Shares of artificial intelligence (AI) data center operator CoreWeave (CRWV 16.30%) were declining on Tuesday following an earnings report that exceeded expectations but featured somewhat weak guidance. A delay related to a single data center was the culprit.
While the data center delay seems like minor news, a new warning about AI hyperscalers from famous investor Michael Burry could be problematic for CoreWeave. While Burry didn't mention CoreWeave directly, the investor accused tech giants of understating depreciation, and thus inflating earnings, by stretching out useful lifetimes of Nvidia GPUs. Burry estimates that the five largest AI hyperscalers will understate depreciation by a cumulative $176 billion between 2026 and 2028.
For CoreWeave investors, it's worth taking a look at the company's practices around depreciation.
Image source: Getty Images.
A pre-IPO increase in useful lifetimes
CoreWeave increased the useful lifetime of technology equipment, which includes all computing, networking, and storage components that go into its AI data centers, from five years to six years at the start of 2023. At the end of June, CoreWeave owned just over $13 billion of assets in this category. The difference between a useful lifetime of six years versus five years effectively lowers depreciation expense by hundreds of millions of dollars annually. Notably, this increase came a couple of years before CoreWeave went public.
CoreWeave's six-year depreciation schedule is on par with other tech giants, according to the data Burry presented. The big question: Is it reasonable? This category includes a lot of different stuff: CPUs, GPUs, solid-state drives, networking equipment, memory chips, and other computing components. Some components last a lot longer than others.
GPUs likely account for a large chunk of CoreWeave's total technology equipment assets. One estimate from analysts put GPUs at 39% of the total cost of building a 1-gigawatt data center.
Here's the problem: Data center GPUs used for AI workloads reportedly have short useful lifetimes. Back in late 2024, Tom's Hardware reported that data center GPUs running AI workloads at utilization rates between 60% and 70% may only have useful lifetimes of one to three years. This is similar to the reported lifetime of GPUs used to mine cryptocurrency, another computationally intensive task.
We don't know exactly how long GPUs are lasting in CoreWeave's data centers. In fact, CoreWeave has scaled up so quickly that many of its data centers haven't been operational for long enough to know for sure. However, if data center GPUs running AI workloads are averaging somewhere between one and three years, and GPUs make up roughly 40% of AI data center capital spending, CoreWeave's six-year depreciation schedule could be overly optimistic.
If that's the case, CoreWeave's earnings would benefit from understated depreciation. The company isn't profitable on a generally accepted accounting principles (GAAP) basis as it stands, and those losses would look even worse if depreciation expense were higher.
NASDAQ: CRWV
Key Data Points
Something to chew on
It's important to note that Burry has short positions in various AI-related companies, and that he would benefit from a decline in AI stocks. It's also important to note that Burry's accusation is just that -- an accusation. There may be a reasonable rationale behind the useful lifetimes CoreWeave and other hyperscalers are using.
However, a six-year useful lifetime for a category that's dominated by data center GPUs does seem at least somewhat optimistic. If CoreWeave eventually reduces the useful lifetime to reflect real-world replacement rates, it would have a large negative impact on the bottom line. Even the most bullish investors should take this possibility seriously.
