This Fund Manager Has a Brilliant Strategy for Investing in AI Stocks

Atreides CIO Gavin Baker argues AI stocks are 'cross-sectionally inefficient,' with valuations too cheap in names like Micron, Sandisk, and Nvidia relative to expensive optical/power/cooling stocks such as Lumentum and Coherent. He suggests the AI trade may be one cycle, meaning cheaper names should outperform over time if the sector continues higher, while pricey multiple-expanded stocks could lag if the boom fades. The piece is commentary rather than a new company-specific catalyst, so near-term market impact is limited.

Analysis

The key takeaway is not that AI is overvalued or undervalued, but that the market is mispricing the dispersion inside the basket. If the same capex cycle is truly driving memory, compute, power, and optics, then the winners are being priced as if they deserve scarcity rents while the cheapest cyclicals are being treated as if the boom is temporary. That creates a second-order opportunity: the most obvious AI beneficiaries may be the laggards with operating leverage, not the highest-multiple names already marked to perfection.

The real fragility is in duration. Optical and power-adjacent stocks are being valued on a narrative of persistent bottlenecks, but bottlenecks invite supply response faster than secular demand stories do. If lead times normalize over the next 2-4 quarters, high-multiple hardware names can de-rate hard even if AI demand stays healthy, because their valuation is now front-running scarcity rather than earnings.

Memory and NVDA look more interesting on a risk-adjusted basis because they have cleaner paths to fundamental growth without requiring further multiple expansion. MU in particular offers asymmetric upside if AI-driven bit demand sustains through the next pricing cycle, while downside is buffered by already-lowered expectations. The contrarian miss in the market is that a single AI cycle does not guarantee a single winner set; it may instead produce a rotating trade where the cheapest names compound from earnings and the expensive names give back through multiple compression.

Key Decisions for Investors

• Long MU vs short COHR/LITE in a 3-6 month pair trade: structure as a basket hedge for AI exposure while expressing mean reversion in valuation dispersion; target 20-30% relative outperformance if multiples normalize.
• Initiate a tactical long NVDA on pullbacks for 1-3 months: use it as the highest-quality exposure to AI capex with less balance-sheet and margin risk than the optics group; stop if forward multiple expands materially without estimate revisions.
• Avoid adding to LITE and COHR after the recent rerating; if already long, trim 25-50% and re-enter only on a 15-20% drawdown or when sell-side revisions catch up to price.
• Buy upside calls on MU into the next earnings window: the setup favors an earnings-driven re-rating if pricing stays firm; risk/reward is better than chasing the high-multiple optical names.
• For hedged books, pair long semis/AI infra with a short basket of the most multiple-expensive optical/power names to isolate valuation compression risk while keeping secular AI exposure.