David Tepper's Appaloosa Management disclosed 40% portfolio concentration across three names in its latest 13F: Amazon, Taiwan Semiconductor, and Micron Technology. The $7.7 billion fund now holds roughly $3.1 billion in those positions, an allocation density unusual even for a manager known for conviction sizing. The filing covers the quarter ending December 31, a period when all three names rallied on enterprise AI capital expenditure guidance.
Tepper sold 41% of his Micron stake during a quarter in which the stock gained 242%. Micron remains Appaloosa's second-largest holding despite the reduction, suggesting the original position was sized at multiple percentage points of the portfolio. The sale came after Micron reported HBM3E memory shipments to Nvidia and guided fiscal 2025 data center revenue above $8 billion, more than double prior-year levels. Tepper's trim follows the pattern of a manager who scales out of momentum after fundamental milestones are priced. Amazon and Taiwan Semiconductor positions were maintained or increased, though exact share counts are not disclosed in the summary data.
The concentration matters because Tepper's Appaloosa has $7.7 billion in disclosed long equity, not the $16 billion peak AUM the fund managed in prior cycles. The current portfolio represents post-redemption capital, and the decision to hold 40% in three correlated infrastructure names indicates Tepper sees the AI build-out as a multi-quarter, not multi-month, capital cycle. Amazon's AWS segment reported $110 billion annualized revenue in Q4 2024, with operating margins above 38% as GPU instance pricing held firm. Taiwan Semiconductor guided March-quarter revenue 10% above consensus, reflecting CoWoS advanced packaging capacity coming online for Nvidia's Blackwell and Google's TPU v6. All three positions share exposure to the same demand vector: hyperscale AI infrastructure spend, which Evercore ISI estimates will exceed $250 billion in 2025 across the top five U.S. cloud providers.
The Micron sale telegraphs Tepper's view that memory pricing has approached mid-cycle equilibrium faster than the market expected six months ago. HBM3E spot prices in Korea fell 8% sequentially in January, the first decline since mid-2023, as Samsung and SK Hynix added capacity ahead of Nvidia's Blackwell ramp. Micron's forward PE sits near 18x, high for a memory cycle three quarters into an upcycle. The retention of a top-three position suggests Tepper expects another leg in 2026 when HBM4 specifications finalize and hyperscalers begin pre-ordering for 2027 deployments. The Amazon and TSM holdings carry less cyclical risk: Amazon's margin structure allows it to absorb slower AI monetization, and TSM's CoWoS monopoly persists until Intel's Foveros packaging scales in late 2026.
Allocators should watch for Appaloosa's Q1 2025 filing in mid-May, which will show whether Tepper adds to Amazon after its April earnings or rotates capital toward software-layer AI plays. The fund has historically moved quickly after thematic exhaustion, exiting energy names within one quarter of crude oil topping in prior cycles. TSM reports April 17 with March-quarter revenue guidance; any CoWoS capacity commentary above 45,000 wafers per month would support Tepper's continued hold. Micron's next earnings on March 19 will clarify whether HBM3E pricing stabilizes or contracts further into spring.
Tepper's $3.1 billion stake is now a public benchmark for how long institutional capital believes the infrastructure phase lasts before the market demands proof of AI application-layer revenue.