David Tepper Sold 41% of His Micron Shares and It Is Still His Second-Biggest Holding

Appaloosa Management trimmed its Micron stake by 41% in Q2, selling 690,000 shares, but the remaining 975,000 shares were still worth about $1.13B (roughly 15% of the fund), after Micron surged ~242% in the quarter. The article attributes Micron’s earnings surge to AI-driven data center demand, with fiscal Q3 2026 revenue rising to $41.5B (from $9.3B a year earlier) and net income up ~15x to $28.2B, alongside management guiding fiscal Q4 revenue near $50B at ~86% gross margin. Despite the massive profit rebound, the piece flags decelerating growth and ongoing memory-cycle risk, positioning Micron as a “hold” rather than a fresh buy at current prices.

Analysis

Appaloosa’s trim reads less like an abandonment signal than classic cycle management: when a position triples, the marginal dollar at risk matters more than the headline sale. For MU, the market is implicitly asking whether the earnings inflection is still compounding or merely plateauing; that question will drive multiple compression/expansion faster than any single-quarter print. The fact that a sophisticated holder kept a very large residual stake suggests the trade is now about duration of peak margins, not whether the business is good.

The second-order issue is that AI memory demand is still being financed through hyperscaler capex, so MU’s near-term sensitivity is tied to AMZN/NVDA spending cadence, not just DRAM pricing. If cloud buildouts stay intact, the cycle can stay elevated longer than legacy memory investors expect; if capex pauses, the earnings air pocket can arrive quickly because the market is paying for peak-ish cash flows on a low multiple. SNDK looks structurally weaker on that framework: NAND lacks the same AI scarcity premium, so any memory normalization should hit it earlier and harder.

Contrarian view: the consensus may be overfocused on the inevitability of a bust and underfocused on supply discipline plus HBM mix shift, which can stretch the cycle for multiple quarters. But the reverse risk is equally real: once sequential growth slows enough, the market will stop anchoring to absolute earnings and start discounting the next downcycle. The key falsifier is not valuation alone; it is whether next-quarter sequential revenue growth and gross margin stay at extreme levels after the current AI ordering wave normalizes.

Key Decisions for Investors

• Do not chase MU outright here; wait for a post-earnings or post-capex-disappointment entry if the stock de-rates 10-15% while forward guide remains intact. Risk/reward is poor if you are paying for peak earnings without a visible reacceleration signal.
• Relative-value idea: long MU / short SNDK over 1-3 months. Thesis: DRAM/HBM scarcity can remain tight even if broader memory normalizes, while NAND is more exposed to inventory digestion and weaker pricing power.
• For existing MU longs, collar or call-overwrite into the next earnings window. The goal is to harvest premium from elevated implied volatility and protect against a fast de-rating if sequential growth slows.
• Use AMZN and NVDA capex commentary as the leading indicator for MU duration. If either company signals slower data-center spend, reduce MU exposure immediately; that would be the cleanest falsifier for the bull case.
• If you want a higher-conviction bearish expression, prefer a small tactical short in MU only after a confirmed miss in sequential growth or gross margin; otherwise the low multiple can keep shorts painful in the near term.