David Tepper's Appaloosa Management has concentrated $3.1 billion — 40% of its $7.7 billion equity book — into three positions: Amazon, Micron Technology, and Taiwan Semiconductor Manufacturing. The 13F filed last week shows a portfolio structure that bets specifically on AI infrastructure margin expansion, not the application layer.
The filing shows Tepper sold 41% of his Micron stake during the quarter that saw the stock climb 242%. Micron remains his second-largest holding. He trimmed into strength, kept the core, and rebalanced without abandoning thesis. The position still represents roughly 13% of disclosed assets. Amazon holds the top slot. Taiwan Semi rounds the triad. Together, the three names account for a concentration level Appaloosa has not printed since the 2020 FAANG run.
The trade is memory pricing power, hyperscaler capex visibility, and foundry margin leverage — not software multiples. Micron sells HBM3E chips that sit inside Nvidia and AMD accelerators. Taiwan Semi fabricates the chips. Amazon buys the infrastructure and runs the largest cloud margin pool in the West. Tepper is positioned in the suppliers and the largest buyer, not the model builders. The portfolio architecture isolates margin from the binary risk of which LLM wins.
The Micron trim matters because it was disciplined. Selling 41% into a 242% move while maintaining it as a top-three position signals rebalancing, not exit. Tepper took profits, brought the weight back to portfolio limits, and kept enough to stay leveraged to the next HBM cycle. The filing shows no new entry into application-layer AI. No Palantir. No Snowflake. No C3.ai. The capital stayed in picks and shovels.
Operators and allocators should track three events. First, Micron reports March earnings with updated HBM shipment forecasts — current Street estimates assume 30% of DRAM revenue shifts to HBM by fiscal 2025. Second, Taiwan Semi's April earnings call will quantify N3E node utilization tied to Nvidia's Blackwell ramp. Third, Amazon's capex guide for the back half of 2025 — consensus sits near $85 billion for the full year — will set the upper boundary for hyperscaler build-out. All three data points arrive within sixty days.
Appaloosa now holds a portfolio that moves with three earnings calls, not thirty. The concentration is the thesis.