The Stock Market's Paradoxical Doomsday: Artificial Intelligence Is Running Out of Gas yet Bound to Replace Software

Equity markets are repricing AI and software exposures after worries that massive AI-related capex by hyperscalers may not deliver commensurate returns and that AI will simultaneously disrupt SaaS business models. Key datapoints cited include a Lawrence Berkeley National Laboratory finding that by 2028 over half of data-center power may be used for AI—potentially equating to electricity for 22% of U.S. households—and a McKinsey estimate of $6.7 trillion in data-center spending by 2030; meanwhile new products such as Anthropic’s Claude Cowork intensify fears of software obsolescence. The net effect is sector-wide selling and a likely rerating of high-valuation SaaS names, though analysts (Bank of America) argue meaningful productization will take years, implying differentiated winners and losers ahead.

Analysis

Market structure: Near-term winners are AI infrastructure and middleware (NVIDIA NVDA, Broadcom AVGO, data‑center REITs EQIX/DLR) and energy/power suppliers that can monetize rising compute demand; losers are high‑multiple, low‑FCF SaaS names (SNOW, ZS) and hyperscalers (AMZN, GOOG, MSFT) facing heavy capex and uncertain incremental returns. Expect pricing power concentrated in GPU designers and specialized foundries for 12–24 months while software incumbents face margin compression as AI removes product development frictions and lowers switching costs.

Risk assessment: Tail risks include regulatory curbs (EU/US AI rules, export controls) that could depress GPU revenues by >20% in a quarter, large capex writedowns from hyperscalers, or energy/water rationing constraining data‑center growth. Immediate (days–weeks) triggers: earnings/capex guidance from NVDA, AMZN, MSFT; short‑term (1–6 months): model performance or Anthropic/OpenAI product launches; long‑term (1–5 years): moat erosion and normalization of SaaS multiples.

Trade implications: Favor 6–12 month exposure to NVDA (core long) and EQIX (data‑center real assets) and reduce/short selected high‑multiple SaaS; implement option collars to control downside. Use pair trades (long NVDA, short SNOW) to express infrastructure upside versus software‑multiple risk and favor energy/utilities (NEE, EXC) as hedge against spiking power costs.

Contrarian angles: Consensus overlooks that productizing model intelligence takes years—profitable, data‑rich SaaS (e.g., CRM owners like CRM) will likely reprice higher relative to speculative SaaS; the sell‑off is likely overdone for profitable SaaS trading <6x revenue with positive FCF. Historical parallel: cloud buildout (2010s) where infrastructure winners consolidated while marginal software margins compressed; unintended consequence—compute scarcity raises barriers for new entrants, reinforcing incumbents in chips and hyperscale.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVDA (AI compute) with 6–12 month horizon; add on any >10% pullback. Target +30–50% total return, set tactical stop‑loss at 18% to limit drawdown.
• Initiate a 1–2% position in EQIX (or DLR) to play data‑center demand over 12 months; target ~20% total return including dividend; trim if utilization falls <3% YoY or FFO guidance is cut >10%.
• Trim or sell 50% of exposure to SaaS names trading >10x trailing revenue and negative FCF (explicitly reduce positions in SNOW, ZS); establish up to 1% portfolio short in SNOW with 6–12 month horizon, profit target -40%, stop +25%.
• Implement options: buy a 3‑month NVDA call spread (buy ATM, sell 10–15% OTM) sizing 0.5–1% notional to capture upside while limiting premium; buy a 3‑month put spread on SNOW (1–2% notional) as a convex hedge vs multiple compression.
• If upcoming hyperscaler capex guidance is cut >15% QoQ (trigger within next 30–90 days), rotate an incremental 2% from hyperscalers (AMZN/GOOG/MSFT) into utilities (NEE/EXC) and long AI hardware names; reverse if guidance proves resilient.