Last month, the financial landscape was shaken by concerns over circularity in Big Tech’s deal-making practices, which left investors feeling uneasy about potential market risks. Now, a new anxiety has emerged: the looming threat of depreciation. Investors are increasingly worried that the high-priced GPUs and semiconductor chips being hoarded by companies may depreciate faster than anticipated, transforming these assets into heavier cost burdens that could negatively impact earnings.
The tech-heavy Nasdaq 100 has seen a decline of 6.3% in recent weeks, while the Technology Select Sector SPDR Fund has plummeted more than 9%. Notably, prominent short-sellers such as Michael Burry and Jim Chanos have highlighted the issue of depreciation as a central reason for their skepticism toward the AI trade. Burry took to social media platform X to share his perspective, estimating that the hyperscalers would understate their depreciation by a staggering $176 billion between 2026 and 2028. He argues that the expected lifecycle of chips is closer to two to three years, far shorter than the six years that many companies are assuming.
This sentiment isn’t limited to well-known short-sellers. Peter Berezin, chief global strategist at BCA Research, also voiced concerns in a LinkedIn post, dissecting the potential financial ramifications of AI asset depreciation. He referred to a chart illustrating that if the hyperscalers hold at least $2.5 trillion in AI assets by the end of this decade, a depreciation rate of 20% could lead to an astronomical $500 billion in annual depreciation expenses—surpassing the total profits projected for these firms in 2025.
In a similar vein, Kai Wu, founder and Chief Investment Officer of Sparkline Capital, indicated that annual depreciation values could soar from $150 billion to $400 billion over the next five years. He further noted that while the so-called “Magnificent 7” tech companies remain highly profitable, their net income may face downward pressure as depreciation costs from increased capital expenditures ramp up. Wu opined that many analysts are overly optimistic regarding the useful life of AI data centers, suggesting that a timeframe of 2-3 years aligns better with Nvidia’s rapid GPU replacement cycle.
While some analysts, such as Bernstein’s Stacy Rasgon, argue that the current depreciation accounting practices of major hyperscalers are reasonable and that GPUs can operate profitably for around six years, the prevailing anxiety around accelerated depreciation is clouding the outlook for the previously booming AI sector. This growing concern appears to be imposing significant challenges on the once vibrant AI trade.
