Billionaire Dan Loeb Sold Shares of AI Leaders Including Amazon, Microsoft, and Meta and Added to His Position in This AI Player That's Soared 453,000% Since Its IPO

Dan Loeb (Third Point) oversees $7.2B in 13F securities and in Q4 reduced stakes in Amazon (-22%) and Microsoft (-15%), closed Meta entirely (previously ~1.8% weight), and increased Nvidia by just under 4% to >7% of the portfolio (now his No.2 holding). The moves come amid Q4 2025 turbulence for AI names (notably November declines) but signal continued conviction in Nvidia and the broader AI infrastructure story despite valuation concerns; Amazon remains ~6.8% and Microsoft ~6.1% of the fund. Meta rose ~140% while in Loeb’s portfolio, and the article highlights Nvidia’s extraordinary long-term IPO gain (~453,000%) and its central role in data-center AI demand.

Analysis

The dominant supplier of high-performance AI accelerators creates an ecosystem where winners are not limited to the chip vendor: hyperscalers capture recurring service margins as they monetize scarce inferencing capacity, OEMs that integrate GPU-heavy racks (networking, power, cooling) see durable orderbooks, and a robust secondary GPU/resale market emerges that can materially shorten real-world upgrade cycles. That oligopolistic supply structure amplifies pricing power but also concentrates tail risk — any single choke point (foundry allocation, export-control edict, or logistics failure) will propagate nonlinearly across cloud capacity and software deployment timelines.

Near-term catalysts cluster around procurement cadence and cadence mismatches between training and inference demand: bookings and cloud capacity guidance will move stock-level realized growth over quarters, while multi-year data-center build plans determine whether revenue growth is front-loaded or back-ended to 2028–2030. Reversals come from two plausible mechanisms — a macro-driven capex pause that defers purchases for 6–18 months, or a faster-than-expected adoption of more specialized inference silicon and software optimizations that cut GPU utilization intensity by 20–40% within 2–4 years.

Consensus positioning looks crowded on pure-play accelerator exposure and thinly hedged beta to AI sentiment. The non-obvious trade is to capture upside from continued capacity tightness while explicitly hedging liquidity/crowding risk: that means structured, time-limited option overlays or relative-value pairs that monetize the oligopoly economics without leaving the portfolio exposed to a headline-forced unwind.

Key Decisions for Investors

• Long NVDA via a 9–12 month 25% OTM call spread (buy 25% OTM, sell 60% OTM) sized to 2–3% of NAV; target asymmetric payoff of ~2:1 upside vs max premium loss if data-center spend stays healthy; roll or take profits at 50% realized gain or if implied vol jumps >40% from current levels.
• Pair trade: long NVDA / short META, equal-dollar, 6–12 month horizon. Size to 2% net delta, set stop-loss if NVDA/META ratio declines 15%; thesis captures GPU oligopoly vs discretionary ad/consumer capex risk — expect 20–40% relative spread expansion if enterprise AI monetization continues.
• Volatility calendar: buy 12-month NVDA calls and finance by selling 1–3 month calls (monthly cadence) to capture realized vol if moves are choppy; limit assignment risk by keeping net-long vega small (1–2% NAV) and close front-month leg on earnings or on >30% directional move.
• Defensive hedge: buy 6–9 month 15% OTM puts on a concentrated AI exposure basket (NVDA, AMZN, MSFT) sized to cap portfolio drawdown at ~8–10% if a sentiment-driven deleveraging occurs; this is insurance for a crowded-risk tail caused by liquidity shocks or regulatory actions.