Billionaire Stanley Druckenmiller Sells Sandisk and Buys an Unstoppable AI Energy Stock Up Over 800% Since Its IPO

Stanley Druckenmiller reportedly sold 166,235 Sandisk shares after a one-quarter hold and rotated into Bloom Energy, framing power rather than memory as the next AI bottleneck. Sandisk has surged more than 400% over the broader period, while Bloom has risen over 800% since its 2018 IPO as investors focus on scalable fuel-cell power for AI data centers. The piece argues that AI infrastructure spending is shifting toward electricity and behind-the-meter solutions, with hyperscaler capex projected at up to $720 billion in 2026.

Analysis

The important read-through is that the AI infrastructure trade is moving from the obvious layer to the constraint layer. Memory/storage beneficiaries can keep rallying on sentiment and backlog, but once pricing power becomes visibly momentum-driven, the market starts discounting a later-cycle setup; that’s exactly when capital rotates to the scarcest input, which is power. BE screens as a second-order winner because it monetizes the bottleneck that is hardest to solve with software, capex efficiency, or chip substitution.

The knock-on effects matter more than the headline trade. If behind-the-meter generation becomes the fastest path to bringing AI campuses online, then grid-centric developers face a longer monetization lag, while firms that can package land, power, and interconnect solutions should gain negotiating leverage. ORCL and EQIX benefit if they can de-risk delivery schedules for tenants; the upside is less about electricity sales and more about winning scarce data-center capacity and improving pre-leasing conversion.

The main contrarian risk is that the market may be front-running an adoption curve that is still pilot-heavy. Fuel-cell economics are highly sensitive to gas prices, utilization, maintenance assumptions, and whether regulators eventually streamline interconnection enough to make the bottleneck less acute over 6-18 months. If hyperscalers are over-ordering capacity today, the next leg could be less about new buildouts and more about digestion, which would hit the most expensive multiple names first.

For SNDK, the issue is not demand collapse but narrative saturation: once a stock becomes the consensus AI memory proxy, incremental good news stops expanding valuation. That makes the risk/reward less attractive than a supplier closer to the power bottleneck, especially if the AI capex cycle shifts from training to deployment, where steady power and uptime matter more than raw storage throughput.

Key Decisions for Investors

• Long BE vs short SNDK as a 3-6 month thematic rotation pair: BE has direct exposure to the scarcest AI bottleneck, while SNDK is more exposed to cyclically priced enthusiasm; target 20-30% relative outperformance if power scarcity remains the dominant narrative.
• Add ORCL and EQIX on weakness over the next 2-4 weeks: they should benefit from any acceleration in behind-the-meter and campus-level power solutions; risk/reward improves if data-center permitting stays tight and pre-leasing stays elevated.
• Buy BE call spreads 6-9 months out rather than outright stock: this captures upside from new contract announcements while limiting multiple-risk if pilot deployments take longer to convert into backlog.
• Trim or hedge SNDK strength after outsized runs: use a collar or partial short-dated call overwrite into any post-earnings bounce, since the stock is vulnerable to valuation compression if memory pricing normalizes.
• Monitor GPU capex and utility/interconnect headlines weekly: a sudden easing in grid queues or permitting would be the main catalyst to take profits in BE and rotate back toward the memory complex.