The Billionaire Who Sold Nvidia Too Early Just Bought 196,000 Shares of Broadcom -- Here's the Thesis Behind the Rotation

Stanley Druckenmiller admitted selling Nvidia was a "big mistake" after its share price rose from his $22-$24 split-adjusted entry to about $190 today, versus a roughly $73.50 exit. He re-entered Broadcom in Q1 2026, buying 196,000 shares at an average of $330, though it is unclear whether this is a trade or a long-term position. The article frames Broadcom as an AI inference beneficiary with AI chip sales of $20 billion in fiscal 2025 and guidance for more than $100 billion by fiscal 2027.

Analysis

The market is treating this as a simple “AI winner rotates from GPUs to ASICs” story, but the more important signal is capital concentration moving one layer up the stack. If hyperscalers keep shifting inference workloads to custom silicon, the economic moat migrates from standalone accelerator vendors toward the firms that can monetize both the chip and the attached networking/software layer. That is structurally favorable for AVGO, but it also compresses the addressable pool for second-tier accelerator and inference-adjacent suppliers over the next 12-24 months.

The key second-order effect is margin durability. ASICs can be more efficient at scale, but they also embed customer-specific design risk and shorten product-cycle visibility; that makes reported growth look smoother than the underlying negotiating leverage actually is. If Meta/Google continue adopting custom parts, AVGO gains share, but the customers also gain bargaining power as the design becomes “must-have” infrastructure rather than a discretionary purchase. The setup is bullish on revenue mix, but less clean on terminal margin than the market’s multiple expansion implies.

For NVDA, this is not a near-term collapse thesis. Training remains the premium demand vector, but the market may be overestimating how much of incremental AI capex will stay in general-purpose GPUs once inference scales. The risk is a slower-than-expected re-acceleration in non-training demand, which would show up over quarters, not days; if that happens, leadership breadth in AI hardware narrows and AVGO/GOOGL/META-related capex wins become more durable relative trades. The contrarian miss here is that Broadcom may be priced as a cyclical trade while the inference shift is actually a multi-year platform transition.

Key Decisions for Investors

• Go long AVGO vs short NVDA on a 6-12 month horizon; the pair captures the inference-ASIC migration while reducing beta to broad AI capex sentiment. Use a 1:1 notional initially, add on any NVDA-led bounce above prior highs.
• Buy AVGO on pullbacks, but size it as a staged entry rather than a full conviction core until the next customer/design-cycle commentary confirms durability. Risk/reward is attractive if the stock stays under ~25x forward earnings; de-rate risk rises materially if growth expectations stop compounding.
• Favor META and GOOGL as secondary beneficiaries of custom-silicon adoption versus pure hardware suppliers; both gain from lower inference cost per query. Use them as lower-vol ways to express the AI capex shift over 12-18 months.
• Avoid chasing NVDA on headlines; instead, use strength to sell covered calls or trim into rallies if portfolio exposure is already large. The asymmetry is no longer as attractive if inference capex increasingly migrates away from general-purpose GPUs.
• If you want convexity, consider a medium-dated AVGO call spread financed by selling out-of-the-money NVDA calls. That structure expresses the regime shift with defined downside and avoids paying full beta for the AI theme.