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

Appaloosa's David Tepper held roughly 40% of his Q2 portfolio in Amazon, Micron and TSMC, with Amazon comprising more than 15%, Micron over 14% and TSMC over 10%. The article cites Amazon's $496 billion AWS backlog and projected $1 trillion cloud-revenue opportunity, while highlighting tight HBM, DRAM and NAND supply as support for Micron's memory-cycle margins; Micron trades below 6.5x forward earnings. TSMC is positioned as a key AI-infrastructure beneficiary due to its leading advanced-node manufacturing scale and pricing power, and Tepper increased his holding by about 24% in Q2.

Analysis

The relevant read-through is not Tepper’s positioning but the emerging AI capex hierarchy: TSM monetizes nearly every accelerator design, while AMZN must convert infrastructure spend into sustained utilization and MU remains exposed to the most cyclical link in the chain. TSM’s advanced-node and CoWoS packaging constraints create pricing power through the next 12-18 months; this favors TSM over single-customer GPU suppliers if hyperscalers continue diversifying into custom ASICs. A second-order beneficiary is ASML, where EUV allocation remains a bottleneck for both leading-edge logic and advanced memory, though order timing can be lumpy.

MU’s bullish setup depends less on HBM revenue than on disciplined industry supply. If Korean peers prioritize HBM capacity, conventional DRAM tightness can lift MU’s blended margins disproportionately over the next 1-3 quarters; however, the low earnings multiple is not independently persuasive because memory-cycle peak earnings routinely screen optically cheap. Watch quarterly bit-supply growth, inventory days, and contract-price trends rather than headline HBM demand. A faster-than-expected Samsung capacity ramp or weaker China handset/server demand would compress the thesis quickly.

AMZN is the least clean AI expression: AWS backlog is not equivalent to recognized, high-margin revenue, and incremental depreciation arrives before customer workloads fully ramp. The near-term stock catalyst is evidence that AWS growth accelerates without another step-up in capex intensity, alongside retail-margin resilience; over 6-18 months, proprietary silicon could improve cloud gross margin but may also pressure NVDA’s hyperscaler wallet share. Consensus may be underweight the risk that enterprise AI pilots remain compute-intensive but low-revenue, delaying cloud ROI.

The article is routine, bullish commentary with limited standalone information value. Avoid chasing any opening move; use upcoming earnings, capex guidance and semiconductor supply disclosures as confirmation points.

Key Decisions for Investors

• Prefer long TSM / short NVDA as a 6-12 month relative-value expression of ASIC diversification and foundry bottleneck economics; size modestly because a renewed CUDA-led GPU demand spike is the principal risk. Exit if TSM’s advanced-node utilization or CoWoS expansion commentary weakens.
• Keep MU on a 1-3 month watch rather than initiate solely on valuation. Go long only if next earnings confirms improving DRAM/NAND pricing and inventory discipline; invalidate on material bit-supply expansion from Samsung or SK Hynix, or declining contract prices.
• For AMZN, wait for AWS revenue growth to outpace incremental capex/depreciation commentary at the next results. If confirmed, long AMZN versus short ORCL offers cleaner exposure to cloud-scale economics; risk is Oracle’s AI cloud backlog converting faster than expected.
• Monitor ASML as a second-order beneficiary of persistent EUV scarcity; add on order-book weakness only if management maintains 2027-2028 demand visibility. The key downside trigger is broad leading-edge fab capex deferral, not an isolated memory-price pullback.