Billionaire Philippe Laffont Trimmed 8 of His Fund's Top 10 Positions Last Quarter. Here's What the Other 2 Have in Common.

Coatue Management trimmed eight of its 10 largest positions last quarter while adding only Taiwan Semiconductor Manufacturing and Lam Research, signaling a shift further up the AI infrastructure supply chain. The piece argues TSMC and Lam Research still look attractive on fundamentals, with TSMC trading at about 25x forward earnings versus Lam Research at roughly 48x after a nearly 60% YTD rally. It also notes Coatue sold stakes in hyperscalers and chipmakers such as Nvidia and Broadcom, reflecting a more selective, macro-aware AI portfolio stance.

Analysis

The market is still misreading where AI economics accrue. The highest-quality setup is not the GPU vendor or the hyperscaler with the loudest capex headline, but the bottleneck providers that sit closest to manufacturing capacity and process control; that is where pricing power is most durable when demand is supply-constrained rather than demand-constrained. If AI infrastructure spending keeps compounding, the second-order effect is margin transfer away from model deployers and toward the equipment/wafer ecosystem, which should keep multiple dispersion wide for at least 2-4 quarters.

The key risk is that this trade has become crowded at the “picks-and-shovels” end just as investors are rotating out of the obvious AI leaders. That creates a near-term setup where strong fundamentals may not translate into immediate upside if positioning is already extended, especially in names with the sharpest year-to-date moves. The more important tell is whether hyperscaler capex growth remains linear or inflects down; if budgets decelerate even modestly, the whole upstream thesis compresses quickly because these businesses are being priced on sustained scarcity, not normalized demand.

A contrarian read: the market is too focused on winner/loser framing inside semis and not enough on duration. If AI infrastructure remains a multi-year buildout, valuation matters less for the first derivative of earnings than for the persistence of free cash flow growth and return on invested capital; that favors TSM and ASML-type franchises over economically later-cycle names with more reflexive multiples. The selloff in downstream exposure may therefore be only partially right, but the more interesting opportunity is to fade overowned platforms while owning the industrialized capacity that converts AI spend into actual shipped units.

Near term, this looks like a relative-value market rather than an outright beta trade: the winners should outperform on revisions, not just momentum, while the former leaders likely need a digestion period. A decisive reversal would require hyperscaler capex discipline, a meaningful China/export shock to equipment demand, or evidence that AI unit economics are not improving fast enough to justify continued infrastructure spend.

Key Decisions for Investors

• Long TSM / short NVDA, 3-6 months: express the view that upstream manufacturing scarcity captures more durable pricing power than downstream chip monetization; target relative outperformance if capex stays elevated and hyperscaler spend remains sticky.
• Long ASML / short AVGO, 3-6 months: pair the most leverageable equipment bottleneck against a higher-multiple chip exposure that is more vulnerable to any pause in AI spend cadence; use on pullbacks after sector strength.
• Add LRCX only on weakness, not strength: the business remains well positioned, but the move has likely outrun near-term revisions; reward/risk improves if the stock de-rates 10-15% while fundamentals stay intact.
• Take partial profits in MSFT/META/AMZN if the group rallies on capex enthusiasm: the market is increasingly treating AI spend as a margin drag, so upside may be capped until monetization evidence improves.
• Consider a basket long TSM/ASML/MU vs short NVDA/AVGO for a 6-9 month relative-value trade: best risk/reward if the market continues to reward supply-chain leverage over platform ownership.