Michael Burry Thinks AI Companies Are Overestimating the Useful Life of Chips. Here's Why That Could Be a Big Problem

Michael Burry warns that tech companies may be overestimating the useful life of AI chips, which would understate depreciation expense and inflate reported earnings and valuations. The article notes Nvidia reported $57 billion in sales for the period ending Oct. 26, up 62% year-over-year, with a trailing P/E of 45 and a forward P/E of 23 (versus the S&P 500 average of 21), and cites strong demand for cloud GPUs. If chip lifespans are closer to 2–3 years rather than the 5+ years assumed by hyperscalers, investors should expect more frequent capex cycles, higher costs for hyperscalers and increased downside risk for richly valued AI-related stocks.

Analysis

Market structure: Shortened useful lives (2–3 years vs. 5+) shift winners to chipmakers (NVDA, AMD, INTC) and equipment suppliers (ASML, LRCX) that can monetize more frequent refresh cycles; hyperscalers (AMZN, GOOG, MSFT) and capex-heavy cloud units absorb immediate pain via higher depreciation and lower reported EPS. Supply/demand remains tight near-term (H100/Blackwell sellouts -> 6–12 month constrained cadence), preserving pricing power for leading GPUs but inviting rapid capacity expansion risk in 12–36 months.

Risk assessment: Tail risks include an AI demand cliff or accounting restatements of useful lives that could compress reported cloud margins by 10–25% and trigger a >30% correction in AI-adjacent equities (probability 10–20% over 12 months). Near-term (days–weeks) volatility will center on NVDA earnings and hyperscaler capex commentary; medium-term (3–12 months) hinge on demonstrated customer ROI from AI projects; long-term (>12 months) outcomes depend on refresh cadence normalizing and competitive supply additions.

Trade implications: Tactical long exposure to NVDA and equipment (ASML, LRCX) with hedges against hyperscaler capex risk; use optioned exposure (3–9 month call spreads on NVDA) rather than outright leverage. Consider relative-value pairings (long NVDA, short AMZN or an AI-exposed software ETF) sized so short leg caps drawdown to <50% of long’s notional.

Contrarian angles: Consensus underprices the upside for semiconductor equipment if refresh cycles shorten — this could double addressable capex for equipment makers over 2–3 years. Conversely, consensus may be complacent about margin pressure at hyperscalers if ROI falters; a synchronized capex pullback would disproportionately punish richly valued software/infra names while supporting cyclical chip-equipment shortsqueeze scenarios.

Key Decisions for Investors

• Establish a core long position: 1.0–1.5% portfolio in NVDA (ticker NVDA) via 6–9 month call spreads (buy 1x 0.5–1.0 delta, sell 1x 0.8–0.9 delta) to capture upside while capping premium; add if NVDA pulls back >10% from today or post-earnings volatility crushes IV by >25%.
• Buy 0.5–1.0% portfolio in semiconductor equipment: equal-weight ASML (ASML) and LRCX for 12–36 months to play sustained capex; scale in on any 15%+ sector pullback and target 25–40% upside over 12–24 months.
• Establish a hedged pair trade: long NVDA (notional X) funded by a 0.4–0.6X short in AMZN (ticker AMZN) or an AI-software ETF for 6–12 months; exit if AMZN reports cloud capex guidance cut >15% YoY or NVDA forward P/E exceeds 30.
• Protect downside with options: buy 3–6 month put spreads on a broad AI/semiconductor ETF sized to cover 30–40% of long exposure (e.g., 10–20% notional) if implied volatility <30%; unwind if ETF falls <20% or volatility normalizes above 40%.
• Monitor: track hyperscaler capex commentary in next 60 days (AWS/Amazon earnings, Google Cloud, Microsoft Azure guidance). If combined capex guidance is cut >10% YoY, reduce AI/infra cyclicals exposure by 30% within 5 trading days.