Why it’s time to reset our expectations for AI

The piece argues that AI has been overhyped and that expectations should be reset: leading models are impressive but have not delivered on extreme claims (e.g., solving climate change or eliminating whole job classes). For investors this suggests a potential shift to a post‑hype phase where capital will be more selectively allocated toward practical, high‑value AI applications, with continued large bets but greater scrutiny on technological maturity, real‑world returns and safety implications.

Analysis

Market structure: The AI "hype correction" shifts economic value from speculative app-layer valuations to firms owning cloud, GPU supply chains, and data-center ops—clear winners include GOOGL/GOOG, MSFT, AMZN and NVDA; losers are small-cap AI pure-plays with little revenue and sky-high EV/NTM multiples. Competitive dynamics will favor incumbents who control compute and data; pricing power in high-end GPUs and cloud compute should persist while marginal AI app vendors face faster multiple compression. Supply/demand: enterprise demand for inferencing/training remains multi-quarter persistent even as investor enthusiasm normalizes, so expect continued double-digit growth in cloud compute demand but tight supply for top-tier accelerators on 6–18 month horizons.

Risk assessment: Tail risks include accelerated regulatory enforcement (EU AI Act/US oversight) within 6–18 months, a major model failure or safety incident that triggers demand pullback, or a VC funding drought causing 30–60% markdowns in private valuations. Immediate risk (days) is sentiment-driven volatility; short-term (weeks–months) is earnings/guide-driven re-rating; long-term (quarters–years) depends on actual productivity gains and compute cost curves. Hidden dependencies: access to top GPUs, electricity/thermal constraints, and concentration of ML talent at a handful of firms—watch GPU spot pricing and hourly cloud instance utilization as second-order indicators.

Trade implications: Prefer long exposure to cloud/infra leaders and select semiconductors while trimming pure-play AI apps; use capital-efficient options to express convexity. Specific mechanisms: accumulation of GOOGL on >15% pullback vs 52-week high or phased 1–2% buys over 4–12 weeks, 0.5–1% allocation to 6–12 month NVDA calls to capture compute tightness, and a 1–2% short basket of small-cap AI SaaS stocks with EV/NTM >15x and negative FCF. Hedge portfolio downside with 3-month GOOG put spreads (5–10% OTM) sized 0.5–1% of NAV; re-assess after next two reporting cycles.

Contrarian angles: The consensus underestimates durable monetization opportunities from enterprise automation that lift cloud margins over 2–4 years, which would disproportionately benefit platform incumbents. Reaction to hype correction may be overdone for quality infra names (GOOGL, NVDA) while underdone for earnings risk at speculative AI apps—this creates pair trade opportunities. Historical parallels (post-2000 tech shakeouts) suggest concentration in a few winners; monitor GPU spot prices falling >30% or major regulatory fines as triggers that would invalidate the infra-long thesis.

Key Decisions for Investors

• Establish a phased long position in GOOGL: start 1–2% of portfolio now, add to 1–2% more if shares drop >15% from 52-week high; reassess after next two quarterly earnings (within 3–6 months).
• Allocate 0.5–1.0% of portfolio to 6–12 month NVDA call options (call spreads if premium expensive) to capture continued GPU tightness; exit or roll if NVDA implied volatility falls >25% or GPU spot prices decline >30%.
• Initiate a 1–2% short basket of 8–12 small/micro-cap AI pure-plays (criteria: EV/NTM revenue >15x, negative FCF, >50% revenue guidance sensitivity to AI claims) and size stops at 30% loss per name; rebalance monthly.
• Buy a defensive 3-month GOOG put spread (5–10% OTM) sized ~0.5–1% of NAV to hedge near-term sentiment/earnings risk; unwind if implied vol for GOOG compresses >40% or post-earnings volatility normalizes.