Imagine that major technology companies such as Google, Microsoft, Amazon, Meta, or Oracle invest billions in specialized artificial intelligence chips, but then present these investments in their accounts in a way that makes them look better than they really are. That is exactly what investor Michael Burry, known from the film The Big Short, is talking about. According to him, these companies are underestimating how quickly their hardware becomes obsolete, which could lead to major financial problems. In this article, we will look at details from articles on Tom's Hardware and CNBC that discuss why the depreciation of Nvidia graphics processing units (GPUs) could be the next major challenge for those building enormous AI data centers.
What GPU depreciation means
Graphics processing units, or GPUs, are at the heart of artificial intelligence systems. Companies such as Nvidia manufacture and sell them at high prices, but these chips evolve rapidly. New, more powerful models arrive every year, causing older versions to lose value. Depreciation is an accounting method that companies use to gradually reduce the value of these chips on their books – as if they were saying: "This hardware has served us, but it is now less valuable." According to articles on Tom's Hardware and CNBC, large companies known as hyperscalers assume that their GPUs will remain in operation for 5 to 6 years. This means that they spread the purchase costs over a long period, which helps them keep their profits higher on paper.
Michael Burry, however, claims that this is a mistake. According to him, depreciation should be faster, over just 2 to 3 years, because Nvidia releases new chips so frequently that older models quickly lose performance and value. For example, A100 chips from 2020 are now less sought after for the most demanding AI tasks, even though they are still used elsewhere. Burry estimates that because of this approach, companies will understate depreciation by $176 billion (approximately CZK 4.048 trillion) between 2026 and 2028. This could mean that their profits are inflated and investors are seeing a distorted picture.
Burry's attack on the major players
Investor Michael Burry, who became famous for predicting the 2008 mortgage crisis, is now targeting the technology sector. The CNBC article describes how Burry accuses companies such as Google, Microsoft, Amazon, Meta, and Oracle of using excessively long depreciation periods. According to him, this causes companies to report higher profits under GAAP accounting standards than they should. Specifically, he estimates that in 2028, Oracle's profits could be overstated by 27%, while Meta's could be overstated by 21%. Burry argues that rapid advances in Nvidia's technology mean that GPUs lose value much sooner than companies admit.
The Tom's Hardware article discusses how these companies are investing enormous sums in building data centers packed with GPUs. After spending billions, faster depreciation would mean higher accounting expenses, which would reduce reported profits. Burry warns that this could lead to future accounting restatements, shaking investor confidence. He also emphasizes that although older chips such as the A100 are still running, their rental prices are falling, reducing the revenue they generate.
What does CoreWeave say?
Not everyone agrees with Burry. CoreWeave, a major provider of GPU cloud services, offers a different perspective. According to the CNBC article, CoreWeave reports that demand for older A100 chips from 2020 is still at full capacity, or 100%. For newer H100 chips, it is 95%. This suggests that companies do not immediately throw older hardware away – they use it for simpler AI tasks that do not require the highest performance.
CoreWeave argues that even though new Nvidia chips are being released rapidly, older models retain value for specific tasks, such as batch data processing, where speed is not as critical. This could justify longer depreciation periods because the hardware generates revenue for longer than Burry assumes. The Tom's Hardware article notes that hyperscalers move older GPUs to less demanding workloads, extending their economic lifespan.
Financial risks for investors
If companies switched to faster depreciation, as Burry suggests, their expenses would rise and their profits would fall. Estimates indicate that this could reduce Oracle's profits by 27% in 2028 and Meta's by 21%. The articles on Tom's Hardware and CNBC describe how this relates to enormous investments in AI infrastructure – companies are spending billions to build centers packed with Nvidia GPUs, but if the hardware ages faster than planned, cash flow problems could arise.
Burry points out that GPU rental prices are falling, reducing returns on investment. For example, older chips such as the A100 generate less revenue than the newer H100, even though they are still in operation. This could mean that companies will soon have to invest in new processor upgrades, increasing pressure on their finances. For investors, this means a risk that the shares of these giants could fall if it turns out that their accounting is not as solid as it appears.
