Billionaire Stanley Druckenmiller Sells Micron and Is Piling Into This Other Unstoppable Artificial Intelligence (AI) Chip Stock Instead

Duquesne Family Office (Stanley Druckenmiller) fully exited Micron (MU) after the stock surged 231% in 2026 to a ~$1T market cap, signaling caution that AI memory may be more capacity-cyclical than a long-duration compounder. In the same Q2 13F move, Duquesne initiated a position in Advanced Micro Devices (AMD), where revenue rose 50% YoY to $11.5B and the data center segment more than doubled to $6.7B (~60% of sales), with management guiding continued acceleration into the second half. The rotation and valuation context (AMD ~63x forward P/E vs. semis ~26x) is likely sentiment-relevant but not necessarily market-wide.

Analysis

The real signal is not the exit itself; it is the rotation from a supply-led trade into a share-gain story. Memory has the nastier setup here: once the market believes HBM is structurally scarce, every incremental wafer or packaging fix invites faster capacity response and lower pricing power, so the upside in MU becomes more linear and more fragile. AMD, by contrast, is still in the phase where each hyperscaler qualification can expand the addressable market, which is why its earnings revisions can outrun the multiple for a few quarters even if the stock already discounts a lot of good news.

Second-order, this is a relative-value warning for the AI semiconductor basket: if capital rotates away from MU and into compute beneficiaries, SMH can stay bid even if memory underperforms. But AMD’s premium valuation makes it vulnerable to any slip in data-center ramp, gross margin, or customer concentration; the stock likely trades more like a high-beta guidance vehicle than a durable compounder over the next 6-12 months. NVDA remains the cleaner secular winner, but that also means AMD’s upside is more dependent on execution and capex diversification than on AI demand itself.

Contrarian view: the market may be underestimating how long HBM scarcity can persist because advanced packaging, not just memory dies, is the bottleneck. If that is right, MU is not a sell-the-story short; it is a hold until pricing data turns. The thesis is falsified if AMD’s data-center growth decelerates below the current run-rate on the next two quarters, or if MU reports another leg of HBM price/margin expansion instead of normalization.

Key Decisions for Investors

• Buy AMD on post-rotation pullbacks over the next 1-3 weeks; target a 15-20% upside if hyperscaler diversification commentary remains intact, but cut exposure if forward data-center growth slows below management’s current trajectory.
• Pair trade: long AMD / short MU for 1-3 months as a relative-value expression on compute share gains versus memory cyclicality; stop out if MU HBM pricing holds firm into the next print.
• Do not short MU outright here; use it as a watchlist name until supply indicators turn, because HBM scarcity could keep margins elevated for another 1-2 quarters.
• If already long semiconductor beta, rotate part of the exposure from MU into NVDA or AMD rather than adding gross; AMD carries higher execution risk but better relative upside if the market keeps rewarding AI compute diversification.