Billionaire David Tepper Sold Most of Appaloosa's Position in Microsoft and Initiated a New Stake in an Artificial Intelligence (AI) Memory Stock Up Over 4,100% in the Past Year

David Tepper’s Appaloosa sold 82% of its Microsoft stake in Q1 and initiated a new $179 million position in Sandisk, which represented about 3% of the portfolio. The move highlights investor rotation within AI-related equities, with Tepper reducing exposure to Microsoft amid AI and software sector concerns while betting on NAND memory demand as a critical AI supply-chain component. The article is largely commentary rather than fresh company-specific news, so near-term market impact should be limited.

Analysis

The important read-through is not “Tepper likes memory,” but that capital is rotating from platform software into a narrower set of AI infrastructure bottlenecks where pricing can re-rate faster than demand decays. That is a sign the market is increasingly separating AI monetization from AI picks-and-shovels: hyperscaler software names now need proof of incremental ROI, while component vendors tied to constrained supply can benefit from immediate scarcity pricing. In other words, the trade is shifting from narrative beta to physical bottlenecks.

The SNDK move is more interesting than the Microsoft trim because it implies a view that NAND is in a sweeter spot than DRAM or compute over the next 12-24 months. If NAND supply stays tight into 2027, the earnings leverage is huge, but so is the reflexive risk: every cycle of outperformance invites capex, inventory rebuild, and eventual margin compression. The market is likely underestimating how quickly AI storage demand can normalize if enterprise model deployment slows or if hyperscalers optimize data retention more aggressively.

The bearish second-order effect lands on the software stack: if AI agents, copilots, and embedded search become good enough, the weakest moat in productivity software becomes usage, not distribution. That pressure is not immediate, but it creates a longer-duration multiple risk for MSFT if investors conclude AI spend is protecting share rather than expanding returns. The near-term catalyst to watch is whether Copilot seat growth accelerates enough to offset capex skepticism; if not, the stock can stay range-bound even with strong fundamentals.

Consensus may be overpricing the durability of the SNDK scarcity story and underpricing how selectively the AI value chain is now being rewarded. The better expression is not an outright chase into the most extended name, but a relative-value trade around where the market is still mispricing duration versus cyclicality. If the memory upcycle extends, SNDK and peers can outperform sharply; if it rolls over, the drawdown will be faster than in software because the business model has less structural multiple support.

Key Decisions for Investors

• Long SNDK on a pullback rather than strength; target entries after a 10-15% retrace, with a 6-12 month hold. Use a defined-risk call spread if implied volatility stays elevated, because the upside is real but the cycle risk is material.
• Short MSFT versus long SNDK as a relative-value pair for 3-6 months. Thesis: the market is rewarding scarce AI inputs more than broad AI platform exposure; stop if MSFT reaccelerates Copilot monetization or Azure growth inflects meaningfully.
• Reduce exposure to crowded AI software winners and rotate toward infrastructure beneficiaries with tighter supply dynamics. The risk/reward is better where pricing power is tied to inventory scarcity, not just AI brand association.
• For aggressive accounts, buy SNDK upside through a call spread into any post-earnings or market-wide tech selloff. The asymmetric setup is strongest if the market temporarily derates AI capex names while supply remains constrained.
• Avoid chasing NVDA here; use it only as a hedge against an AI demand rebound. The article’s signal is that the next marginal buyer is moving one layer deeper into the stack, not further up the compute ladder.