Billionaire Philippe Laffont Has Sold Shares of Nvidia in 10 of the Last 11 Quarters -- What Does He Know That Wall Street Doesn't?

Philippe Laffont has sold >40 million split-adjusted Nvidia shares since 31-Mar-2023, cutting Coatue’s holdings from 49.802M to 9.203M shares (≈81.5% decline) through Q4 2025. While some sales likely reflect profit-taking after Nvidia’s >1,200% rally since Q3 2022, the article highlights material risks that could justify continued selling: rising GPU competition from large customers (cheaper in-house chips), China’s reluctance to buy H200 GPUs, and tariff/trade uncertainty—all threats to Nvidia’s pricing power and gross margins.

Analysis

Hyperscaler vertical integration is the clearest second‑order threat to NVDA’s pricing power: if the largest customers shift 20–30% of incremental AI capacity to in‑house silicon over 12–36 months, that could mechanically shave 200–600bps off NVDA data‑center gross margins as high‑ASP GPU mix falls. That rotation would also create a bifurcated aftermarket — elevated supply of used datacenter GPUs (pressuring new unit pricing) and accelerated demand for orchestration, compiler and model‑optimization software that extracts more from cheaper silicon.

Trade policy and China availability act like a volatility amplifier rather than a binary risk. A 6–18 month tightening of exports or higher tariffs would turn near‑term revenue into a stop‑start cadence, magnifying quarter‑to‑quarter EPS volatility even if secular demand remains intact — this increases option premia and short‑dated event risks while lengthening the time horizon for any valuation rerating.

Given current positioning, the path to downside is likely faster than the path back up: a 30%+ drawdown in NVDA within 6–12 months is plausible if macro slows and cloud customers pause optimization spend, but recovery would require multi‑quarter proof points (repeat orders, higher ASPs, or China re‑entry). That asymmetry favors protective structures and relative value trades (long other parts of the stack or shorts into gamma) over naked long exposure at current multiples.

Key Decisions for Investors

• Pair trade (6–12 months): Short NVDA vs Long INTC, notional ~0.5–1.0% NAV each leg. Thesis: capture relative weakness from margin erosion / vertical integration while participating in Intel’s potential share gain on lower‑end inference and edge deployments. Target relative return 25–40%; hard stop if pair moves 12–15% against position on a mark‑to‑market basis.
• Protective option structure (9–12 months): Buy 25% OTM NVDA puts sized to cover 50–75% of existing NVDA long exposure, financed by selling 2–3 month call spreads (roll monthly into event windows). Expect to pay a small carry (~1–3% annualized) for tail insurance while harvesting rich short‑dated IV; this reduces left‑tail risk while retaining upside participation.
• Tactical income trade (30–60 days around catalysts): Sell short‑dated NVDA call spreads (2–4 week tenor) into anticipated positive headlines or earnings when implied vol spikes, size as <0.25% NAV. Capture theta; keep risk capped to the width of the spread (max loss known) and exit into any adverse directional move >10%.
• Long asymmetric exposure to AI software/optimization names (12–36 months): Reallocate 0.5–1.0% NAV from naked NVDA longs into companies that monetize optimizations on cheaper silicon (compilers, model optimizers, deploy tooling). Rationale: these businesses gain if customers prioritize cost/efficiency over raw FLOPS; target 2–4x upside vs 1x downside on a multi‑year horizon.
• Liquidity & monitoring rule: Reduce NVDA gross exposure ahead of any new China/export guidance and re‑establish only after two consecutive quarters of order stability or visible customer reorder cadence. Treat trade policy announcements as 48–96 hour event windows — tighten stops and scale hedges during them.