The volatile AI trade marched higher, but oil kept Wall Street on edge last week

AI and semiconductors drove a choppy week: the VanEck Semiconductor ETF moved from about +2% Monday to ~-4% Tuesday after Samsung’s results and DeepSeek’s AI-chip plans, then rebounded as Apple expanded its Broadcom partnership to exceed $30B and 15B+ U.S.-made chips (Broadcom +~5%). Meta surged +15% for the week on cloud-compute plans and AI model launches (Muse Image, Muse Spark 1.1) alongside plans to begin manufacturing a custom AI chip in September. Meanwhile, renewed U.S.-Iran tensions pushed WTI back toward $76/bbl and lifted the 10-year Treasury yield to its highest since May, pressuring rate-sensitive and higher-input-cost names and keeping overall direction fragile despite the S&P 500’s +1.23% weekly gain.

Analysis

The biggest signal is not that AI demand is weakening; it is that the marginal winners are shifting from pure model exposure to the companies that can monetize infrastructure and control silicon design. That favors META and AVGO, with TSM as the quieter toll collector: custom chips and long-cycle manufacturing commitments reduce pricing elasticity and improve visibility, while merchant accelerator names become more exposed to mix risk if customers keep in-sourcing workload-specific silicon. Over 1-3 months, the market should reward firms turning capex into either external revenue or lower unit compute cost, and punish those still selling only commodity AI exposure.

The oil move matters less as an outright energy thesis than as a cross-asset inflation shock. If crude holds near the mid-70s for several weeks, the second-order losers are rate-sensitive, fuel- and travel-adjacent names like HD and HONA, because higher yields delay the housing rebound and compress aftermarket activity tied to flight miles. Energy equities can work tactically, but this is more of a hedge against macro re-pricing than a clean long; if headlines fade and crude slips back below recent breakout levels, the trade reverses quickly.

The contrarian miss is that investors are treating the AI complex as one trade when the fundamentals are fragmenting. Broad selloffs on chip headlines may be overdone for AVGO/TSM/META because those businesses are becoming more vertically integrated and less dependent on generic GPU scarcity, while the more fragile part of the trade is any name whose valuation depends on uninterrupted third-party capex intensity. Watch for next earnings to confirm whether META's compute monetization shows up in revenue line items or just narrative; without that, the move can still retrace, but the direction of travel is toward better capital efficiency, not peak AI fatigue.

Key Decisions for Investors

• Long META on any pullback over the next 1-3 weeks; target 2:1 risk/reward into the next earnings print if management keeps monetizing compute and ad tooling. Falsify if capex rises without a credible external revenue bridge.
• Long AVGO versus short AMD as a 1-3 month pair trade; thesis is that custom ASIC share and platform control matter more than generic accelerator exposure. Stop if AMD's AI guidance reaccelerates or AVGO fails to convert design wins into margin.
• Add TSM as a quieter beneficiary of the custom-silicon cycle for a 6-18 month horizon; this is a lower-beta way to own the supply-chain reconfiguration. Falsify if advanced-node utilization or packaging demand softens.
• Use HD as a short or underweight against a rising-yield / rising-oil backdrop for the next 1-3 months. The trade works if 10-year yields stay elevated and fuel costs keep pressure on discretionary home-improvement demand; cover if rates roll over decisively.
• Keep COP/CVX only as a tactical geopolitical hedge, not a core long. Size small and reassess if WTI fails to hold the recent spike, because the inflation impulse can fade faster than equity positioning resets.