Billionaires David Tepper and Michael Platt Sold Nvidia Shares and Bought This AI Stock That's Climbed 40,000% Since its IPO.

Tepper cut Nvidia by 10% (now 4.6% of his $6.9B portfolio) and increased Micron by 200% (now 6.2%), while Platt slashed Nvidia by 96% (now 0.2% of his $3.3B portfolio) and opened a small Micron position (~0.1%). Micron reported record revenue and expects additional record quarters, with the company positioned to benefit from AI-driven memory demand; the stock trades at ~11x forward earnings versus ~20x+ for many AI peers. These disclosed 13F moves and Micron's results are likely to affect MU and NVDA flows and investor positioning near term, though filings reflect activity from a few months ago.

Analysis

The structural shift from compute-limited models to memory-limited inference is underappreciated by consensus and creates a multi-year demand tail for higher-density DRAM and HBM class products. As large-scale LLM deployments move from research clusters to production inference at scale, per-rack memory content can rise materially (think 2-4x in the first 12–36 months for racks running large-context models), which favors memory suppliers with flexible capacity and pricing power during tight cycles.

Second-order winners are not just DRAM makers: board-level component suppliers (memory packaging, interposer substrates, and CXL-enabled memory pooling technologies) will see step-function revenue growth as customers trade CPU/GPU optimization for memory bandwidth and persistence. Conversely, firms relying on single-sourced HBM or long OEM qualification cycles face share risks — fast-to-market, high-yield fabs will capture the upgrade wave and squeeze incumbents’ mix and margins.

Risks are highly idiosyncratic and timing-sensitive: inventory digestion, model compression breakthroughs, or a rapid move to edge inference could compress ASPs within 1–2 quarters and reverse the rally. Practically, the trade is a time arbitrage on a capital-intensive supply cycle: if capex remains disciplined, price recovery is plausible over 6–18 months; if suppliers flood the market, downside can be >30% over the same window.

Key Decisions for Investors

• Initiate a tactical long in MU (size 2–3% NAV) via a 12-month call spread to capture the DRAM upswing: buy MU 12-month ATM call and sell a further OTM call to finance ~50–60% of the premium. Risk: cap to premium paid (~-100% of option cost); Reward: directional 2–3x exposure to USD DRAM ASP recovery over 6–18 months. Trim into strength at +40–60%.
• Pair trade to express rotation into memory: long MU equity (1.5% NAV) vs short NVDA (1.5% NAV) on a dollar-neutral basis for 3–9 months. Rationale: capture relative rerating if memory content growth outpaces GPU multiple expansion; Risk: NVDA multiple re-expands or NVDA outperforms during GPU-led spending wave — cap pair size and use weekly rebalancing.
• Event hedge / downside protection: buy 3–6 month put protection on MU equal to ~25% notional (cost ~1–2% NAV) ahead of key industry supply/capex updates. This limits tail risk from an ASP shock or sudden inventory destocking that could erase expected cyclical gains within 1–2 quarters.