Warren Buffett Says This Move Is the Key to Surviving a Stock Market Crash. Here's What History Says Comes Next

The article cautions that the generative-AI investment boom may exhibit bubble-like characteristics despite the S&P 500 gaining 70% over the past five years, led largely by technology companies. Applying Warren Buffett's emphasis on durable economic moats, it flags Micron Technology's AI-driven memory boom as potentially vulnerable to commoditization, competing supply from Samsung, SK Hynix and Chinese entrant CMXT, and a future supply glut. Investors are advised to prioritize sustainable competitive advantages over near-term AI-driven growth.

Analysis

The relevant differentiation within AI hardware is not whether demand remains strong, but where incremental returns accrue once capacity catches up. MU has unusually high operating leverage to HBM pricing and bit growth, yet its product economics remain exposed to qualification-driven share shifts among Samsung and SK Hynix; a modest oversupply can erase a disproportionate share of peak-cycle gross margin. The market should therefore value MU on mid-cycle earnings power rather than current AI-memory spot tightness, creating downside risk over the next 6-18 months if capex plans convert into supply faster than inference demand broadens.

NVDA is less directly vulnerable to memory commoditization because its software ecosystem, system integration and developer lock-in preserve pricing power, but it is not immune to a second-order memory constraint: HBM availability and pricing can determine rack shipment timing and customer ROI. If HBM supply normalizes, lower system BOM costs could support hyperscaler AI deployment and benefit NVDA volumes; if Chinese HBM reaches credible qualification, the near-term effect is more likely margin pressure on memory vendors than disruption to NVDA's premium platform.

Near-term positioning risk is asymmetric for MU: earnings upside requires continued elevated HBM pricing and clean execution, while downside can emerge from any indication that customer inventory is rebuilding or that 2027 supply commitments are expanding. The contrarian point is that a broad "AI bubble" short is low quality while cash-generative platform owners retain monetization optionality; the more attractive expression is to separate cyclical component exposure from durable ecosystem exposure. BRK.A is a defensive relative beneficiary if AI-capex multiples compress, though its exposure is too diffuse for this to be a primary catalyst.

Key Decisions for Investors

• Initiate a 6-12 month pair: long NVDA / short MU in equal dollar beta-adjusted amounts. Thesis is moat and recurring software economics versus peak-cycle memory margins; target 15-25% relative return. Exit if MU guides sustained gross-margin expansion despite rising industry capacity, or if NVDA data-center growth decelerates materially for two consecutive quarters.
• For existing MU longs, reduce into strength and replace directional exposure with a 3-6 month collar or put spread around earnings. The key watch items are HBM contract-price direction, Samsung qualification progress, SK Hynix capacity additions, and evidence of Chinese supplier qualification at major hyperscalers; absent these data, do not add on AI-demand headlines.
• Maintain NVDA core exposure but avoid chasing near-term beta: add only after post-earnings volatility or on evidence that inference revenue, rather than training capex alone, is supporting customer ROI. Falsification is a material reduction in hyperscaler capex plans coupled with inventory growth in NVDA's supply chain.
• Use BRK.B as a modest 6-18 month quality/valuation hedge against a broad AI multiple reset rather than shorting AI indices outright. A sharp rate decline and continued AI-led index concentration would weaken this relative-defense trade.