Nvidia name-checks Michael Burry in secret memo pushing back on AI bubble allegations

Investor Michael Burry has publicly accused Nvidia of exhibiting Cisco-era overbuild dynamics and questionable capital-allocation practices amid an AI infrastructure boom, while Nvidia circulated a seven-page memo rebutting his claims. Nvidia says it repurchased $91B of shares since 2018 (not $112.5B, which Burry reportedly included RSU taxes in), defends its employee equity grants, and states customers depreciate GPUs over four-to-six years with A100s retaining value beyond two-to-three years. Burry counters that hyperscalers plan nearly $3 trillion of AI infrastructure spending and warns of extended depreciation schedules, massive capex and potential overcapacity — a governance and valuation debate likely to influence investor positioning around Nvidia and AI suppliers.

Analysis

Market structure: The tussle amplifies tail-risk perception for AI-capex-dependent hardware (NVDA, AMD, INTC suppliers) and strengthens flight-to-software/cloud winners (MSFT, AMZN, GOOGL) if hyperscaler spend softens. If Burry’s Cisco-like scenario materializes, expect meaningful share re-pricing in hardware: a 20–40% re-rating is plausible over 6–18 months versus 5–15% for cloud/software which capture recurring economics. Cross-asset: rising equity risk aversion should push US 10y yields down ~10–30bp near-term, lift gold and defensive FX (JPY, CHF), and pressure copper/power futures if datacenter build projections are cut.

Risk assessment: Tail risks include a sudden hyperscaler capex pause, a regulatory probe into NVDA accounting/RSU treatment, or a secondary market GPU glut driving ASP declines of 30%+ — any of which could halve forward EPS multiples. Immediate (days) risk: vol spikes and retail-driven repricing; short-term (weeks/months): guidance cuts and order delays; long-term (quarters/years): structural demand miss vs. current $2.5–3T capex expectations. Hidden deps: resale market for used A100/T4 units, hyperscaler internal capex smoothing, and covenant-driven vendor financing.

Trade implications: Defensive tilt to software/cloud productors and volatility-insured exposure to NVDA. Near-term (0–3 months) buy put spreads or straddles around NVDA to monetize elevated event risk; medium-term (3–12 months) favor secular cloud exposure (MSFT, AMZN) via longs and reduce direct hardware beta. Use pair trades to neutralize market beta and capture dispersion: long MSFT vs short NVDA if you prefer durable margins over hardware cyclicality.

Contrarian angles: Consensus equates NVDA solely with overbuild; it understates durable TAM capture via software hooks (SDKs, models, licensing) and aftermarket pricing power for premium L40/H100 classes. Reaction is likely uneven — a 10–25% pullback in NVDA could be overdone relative to fundamentals if utilization stays >60% and resale values hold. Historical parallel: Cisco 2000 re-rated 60%+ as telco capex collapsed; differences now are larger recurring software revenue and concentrated enterprise AI spend, which could limit downside.

Key Decisions for Investors

• Establish a 1.0% notional position buying a 3-month NVDA put spread (sell 5% OTM, buy 12% OTM) to hedge downside risk ahead of the next earnings/capex updates; size to limit cost to ~0.15–0.25% of portfolio.
• Implement a 1.5% long MSFT position funded by a 1.0% short NVDA exposure (dollar-neutral) over a 3–12 month horizon to express preference for durable cloud economics vs. hardware capex cyclicality; rebalance if NVDA moves >15%.
• If NVDA falls >15% on capex concerns, scale into a tactical 1.0–2.0% long NVDA position using a staggered buy strategy (50% at -15%, 50% at -30%) and hedge with 6–12 month OTM puts to protect against deeper downside.
• Reduce direct exposure to small-cap AI hardware suppliers (reallocate 2–4% of portfolio away from pure-play ASIC/board vendors) into software/MLops names (MSFT, SNOW, DATA) over the next 6 months; prioritize companies with >40% recurring revenue and positive FCF.