Are tech giants really "playing financial tricks" with the depreciation of AI chips?

As technology giants invest hundreds of billions of dollars in artificial intelligence, a seemingly mundane accounting practice—the depreciation method for key equipment such as AI chips—has unexpectedly become the focus of market controversy, sparking intense debate about whether companies are using it to embellish profits.

According to a report by The Wall Street Journal on December 8, companies including Meta, Alphabet, Microsoft, and Amazon have recently extended the estimated useful life of their server and network assets. This accounting change directly reduces current depreciation expenses, thereby inflating reported profits, which has immediately raised concerns and scrutiny among some investors.

Notable "big short" investor Michael Burry has labeled this move as "one of the modern common frauds." In an article last month, he wrote that extending the useful life of assets reduces depreciation expenses and increases apparent profits, leading to overvaluation of assets and inflated profits. When massive capital expenditures are at stake, such adjustments in accounting estimates are enough to trigger sensitive nerves in the market.

However, for investors focused on these tech giants, the crux of the issue may not be a simple matter of "right or wrong." While extending the depreciation period can indeed boost profit figures immediately, its substantive impact on the company's fundamentals may be limited. The market is more concerned with the ultimate returns on AI investments rather than the details of accounting treatment. A deeper discussion has shifted towards the choice of depreciation methods and whether these accounting treatments will ultimately affect the assessment of the long-term value of AI investments.

The "Profit Magic" of Accounting Adjustments

Depreciation is an accounting practice that ensures the cost of capital investments is recognized over time in financial statements. When a company extends the "estimated useful life" of an asset, it means spreading the total cost over a longer period, thereby reducing the expense for each year and directly boosting current profits.

Technology giants have indeed been doing this in recent years. For example, Meta has increased the estimated useful life of most of its server and network assets to 5.5 years by 2025, up from 4 to 5 years, and as low as 3 years in 2020. This change has reduced Meta's depreciation expenses by $2.3 billion in the first nine months of 2025. However, it is worth noting that its total depreciation during the same period was nearly $13 billion, with pre-tax profits exceeding $60 billion.

A similar trend is observed among other giants. Alphabet and Microsoft currently use a 6-year depreciation period for similar assets, significantly higher than the 3 years in 2020. Amazon used 4 years in 2020, increased it to 6 years by 2024, but then reduced the depreciation period for some servers and network equipment back to 5 years in 2025. This practice of influencing billions of dollars in profits through adjustments in accounting estimates naturally invites market scrutiny.

Straight-Line vs. Accelerated Depreciation: Which is More Accurate?

The core of this controversy may not lie in the specific numbers of depreciation periods, but in the choice of depreciation methods. Currently, the vast majority of companies use the "straight-line depreciation method," which keeps the annual depreciation expense constant over the asset's entire useful life However, for rapidly evolving hardware such as AI chips, this method may not accurately reflect the true depreciation trajectory of their value. According to reported data, the organization Silicon Data, which tracks NVIDIA chip prices, found that the average resale value of a three-year-old H100 system is about 45% of the new H100 price. This indicates that the value of such assets declines more rapidly in the early stages of use and then stabilizes.

In this case, the "accelerated depreciation method" may better reflect economic reality. By adopting this method, the depreciation expense for the asset is higher in the early stages and lower later on, which aligns more closely with the curve of rapid value decline. However, analysis also points out that even with the accelerated depreciation method, the differences from the straight-line method may not be drastically different, insufficient to overturn the overall financial situation.

What should the market pay attention to?

Fundamentally, many numbers in financial statements are based on estimates, guesses, and assumptions; depreciation expense itself is an accounting construct, and precise calculation is nearly impossible. For high-tech equipment with strong demand, management also finds it difficult to accurately predict its exact effective lifespan.

For the market, a more concerning signal may be asset impairment. According to accounting standards, if the value of an asset is severely impaired, management should make a significant write-down. However, this situation usually occurs after the company's own stock price has already collapsed, which is far from the current market position of the "Tech Seven."

Therefore, although the accounting debate surrounding depreciation is intense, it may not become a key factor influencing investment decisions. As the analysis reported suggests, if investors conclude one day that a large amount of AI investment is being wasted, the reason will not be the choice of depreciation period by the company. For these tech giants' AI ventures, the ultimate criterion for the market will be whether these investments can yield substantial returns in the future, rather than the accounting estimation methods used along the way.