (MSFT), Amazon (AMZN), and Meta Platforms (META) are all making significant investments in their AI capabilities. Corporate clients and AI developers pay to use that processing power on the other end.
Even if accountants disagree on whether equipment should be depreciated over four, five, or six years, the industry may have enough revenue to support significant investment if demand continues to grow quickly.
Eisman is concerned about what will happen if that demand doesn’t come through.
He has described OpenAI and Anthropic as the “Achilles heel” of the AI trade, according to Business Insider. If companies at the center of AI demand stumble, the financial consequences could move backward through data centers, cloud providers, and chip companies.
Investors who do not consider themselves AI traders should also be concerned about this issue.
These days, technology and AI stocks make up a disproportionately large portion of major U.S. indexes. This week, Reuters reported that companies related to technology and artificial intelligence make up more than half of the market capitalization of the S&P 500.
As broad-market portfolios become more dependent on the performance of these companies, an investor who owns a regular index fund is significantly exposed to the success of the AI buildout.
Burry and Eisman just can’t agree on where the weakest link could be.
Burry is watching how quickly the hardware loses value. Eisman is monitoring whether the number of consumers purchasing all that processing power continues to increase quickly enough to sustain the surrounding infrastructure.
What the Burry-Eisman debate means for investors
Investors are not required to choose between Eisman and Burry.
Rather, their debate offers a helpful checklist for tracking the AI growth.
Older hardware is the first thing to consider. Eisman’s claim that AI infrastructure has a longer economic life than the most pessimistic predictions imply would be supported if earlier generations of Nvidia GPUs continued to draw buyers at high prices years after their launch.
One significant piece of information is CoreWeave’s A100 contract until 2029. However, cost is just as important as use. Burry’s fears about depreciation might be justified, since an outdated GPU can continue to be active but generate much less revenue.
Demand is the second signal.
In the end, AI firms and corporate clients must produce enough revenue to sustain the data centers, processors, power infrastructure, and funding being developed around them. Arguments about many years’ worth of depreciation schedules may become less important if that demand continues to rise.
These presumptions can suddenly become considerably more important if demand declines.
Because of this, the dispute between the two “Big Short” investors is significant outside of Wall Street figures.
Both gained notoriety for challenging presumptions that, before the housing crisis, a large portion of the market took for granted. This time, inside the same technological boom, they are challenging several presumptions.
According to Burry, investors could be underestimating the actual cost of the devices that drive AI.
The real concern, according to Eisman, is whether enough people will continue to use them.
Investors who own Nvidia, hyperscalers, or just a broad-market fund face the same question: Will the AI boom ultimately generate enough revenue to support the amount of money businesses are spending on it?
Related: Michael Burry pulls an old playbook into the Nvidia fight
