Billionaire David Tepper Has 40% of His Fund in Just 3 AI Stocks. Are They Still Buys?

David Tepper held roughly 40% of Appaloosa Management's Q2 portfolio in Amazon, Micron and TSMC, with Amazon exceeding 15%, Micron over 14% and TSMC over 10%; he increased TSMC by about 24%. The article highlights Amazon's $496B AWS backlog and potential $1T cloud-revenue opportunity, Micron's sub-6.5x forward P/E amid a prolonged AI-driven memory upcycle, and TSMC's advanced-foundry pricing power as key AI infrastructure investment opportunities. The outlook is bullish on sustained demand for cloud capacity, HBM/DRAM memory and leading-edge chip fabrication, though it represents investment commentary rather than a new company disclosure.

Analysis

This is low-information, promotional flow rather than a new fundamental catalyst; the disclosed portfolio positioning is stale and should not be traded as a standalone signal. The investable implication is the relative AI-capex stack: AMZN monetizes capacity with customer contracts, TSM captures leading-edge wafer and packaging scarcity, while MU remains the highest-beta expression of memory pricing. In a modest AI-spend deceleration, MU's earnings revisions and multiple will react first; AMZN's advertising/retail profit pool provides the greatest downside offset.

The underappreciated bottleneck is not simply leading-edge compute but advanced packaging, EUV tool allocation, and power availability at hyperscale campuses. This favors TSM's pricing durability and potentially benefits ASML and packaging suppliers more directly than a broad semiconductor basket; it also limits the likelihood that custom ASICs rapidly displace NVDA, since incremental accelerator supply still consumes the same constrained manufacturing ecosystem. Conversely, a shift from training clusters toward inference-efficient models could reduce memory-content intensity sooner than it reduces cloud demand, challenging the strongest version of the MU thesis over 6-18 months.

Near term, focus on earnings revisions and capex commitments rather than backward-looking HBM commentary. Over the next 1-3 months, AWS growth, AMZN capex guidance, TSM node utilization/CoWoS commentary, and MU's DRAM contract-price outlook are the relevant checkpoints. The consensus risk is that AI infrastructure returns remain adequate at the platform level but unevenly distributed: cloud operators can absorb lower hardware ROI through broader product monetization, whereas memory suppliers cannot absorb a price reset without sharp operating leverage reversal.

Key Decisions for Investors

• Prefer long AMZN / short MU as a 3-6 month relative-value expression of AI monetization versus memory-cycle beta. Size for a 10-15% MU drawdown on a DRAM-price inflection; exit if MU raises forward gross-margin guidance while AMZN AWS growth decelerates materially.
• Accumulate TSM on semiconductor-led pullbacks for a 6-18 month horizon; use a stop/reassessment trigger if advanced-node utilization or advanced-packaging capacity commentary weakens for two consecutive reporting periods. TSM offers cleaner scarcity economics than NVDA at a potentially less crowded point in the AI stack.
• Do not chase MU solely on a low forward P/E: establish an alert around monthly DRAM/NAND contract pricing and inventory days. A sequential contract-price decline or renewed capacity-expansion plans by Samsung/SK Hynix would falsify the cycle-duration premise and warrants reducing exposure quickly.
• For a more diversified bottleneck basket, pair TSM with ASML rather than broad SOXX exposure over 6-12 months. The trade is vulnerable to a China-related export-control escalation or Taiwan risk premium shock; hedge that tail with defined-risk semiconductor index puts rather than reducing the core position prematurely.