If a Bear Market Is Coming, These Are the 2 Stocks to Avoid -- and the 1 to Own

The article warns that an AI-spending slowdown could amplify a potential bear market, citing Oracle's negative $5.4B quarterly free cash flow, $125B of long-term debt and $288B of unstarted data-center lease commitments. CoreWeave is portrayed as more exposed, with $35.6B of debt versus $5.5B of cash, $640M in quarterly interest expense, and a recent $4.2B convertible-debt offering. Berkshire Hathaway is positioned as a relative defensive beneficiary, holding roughly $365B in cash and U.S. Treasury bills available for deployment during a market downturn.

Analysis

The relevant fault line is not AI demand alone but the financing spread between contracted capacity and cash-generative capacity. Backlog is not equivalent to funded, cancellable-proof revenue: if customer credit quality weakens or deployment schedules slip, fixed lease, interest, and hardware obligations remain while utilization falls. This creates materially higher equity convexity at CRWV than ORCL; ORCL’s legacy software cash flows provide a cushion, but also make a capex-led FCF reset more likely to trigger multiple compression rather than an immediate solvency event.

Over the next 1-3 months, the key catalyst is whether AI infrastructure operators demonstrate improving operating cash flow before requiring incremental debt or equity-like financing. Watch CRWV unsecured/convertible pricing, ORCL credit spreads, customer prepayments, capex guidance, and RPO conversion into recognized cloud revenue; widening spreads or another financing round would challenge the "demand-funded growth" narrative. The falsifier is sustained utilization and gross-margin expansion sufficient to produce positive post-capex FCF while customer concentration declines.

A neocloud retrenchment would not necessarily be bearish for all AI infrastructure. Compute demand could migrate toward better-capitalized platforms such as MSFT, AMZN, and GOOGL, which can absorb capacity internally and bundle AI services with existing enterprise relationships. NVDA faces a second-order risk only if cancelled or deferred data-center builds reduce aggregate GPU procurement; in a more likely consolidation scenario, its orders shift from leveraged intermediaries to hyperscalers, preserving volumes but reducing urgency-premium and potentially lengthening order cycles.

BRK.A is a useful liquidity hedge rather than a pure beta hedge: its optionality becomes valuable only after asset prices and credit availability deteriorate enough to create attractive deployment opportunities. Consensus may overstate its immediate downside protection in a shallow growth scare, where cash earns less relative to risk assets and no distressed opportunities emerge; the relative trade works best in a genuine credit-tightening regime, not a soft landing.

Key Decisions for Investors

• Initiate a 3-6 month relative-value position: long BRK.B / short CRWV, sized beta-neutral. Target 15-25% relative return if financing conditions tighten; cover the short if CRWV reports two consecutive quarters of positive post-capex FCF, stable or improving customer concentration, and no need for incremental external financing.
• Use CRWV put spreads rather than outright puts ahead of its next earnings/financing update: buy 3-6 month at-the-money puts and sell 20-30% out-of-the-money puts. This captures downside from credit repricing while limiting premium loss if the stock remains supported by AI momentum; avoid naked short exposure given high squeeze risk around large contract announcements.
• Maintain ORCL as an underweight or hedge against AI-capex longs rather than a core directional short. Add only if management raises capex or lease commitments without a corresponding increase in customer prepayments, OCI margin guidance, or post-capex FCF visibility; a verified acceleration in RPO conversion and improving FCF would invalidate the thesis.
• For AI exposure, prefer hyperscaler beneficiaries MSFT and AMZN over leveraged GPU-rental intermediaries for the next 6-18 months. Reassess NVDA only if cloud-provider capex guidance turns negative in aggregate; a slowdown isolated to CRWV/ORCL would more likely represent demand migration than a broad GPU demand collapse.