Billionaire Stanley Druckenmiller Sold Broadcom and Bought These Artificial Intelligence (AI) Giants Instead

Druckenmiller rebalanced his AI exposure by fully exiting Broadcom and adding sizeable stakes in Amazon and Alphabet, reflecting a shift toward hyperscalers taking more control of their custom AI chips. The article argues both Amazon (20.5x forward P/E) and Alphabet (16.4x forward P/E) trade at historically low multiples despite heavy AI data-center capex. However, it flags near-term free-cash-flow strain—Amazon burned $7.6B over the last four quarters and Alphabet posted -$5.9B FCF last quarter—though both have large contracted backlogs ($496B for Amazon, $514B for Alphabet). Overall, the piece frames the setup as a contrarian valuation opportunity with long-run utilization and returns potential.

Analysis

Near term, the market should keep rewarding the platform owners over the component vendors because AI economics are shifting from selling the fastest chip to controlling utilization, routing, and procurement inside the data center. That favors AMZN and GOOG/GOOGL: custom silicon can lower inference cost and improve customer stickiness even while reported free cash flow stays under pressure for a few quarters. It is a relative headwind for AVGO, where bargaining power can erode if hyperscalers increasingly dual-source and treat merchant silicon as a fallback rather than a primary dependency.

The bigger issue is not whether capex is high, but who captures the operating leverage when that capex is deployed. If the new builds ramp as planned, AMZN and GOOG can convert backlog into higher-margin cloud revenue over 6-18 months, while the semiconductor group risks multiple compression if investors start viewing AI spend as a low-return utility buildout rather than a scarcity-driven supercycle. The key falsifier is any sign that capex is being cut or deferred because utilization and monetization are lagging; that would hit the hyperscalers first and also drag the whole AI chain lower.

Contrarian take: the consensus is too binary on chip displacement. Self-designed silicon does not eliminate the supply chain; it shifts value toward networking, memory, power, packaging, and systems integration, so a clean short on semis is too crowded and probably wrong. The more durable trade is that hyperscalers gain negotiating leverage, not that NVDA or AVGO lose all relevance, and that leverage should show up first in pricing discipline and supplier mix before it shows up in revenue declines.

Key Decisions for Investors

• Long AMZN and GOOG/GOOGL on pullbacks into earnings: 6-18 month thesis on higher cloud utilization and better control of AI economics; downside is a continued FCF trough, but upside is multiple re-rating if capex translates into faster revenue growth.
• Pair trade: long AMZN / GOOG vs short AVGO for the next 1-3 months to express the shift in bargaining power from chip suppliers to hyperscalers; stop if Broadcom commentary shows accelerating custom-silicon bookings or pricing remains firm.
• Avoid chasing a broad short NVDA basket here; the article is more about mix shift than demand destruction. If anything, wait for evidence of AI capex deceleration or a guide-down in hyperscaler spending before shorting the complex.
• Watch MRVL as a secondary beneficiary/loser depending on inference mix: if it wins more specialized TPU-related sockets, it can outperform on beta, but any sign of hyperscaler insourcing should cap the upside. Treat as a tactical trading vehicle, not a core long.
• Set a catalyst alert on the next cloud capex updates: if AMZN or GOOG reduces 2026 spend growth while backlog remains high, the thesis breaks and the preferred trade is to fade the entire AI infrastructure basket.