Warren Buffett Has a Stark Warning for Investors as the Market Repeats a Pattern Seen Only Twice in 155 Years

The S&P 500's CAPE ratio stands at 41.6, near its 1999 dot-com-bubble peak of 44.2 and well above the 1929 peak of 32.6, prompting warnings that AI-driven market enthusiasm may be outpacing fundamentals. Generative AI has added an estimated $19 trillion of global market value, while OpenAI is projected to burn $280 billion in cash by 2030. Higher policy rates of 3.75%-4.0% and a 10-year Treasury yield just below 5% add valuation and financing pressure; Buffett urged investors to favor sustainable moats and reasonable valuations rather than speculative momentum.

Analysis

The relevant risk is not broad AI adoption but a narrowing of the AI investment loop: hyperscaler capex is currently both NVDA’s demand engine and the source of incremental earnings expectations embedded across semis, power infrastructure and data-center supply chains. A modest reduction in MSFT, AMZN, GOOGL or META capex guidance would likely produce a nonlinear de-rating in NVDA because investors would reassess both shipment durability and terminal margins; the first warning signs would be lead-time normalization, rising GPU secondary-market supply, or materially lower networking attach rates. This is a 1-3 month earnings-call risk, not necessarily a 6-18 month technology-adoption bear case.

High long-duration equity valuations are particularly exposed if real yields remain elevated: the mechanical impact is greatest for unprofitable AI software and private-market-adjacent names, but a broad multiple reset would also pressure profitable mega-cap beneficiaries. BRK.A is comparatively attractive as a volatility absorber because its operating earnings, insurance float and deployable liquidity create optionality if equity dislocations force attractive acquisitions or buybacks. The second-order winner from an AI capex pause is not necessarily value broadly, but cash-generative, low-capex businesses whose earnings are less dependent on financing conditions.

Consensus is too binary on NVDA: a capex digestion event need not imply collapsing AI demand, but it can still make the equity vulnerable because the market prices sustained exceptional growth and assumes customers continue funding capacity ahead of monetization. Conversely, a shallow pullback should not be treated as confirmation of a bubble; the bear thesis is falsified if the major hyperscalers collectively maintain or raise 2027 AI capex while NVDA sustains gross margin and data-center revenue growth above expectations. The cleaner expression is relative rather than an outright index short, given concentration in profitable large-cap technology and uncertain timing of a valuation catalyst.

Key Decisions for Investors

• Initiate a 1-3 month hedge: buy NVDA put spreads 10-15% below spot, funded only partially with far-out-of-the-money puts rather than naked short stock. Target a 2-3x payoff on a post-earnings capex-guidance reset; exit if NVDA guidance and major customer capex outlooks remain intact.
• Pair long BRK.B / short an equal-dollar basket of high-multiple AI infrastructure proxies, with NVDA as the liquid hedge leg if basket construction is unavailable. Hold 3-6 months; the thesis is resilience to higher-for-longer yields and capital redeployment optionality, not a near-term earnings surprise.
• Do not add to broad semiconductor beta ahead of hyperscaler reporting without verifying aggregate capex guidance from MSFT, AMZN, GOOGL and META. Set an alert for two or more firms cutting forward AI capex or signaling lower deployment velocity; that would justify expanding the NVDA hedge and adding short SOXX exposure.
• For long-only AI exposure, favor staged entries after earnings rather than chasing momentum: require evidence that AI revenue monetization is improving alongside capex, not merely continued infrastructure spend. A sustained rise in 10-year yields above 5% would warrant reducing long-duration growth exposure regardless of near-term AI demand.