Patrick Freyne: AI’s weird billionaires want us all to be atomised units of labour and consumption

A sharply negative critique argues mass-market large language models centralize power among wealthy tech actors, impose substantial environmental costs from large data centres, and rest on unprofitable, bubble-prone private-market dynamics (noting examples such as Nvidia’s investment ties to OpenAI and industry leaders warning of investor overexcitement). The piece highlights operational risks — hallucinations/misinformation, manipulative UX design, limited productivity upside (citing an MIT finding that 95% of adopters saw no productivity gains), copyright litigation and potential labour displacement — signaling elevated regulatory, legal and reputational risk for investors in AI platforms.

Analysis

Market structure: Big incumbents (NVIDIA NVDA, Google GOOGL/GOOG, cloud providers) remain the likely ultimate winners because network scale, chip access and capital intensity create high barriers; smaller AI-native startups and content creators face value capture risk and margin collapse. Demand signals point to sustained appetite for datacenter capacity and Nvidia H100/A100-class GPUs — expect supply tightness for leading nodes through H1 2026 and pricing power for top-tier fabs/vendors.

Risk assessment: Tail risks include rapid regulatory intervention (antitrust/copyright/energy taxes) or a private-market funding crunch that deflates valuations — probability materially rises if a major copyright ruling goes against LLM training within 3–9 months. Hidden dependency: industry profitability hinges on a few chip suppliers and power grids; a single large-capacity outage or Nvidia supply shock could cause >20% re-rating of model-driven names.

Trade implications: Favor concentrated exposure to deep-moat infrastructure (NVDA long, EQIX/colocation) and underweight/short highly-marketed, loss-making AI apps that trade on narrative (small-cap AI names, selective private-to-public proxies like RDDT). Use options to express asymmetric views: defined-risk call spreads on NVDA ahead of catalysts and put spreads on speculative AI names to hedge downside over 3–6 month windows.

Contrarian angles: Consensus conflates AI hype with productivity gains — history (2000s tech bust) shows infrastructure providers consolidate market share while marginal incumbents fail; the market may be over-discounting Google’s diversification and under-discounting Nvidia’s choke-point role. If carbon/regulatory costs force consolidation, winners’ free-cash-flow yields could expand by 300–500bp over 12–36 months.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA within 4–6 weeks; implement a 3-month call spread (buy ATM, sell +20% OTM) sized to 2% notional, target +15–25% on a positive earnings/capacity beat, stop-loss -12%.
• Initiate a 1–2% short exposure to speculative/loss-making AI-branded equities; prefer 3-month put spreads (buy 20% OTM, sell 40% OTM) on names that rally >15% without revenue guidance improvements—example: size a 1% notional short in RDDT via put spread if fundamentals don’t improve by next quarter.
• Rotate 1.5–2% into datacenter infrastructure: buy EQIX (1–1.5%) and select regulated utility exposure (NEE or AES 0.5%) to capture persistent power demand; horizon 6–24 months, take profits if shares appreciate >20% or if wholesale power costs spike >30%.
• Monitor regulatory catalysts over the next 60 days (EU AI Act votes, major US copyright rulings, DOJ/FTC filings); if probability of strict regulation >50% (per press/legal filings), expand puts on small-cap AI names by +50% notional and buy 6–12 month protective puts on GOOGL/GOOG sized 1%.