The AI "Capex Trap" Is Wall Street's Biggest Fear Right Now. Here's Why I Think It's Overblown.

Amazon plans ~$200B in capex this year after $132B in 2025; revenue rose 12% (AWS +20%) while free cash flow fell to $11B in 2025 from $38B in 2024, yet book value increased to $411B from $286B. Alphabet plans $175–185B in capex (vs $91B in 2025); 2025 revenue grew 18% (Google Cloud +48%), free cash flow edged up to $73.3B from $72.8B, and book value rose to $180B from $125B. The article argues these large capex increases appear accretive for both companies and may create buying opportunities if share prices fall.

Analysis

The capex race is increasingly a scale-and-irreversibility story: winners will be those who convert large, lumpy investment into durable unit-cost advantages (lower $/inference, higher utilization of proprietary stacks) while losers are firms that face commoditization of services and underutilized datacenter capacity. Expect the market to bifurcate between integrated hyperscalers able to monetize bespoke infrastructure and smaller cloud/AI providers that must either raise prices, sell into narrow niches, or consolidate.

Second-order supply-chain effects matter more than headline chip demand: power-delivery, liquid-cooling, high-end networking, and specialty OS/stack software see multi-year tailwinds and pricing power; foundry constraints will keep GPU and accelerator pricing sticky while ramp cycles create discrete windows of margin pressure. Watch capex cadence vs. utilization — sustained overbuild will first show up in slower revenue/kw, then in negative working-capital turns and finally in one-off impairments.

Main tail risks are macro tightening, slower enterprise AI adoption, and regulatory interventions that limit data mobility or multilocation deployments; these can compress multiples quickly because a big portion of the upside is option-like (future monetization of AI services). For investors, the highest-probability re-rating window is 6–18 months after demonstrable improvement in incremental ROIC or cloud gross margins, not immediately on capex announcements. Consensus is underweight the durable asset-side benefits (capitalized scale) while overestimating straight-line impairment risk — position sizing should skew toward optionality rather than outright leverage.

Key Decisions for Investors

• Long AMZN via calendar/LEAP call spread: buy a 12–24 month call spread to capture re-rating if AWS and consumer mix show sustained margin expansion; risk = premium paid (~2–4% of notional), reward = 25–50% upside to target if re-rating occurs within 12–18 months. Add a 10% OTM protective put if funding is available (limits tail risk on macro shock).
• Long GOOGL outright with phased entries: accumulate on any 5–12% pullback over 3 months to target a 20–40% return over 6–18 months tied to cloud monetization milestones; size to tolerate 15% drawdown and add if sequential cloud revenue/adj. margins beat.
• Relative pair: long NVDA exposure (calls/dispersion) vs short INTC (put spread) — play secular GPU share gain + tight foundry supply. Structure: buy NVDA 9–18 month calls (limited premium) and fund with an INTC 6–12 month bear put spread. Expect asymmetric payoff: limited cost with 2–4x upside if NVDA sustains premium and INTC misses volumetric recovery.
• Defensive hedge: buy NDAQ exposure or short-duration puts on exchanges as a volatility hedge to portfolio tech longs — exchange fees typically hold up in volatility and offer idiosyncratic insurance if market liquidity tightens. Target a 6–12 month horizon and keep position size small (1–3% NAV) as tail protection.