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—particularly hyperscalers—may be inflating profits by assigning overly long useful lives to AI chips, reducing annual depreciation and understating expenses; if chip useful lives are actually closer to 2–3 years rather than 5+ years, firms could face more frequent refresh cycles and higher recurring capital expenditures. Nvidia reported $57 billion in sales for the period ending Oct. 26, up 62% year-over-year, and trades at a trailing P/E of ~45 and a forward P/E of ~23 (S&P ~21); while strong demand could benefit chip vendors, overstated asset lives would lift reported earnings and valuations across AI-linked stocks and increase downside risk if demand softens.

Analysis

Market structure: Shortening assumed useful lives for AI chips raises near-term winners (chip OEMs like NVDA capturing repeated replacement demand) and losers (cloud hyperscalers and AI software vendors facing higher recurring CapEx and margin compression). If useful life collapses from 5+ years to ~2–3 years, annualized depreciation doubles to triples, which for hyperscalers could cut free cash flow margins by 200–500 bps on AI-capex-heavy workloads within 12–24 months. Secondary markets (used GPU channels) may expand but won’t offset OEM ASP power if supply remains tight.

Risk assessment: Tail risks include SEC/accounting restatements (earnings revisions) and a demand shock if AI projects fail to monetize — both could trigger >30% re-rates for overvalued AI plays within 1–3 quarters. Immediate volatility will spike around NVDA and hyperscaler earnings (next 30–90 days); medium-term (6–18 months) the key is actual replacement orders and inventory disclosures. Hidden dependencies: software layer longevity, virtualization, and resale ecosystems could extend physical useful life and blunt worst-case capex ramp scenarios.

Trade implications: Favor direct exposure to semiconductor pricing power but hedge execution risk — NVDA benefits from frequent refreshes, while AMZN/MSFT/GOOGL suffer margin pressure if they internalize replacement costs. Options are efficient: buy calendar/LEAP call spreads on NVDA (6–12 months) and finance with short-dated put spreads on selected hyperscalers; rotate 3–5% of portfolio from pure-AI names into broad market ETFs (SPY/VOO) if cross-asset volatility rises.

Key Decisions for Investors

• Establish a 2–4% net long position in NVDA via a 6–12 month call spread (buy 10–20% ITM LEAP or near-ATM call, sell 30–40% OTM call) to capture replacement-cycle upside while capping premium paid; reassess after next NVDA earnings/GPU guidance (30–90 days).
• Initiate a 1–2% short-via-put-spread position against cloud hyperscalers (AMZN or MSFT) using 3–6 month 10–20% OTM put spreads to hedge for 200–500 bps margin compression risk if useful lives prove shorter; size to loss-limit of ~1% portfolio per name.
• Execute a pair trade: long NVDA (1.5–2%) and short AMZN or GOOGL (1.5–2%) — use options to express asymmetric risk (NVDA call spread funded by short put spreads on hyperscaler) and target a 1.5–2x directional hedge ratio.
• Rotate 3–5% from high-multiple pure-AI equities into SPY/VOO or IG corporate bonds if market-implied volatility (VIX) spikes >25 or if NVDA/hyperscaler guidance misses; rebalance when VIX normalizes below 18 or margin guidance improves.
• Monitor three specific catalysts in the next 60–120 days before scaling positions: (1) NVDA sales/guidance for Blackwell/cloud GPUs, (2) AMZN/MSFT/GOOGL capex & depreciation policy disclosures, and (3) any SEC/accounting inquiries or restatement notices; adjust positions by ±50% if any catalyst triggers adverse revisions.