Warren Buffett Broke One of His Most Important Investing Rules, and It's Cost Berkshire Hathaway $16 Billion (and Counting)

Berkshire Hathaway, under Warren Buffett, bought 60,060,880 shares of Taiwan Semiconductor Manufacturing (TSMC) in Q3 2022 for about $4.12 billion, then sold ~86% (51,768,156 shares) in Q4 2022 and fully exited in Q1 2023—citing location/export concerns after the CHIPS Act. TSMC subsequently benefited from surging AI GPU demand and CoWoS capacity expansion, joined the $1 trillion market-cap club in July 2025, and the initial stake would be worth nearly $20 billion as of Jan. 26, implying Berkshire forewent roughly $16 billion in unrealized gains. The episode underscores conflict between short-term geopolitical/regulatory risk assessment and a long-term investment mandate, and signals governance and positioning considerations for Berkshire under new CEO Greg Abel.

Analysis

Market structure: The article underscores a durable winners-take-most dynamic in advanced-node foundry economics — TSM (TSM) and its GPU customers (NVDA) capture outsized margin expansion because CoWoS capacity is scarce and high-ASP. Losers are incumbent, legacy-capacity suppliers (INTC) and any fab that cannot secure EUV/advanced packaging tools; expect pricing power for advanced wafers to persist for 12–24 months while backlog clears.

Risk assessment: Key tail risks are geopolitical (cross‑strait military escalation) and regulatory (expanded export controls beyond high‑end GPUs), both low probability but high impact (>=50% equity drawdown in Taiwan assets). Near-term (days–months) volatility will be driven by policy headlines and 2–4 quarters of capacity announcements; multi‑year upside relies on sustained AI GPU demand versus capex execution and equipment supply (ASML constraint risk).

Trade implications: Position sizing should be asymmetric — express AI/TSM exposure but hedge geopolitical/regulatory tails. Prefer concentrated, time‑bounded option structures to capture upside while capping downside; consider relative-value trades that long foundry/GPU exposure and short legacy CPU/IDM names that face secular share loss (INTC). Rebalance as TSM monthly CoWoS capacity updates and NVDA booking commentary arrive (next 30–90 days).

Contrarian angles: Consensus exalts TSM as unassailable; the miss is underpricing policy and concentration risk (NVDA represents large share of advanced node demand). History (Intel’s mobile miss) shows technological leadership can be transient if customers vertically diversify or policy forces onshore builds — price in a 20–40% geopolitical haircut when sizing positions.

Key Decisions for Investors

• Establish a 2–3% portfolio long in TSM (TSM) via dollar‑cost averaging over 6–12 weeks; add another 1% if TSM retraces 10–15% from current levels. Hold horizon 12–36 months; if Taiwanese geopolitical tensions escalate (measured by US/Taiwan military alerts or sanctions announcements) cut exposure by 50% immediately.
• Implement a 1–2% notional 9–12 month call‑spread on NVDA (buy near‑ATM long call, sell 20–30% OTM call) to capture continued GPU demand while capping premium spend; roll or convert to equity on materially positive bookings/earnings surprises.
• Execute a pair trade: long 1.5–2% TSM (equity or LEAP) and short 1–1.5% INTC (equity) to express structural foundry share gains vs IDM headwinds; tighten short if Intel announces credible advanced‑node capacity commitments within 6 months.
• Buy asymmetric tail insurance: allocate 0.3–0.6% of portfolio to 9–12 month deep OTM puts on TSM or a Taiwan ETF as geopolitical hedge. Monitor CHIPS Act policy updates and export‑control communications over next 30–60 days — if US policy expands export curbs to TSM inputs/fabs, increase hedge to 1% and reduce longs by 30%.