Tepper Stakes $3.1B on Amazon, Micron, TSMC in Sharpest AI Concentration Yet

David Tepper's Appaloosa Management disclosed a $7.7 billion long equity portfolio in its latest 13F filing, with $3.1 billion—roughly 40%—locked into three positions: Amazon, Micron Technology, and Taiwan Semiconductor Manufacturing. The allocation represents the fund's tightest concentration bet in at least four quarters, and the names share a single thesis: ownership of the physical layer beneath generative AI workloads.

Amazon holds the largest single position at approximately $1.3 billion, justified by AWS's rising capital expenditure on custom silicon and datacenter buildout—Amazon guided to $100 billion in capex over the next two years during its February earnings call. Micron Technology, weighted near $950 million, supplies high-bandwidth memory (HBM3E) for Nvidia's H200 and Blackwell GPU clusters, where lead times remain at 26 weeks and ASPs are running 35% above DRAM commodity pricing. Taiwan Semiconductor, at roughly $850 million, fabricates the logic dies for every frontier AI accelerator except Intel's Gaudi line. TSMC's Arizona Fab 21 begins 3-nanometer production in Q2 2025, and the company has locked $20 billion in U.S. CHIPS Act subsidies—Tepper is betting the subsidy flow stabilizes margin even if hyperscaler orders moderate.

The concentration matters because Tepper rarely clusters above 30% in fewer than five names. His prior playbook—visible in 2021 and 2022 filings—favored diversified tech exposure across fifteen to twenty positions, with consumer discretionary and financials providing portfolio ballast. This shift into semiconductor infrastructure suggests Appaloosa's research desk believes AI capex cycles will persist through at least 2026, despite Azure and GCP both signaling deceleration in new datacenter leases last quarter. The bet also isolates Tepper from application-layer risk: he holds no meaningful stakes in OpenAI-adjacent software companies or enterprise AI resellers, where gross margins remain under pressure and customer acquisition costs are climbing. Instead, he owns the picks-and-shovels layer where volume contracts are already signed and where hyperscalers cannot easily substitute suppliers. Micron and TSMC both trade at 18x forward earnings, well below the 28x carried by the Magnificent Seven median, yet their revenue guidance for fiscal 2025 implies 22% growth—a 4-point spread that Tepper appears to view as mispriced insurance.

Allocators should monitor three follow-on signals over the next sixty days. First, Micron's May earnings call will disclose HBM3E shipment volumes for Q3 fiscal 2025—any sequential decline below 15% quarter-over-quarter growth would pressure the memory supply thesis and likely trigger position trimming. Second, Amazon's April capex update, expected during its Q1 release, will clarify whether AWS maintains its $50 billion annual run rate or scales back in response to slower enterprise AI adoption. Third, TSMC's April technology symposium in San Jose will detail 2-nanometer yield rates and customer commitments—any delay beyond the planned H2 2025 ramp would ripple through Apple and Nvidia's product roadmaps. Tepper's 13F snapshot reflects positions as of December 31, 2024, meaning he has held through January's semiconductor selloff when the SOX index dropped 9%—suggesting conviction, not momentum chasing.

The filing also reveals what Tepper exited: he liquidated a $430 million position in Alibaba and trimmed Meta by $290 million, reallocating proceeds into the semiconductor trio. That rotation—away from Chinese e-commerce regulatory risk and social platform margin compression, toward U.S.-subsidized fabrication and memory oligopoly—tells allocators that Appaloosa is trading geopolitical and competitive moats, not growth narratives. TSMC's Phoenix facility begins commercial wafer shipments in eleven weeks.