The Stock Market's New Most-Hated Word Is Pummeling the AI Trade

Fears that expensive GPUs and AI chips will depreciate much faster than hyperscalers assume have become a new drag on the AI trade, helping to push the Nasdaq 100 down ~6.3% and the Tech Select Sector SPDR over 9%; prominent skeptics including Michael Burry and Jim Chanos and analysts such as Kai Wu and Peter Berezin warn that 2–3 year useful lives (vs. the 5–6 years many firms use) could boost annual depreciation from roughly $150bn today to $400–500bn, with Burry estimating $176bn of understated depreciation in 2026–28 and Berezin noting $500bn a year would exceed combined 2025 profits. The implication is meaningful earnings compression for the Magnificent 7 and hyperscalers if accounting and economics catch up to capex reality, though the thesis is not yet consensus—mainstream strategists like Bernstein’s Stacy Rasgon continue to defend longer six-year GPU useful lives.

Analysis

Depreciation fears have emerged as a material headwind for the AI trade, with the Nasdaq 100 down 6.3% over the last few weeks and the Technology Select Sector SPDR (XLK) off more than 9%. Prominent skeptics including Michael Burry and Jim Chanos argue GPUs and AI chips will lose value faster than hyperscalers assume, adding to earlier circularity concerns and amplifying risk-off positioning.

Analysts quantify the potential hit: Burry estimates hyperscalers will understate depreciation by $176 billion in 2026–2028, Peter Berezin warns that $2.5 trillion in AI assets with a 20% depreciation rate implies $500 billion of annual charges that would exceed combined 2025 profits, and Kai Wu projects depreciation rising from ~$150 billion today to ~$400 billion over the next five years. Those scenarios would meaningfully compress net income for the Magnificent 7 and major hyperscalers if useful lives shorten to 2–3 years from the commonly assumed 5–6 years.

The thesis is not yet consensus—Bernstein’s Stacy Rasgon maintains GPUs can run profitably for about six years and current depreciation accounting is reasonable—so market moves may reflect fear of a regime change rather than confirmed accounting revisions. Investors should therefore focus on hyperscalers’ capex disclosures, stated useful-life assumptions, replacement-cycle commentary and any shifts in analyst models as triggers for re-rating high-capex tech names.

Key Decisions for Investors

• Reassess exposure to high-capex AI leaders (including NVDA and the Magnificent 7) and consider trimming positions or buying downside protection given the risk of accelerated depreciation compressing net income
• Monitor upcoming earnings and capital-expenditure disclosures for explicit useful-life assumptions and replacement-cycle guidance; if companies shorten lifecycles toward 2–3 years, downshift valuations and earnings estimates
• Implement tactical hedges on XLK and QQQ (for example, put spreads or inverse ETFs) to protect portfolios from further sector downside while maintaining selective long exposure to durable cash-flow names
• Stress-test portfolio models under scenarios where annual depreciation rises to the $400–500 billion range and set pre-defined trigger prices to rebalance if analyst consensus and company disclosures begin to shift