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

The article warns that the generative-AI rally, which has helped lift the S&P 500 70% over the past five years, carries characteristics of a speculative bubble despite being led by profitable technology incumbents. Applying Warren Buffett's dot-com-era framework, it argues investors should prioritize durable competitive moats over near-term AI-driven growth. Micron is highlighted as exposed to a potential memory-chip boom-and-bust cycle, as commoditization, competition from Samsung and SK Hynix, and potential Chinese HBM supply could pressure future profitability.

Analysis

The relevant fault line is not whether AI compute demand persists, but where excess returns settle once the hardware stack normalizes. MU's earnings are unusually exposed to the HBM pricing/supply balance: qualification gains and tight supply can sustain elevated margins over the next 1-3 quarters, but memory has historically converted scarcity into capex and inventory overshoot faster than logic semiconductors. A shift in customer inventory days, HBM yield improvement at SK Hynix/Samsung, or credible Chinese qualification would matter more to MU's multiple than another incremental AI-demand headline.

NVDA is relatively insulated from a memory downcycle because its economic moat is system-level—software, developer lock-in, networking, and platform integration—but it is not immune to a lower return-on-invested-capital equation for hyperscalers. If memory and server-component costs fall, cloud customers may preserve AI build rates by expanding units; alternatively, falling component prices could expose that AI infrastructure spending is being pulled forward. The differentiator is hyperscaler capex guidance and GPU utilization, not aggregate semiconductor revenue.

Consensus skepticism around AI hardware is directionally sensible but may be early for MU: HBM supply is qualification-constrained, so a standard DRAM-cycle playbook can miss the duration of tightness. The more attractive contrarian framing is to separate a 6-12 month memory-cycle hedge from a 12-24 month short thesis: HBM commoditization requires verified competitor yields, customer approval, and sufficient packaging capacity. BRK.A is a modest defensive beneficiary only if a risk-off rotation rewards balance-sheet optionality; it is not a direct AI-bubble hedge.

Key Decisions for Investors

• Maintain a tactical long MU only through the next 1-2 earnings cycles if HBM bit shipments, pricing, and gross-margin guidance continue to rise; use a 10-15% downside stop or exit on the first material guide-down in HBM revenue/margin. Upside is continued estimate revisions, while the key risk is accelerated Samsung/SK Hynix supply qualification.
• For a 6-18 month cycle hedge, consider a MU short versus long NVDA pair after MU materially outperforms SOX on peak-margin expectations. The thesis is that MU's earnings multiple compresses first when memory supply loosens, while NVDA retains greater software/platform pricing power; cover if NVDA hyperscaler demand or GPU utilization weakens materially.
• Do not short AI semiconductors solely on bubble rhetoric. Set an alert for sequential declines in hyperscaler capex guidance, AI-server lead times, or HBM contract pricing; those are the evidence-based triggers for reducing NVDA/MU exposure rather than valuation comparisons alone.
• Add BRK.A as a modest portfolio ballast rather than a directional AI expression over the next 3-6 months if AI concentration risk rises; reassess if equity-market volatility fails to produce deployable capital opportunities or if insurance underwriting trends deteriorate.