Does Philippe Laffont Know Something Wall Street Doesn't? The Billionaire Investor Just Sold Nvidia and AMD, and Bought These Other Chip Stocks Instead

Coatue Management materially rebalanced its Q3 equity exposure, trimming stakes in GPU leaders Nvidia and AMD by 14% and 19% respectively while more than tripling positions in Alphabet and Marvell Technology, signaling a shift from general-purpose accelerators toward broader AI infrastructure plays. The move underlines confidence in expanding AI data-center demand (McKinsey forecasts nearly $7 trillion in AI infrastructure investment over five years) and highlights Marvell's role in HBM, networking and storage, and Alphabet's TPU-driven cloud push; Nvidia and AMD have still delivered outsized gains (~900% and ~200%) during the AI cycle, indicating this is a portfolio diversification rather than a capitulation on GPUs.

Analysis

Market structure: Coatue’s trim of NVDA (-14%) and AMD (-19%) in favor of GOOGL and MRVL signals a rotation from concentrated GPU exposure to broader AI-infrastructure stacks. Winners: GOOGL (cloud + TPUs) and MRVL (HBM, networking, storage) capture incremental wallet share from hyperscalers; marginal losers are pure-play GPU/accelerator suppliers if hyperscalers internalize or diversify suppliers. The McKinsey $7T AI infra estimate over five years implies multi-year demand but distribution of revenue will shift toward memory, interconnect, and custom silicon.

Risk assessment: Key tail risks are an AI-capex slowdown (20–40% downside to consensus over 12–24 months), renewed US–China export controls, and Taiwan/TSMC geopolitical disruptions; these can compress multiples across NVDA/AMD/ MRVL. Immediate (days-weeks): 13F-driven flows and options gamma; short-term (1–6 months): quarterly earnings, design-win announcements and HBM supply data; long-term (1–3+ years): architecture shifts (TPU alternatives, specialized silicon) and hyperscaler vertical integration. Hidden dependencies include MRVL’s revenue cadence tied to hyperscaler procurement windows and TSMC capacity.

Trade implications: Tactical tilt to data-center infra over raw GPU exposure. Favor MRVL and GOOGL for 6–24 month appreciation; size NVDA/AMD exposure as growth-hedged positions rather than core buys. Use options to buy asymmetric upside (MRVL calls) and to protect valuation risk in NVDA/AMD (put spreads). Rebalance sector exposure toward Networking/Storage (+2–5% absolute weight) and trim high-multiple GPU longs by 10–20%.

Contrarian angles: The market may underprice Marvell’s HBM/connectivity leverage — design wins can re-rate shares quickly (30–60% re-rating within 6–12 months). Conversely, consensus may be overestimating TPU stickiness; GOOGL’s cloud contracts could be renegotiated or internalized by customers. Historical parallel: 2016–18 GPU cycles where ecosystem suppliers outperformed GPUs once memory/interconnect shortages emerged. Unintended consequence: a sprint to secure HBM could create temporary supply-driven margin compression for smaller vendors, compressing near-term earnings expectations.

Key Decisions for Investors

• Establish a 1.5–3.0% portfolio long in MRVL for 12–24 months—target +40% upside, stop-loss 25%; consider buying 9–12 month ATM/SLIGHTLY OTM calls (1:1) instead of outright equity if seeking asymmetric payoff.
• Initiate a 2.0–4.0% long in GOOGL for cloud/TPU exposure with a 12-month target return +25% and monitor quarterly cloud revenue growth and Anthropic/OpenAI customer disclosures; trim if YoY cloud growth falls below 20%.
• If NVDA or AMD positions exceed 3% portfolio weight, reduce exposure by 10–15% and buy 3-month put spreads (e.g., 1x 7–10% OTM put / sell 1x 15–20% OTM put) to hedge a 20–35% drawdown risk while retaining upside.
• Run a relative-value pair: long MRVL (1–2%) vs short AMD (1–2%) notional for 6–12 months — thesis: MRVL benefits from memory/interconnect scarcity while AMD remains GPU-valuation sensitive; close if MRVL outperforms by 25% or pair diverges beyond 40% volatility.
• Reduce cyclical/consumer GPU exposure in broader tech sleeve by 2–4% and reallocate to Networking/Storage ETFs or individual names (e.g., MRVL) to capture secular AI-infra wallet-share shift over the next 12–36 months.