'Big Short' investor Burry targets AI boom with new blog after hedge fund closure

Michael Burry has launched a paid Substack newsletter, Cassandra Unchained, with over 21,000 subscribers at $39/month and signaled he will continue publicly critiquing markets after shutting his hedge fund Scion Asset Management and returning investor capital. In early posts he likens the current AI boom to the 1990s dot‑com bubble, singled out Nvidia as the central “picks and shovels” play for AI infrastructure and criticized major cloud providers and tech names for aggressive accounting and frothy investment rhetoric. Burry’s public stance and historical influence on investor sentiment add a cautionary voice to AI and big‑tech positioning, though the move is more likely to sway sentiment than trigger immediate market dislocations.

Analysis

Market structure: The immediate beneficiary is GPU/infrastructure exposure (NVDA, key ODMs, memory vendors) where constrained supply and enterprise capex can sustain 20–40% gross margin expansion vs legacy CPUs. Losses cluster in high multiple cloud/apps (MSFT, META, PLTR-unproven monetization) where aggressive accounting and re-rating risk make multiples vulnerable if AI ROI timelines slip. Cross-asset: expect a 15–40% rise in short-dated tech option IV on headline shocks, transient safe-haven flows into Treasuries (yields down 10–30bps) and cash bid pressure in industrial materials tied to data-center builds (copper, power equipment).

Risk assessment: Tail scenarios include a rapid demand collapse (20–50% revenue downgrades for AI services), export controls or antitrust action that cuts China-accessible GPU revenue by >15%, or a liquidity-driven gamma squeeze amplifying moves. Near-term (days) risk is sentiment-driven volatility; 1–3 months is guidance/earnings repricing; 12–24 months is capex cycle and wafer capacity changes. Hidden dependencies: hyperscaler ordering cadence, foundry lead-times (12–24 months), and China exposure; monitor customer concentration where top 3 buyers account for >40% of GPU orders.

Trade implications: Favor concentrated, hedged NVDA exposure: 2–4% portfolio long via calendar call spreads (3–6 month) with a 5–15% OTM bought call and 10–20% OTM sold call to cap cost, plus a 5% cost buy-protective put for downside. Pair trade: long NVDA vs short MSFT (1–2% notional) via buying NVDA call spreads and MSFT 3-month 5–10% OTM put spreads; reduce broad mega-cap growth weight by 3–5% and redeploy to semis and infrastructure stocks with free cash flow yield >4%.

Contrarian angles: The market underestimates durable pricing power for proprietary accelerators — survivors can sustain 30–50% incremental margins vs consensus. Conversely, consensus may be complacent about hyperscaler margin erosion; if MSFT/META guidance slips by >10% in next two quarters, expect a rapid 20–35% multiple compression. Watch short interest thresholds (>8–10%) and order-flow spikes as triggers for short squeezes or capitulation buying.

Key Decisions for Investors

• Establish a 2–4% portfolio long in NVDA using 3–6 month call spread structures: buy a 7–12% OTM call and sell a 15–25% OTM call to cap premium; simultaneously buy a 5% cost protective 3-month put. Trim/lock 50% of position if NVDA rallies >25% or implied volatility falls >30% from entry.
• Implement a pair trade: short MSFT equal to 50–60% dollar exposure of NVDA long via 3-month 5–10% OTM put spreads (size 1–2% portfolio). Close if MSFT reports quarter-on-quarter AI-related revenue beat >5% or if MSFT down >15% from entry (stop-loss).
• Reduce broad mega-cap growth exposure by 3–5% within 2–6 weeks and redeploy into semiconductor suppliers and data-center infrastructure names with FCF yield >4% (target ORCL if trading >20% below historical EV/EBITDA and free cash flow remains stable).
• Monitor three triggers for active management: (1) hyperscaler order cadence announcements or downgrades within 30–90 days, (2) NVDA short interest >10% or daily retail buy/sell imbalance >3x average, (3) regulatory actions (export/antitrust) filed publicly — act within 48–72 hours of any trigger with pre-defined cut/loss limits of 10–15%.