This Under-the-Radar Top Tech Investor Has 40% of His Portfolio in These 4 AI Stocks

The article is bullish on TSMC, Nvidia, Broadcom, and AMD as key semiconductor winners from AI infrastructure, inference, and agentic AI demand. TSMC is highlighted as a virtual monopoly in advanced chip manufacturing with 14.4% portfolio weight, while Nvidia (8.9%), Broadcom (8.7%), and AMD (8.4%) are cited for strong positioning in GPUs, custom AI chips, and data center CPUs. Broadcom’s custom chip revenues are projected to exceed $100B in fiscal 2027, and Citi sees its AI revenue reaching $180B in fiscal 2028.

Analysis

The common thread is not “AI beneficiaries” broadly; it is the tightening concentration of value in the semiconductor stack around manufacturing capacity, custom silicon, and memory bandwidth. That favors the picks-and-shovels layer first: foundry and advanced packaging should keep capturing pricing power even if the eventual winners at the model layer rotate. In other words, the market is still underappreciating that every incremental inference dollar likely requires more heterogeneous compute, more advanced nodes, and more outsourced manufacturing complexity, which extends the runway for TSM and keeps supply chains tight.

The second-order winner is Broadcom, because custom ASIC adoption is a budget-allocation story, not just an AI story. As hyperscalers push beyond generic GPU deployments, they are effectively trading capex concentration for power efficiency and cost control; that creates a multi-year revenue annuity for the few vendors able to co-design and manufacture at scale. The risk is that this becomes a crowded consensus long: if hyperscaler capex growth slows even modestly, the market may compress multiples before fundamentals roll over, especially in names with the most aggressive forward expectations.

AMD sits in a more interesting middle ground. The bull case is not share gains versus Nvidia in training; it is that inference and agentic workloads force a re-rating of memory-centric architectures and CPU attach rates, where AMD’s chiplet strategy matters more than headline GPU benchmarks. The contrarian miss is that the market may be too focused on product cycles and not enough on ecosystem lock-in: Nvidia’s software moat is harder to displace in software-defined workloads than in fixed-function inference, so AMD’s path is likely adoption-led and lumpy over 6-18 months rather than linear.

Key Decisions for Investors

• Long TSM into any 5-10% pullback over the next 1-3 months; best risk/reward in the group because it monetizes industry growth regardless of share shifts and should hold up best if AI capex sentiment cools
• Pair trade: long TSM / short the most crowded high-multiple AI-semicap name over 3-6 months; thesis is that manufacturing bottlenecks and packaging scarcity persist while valuation dispersion widens as earnings estimates get cut unevenly
• Add AVGO on weakness, but size for expectation risk; use a 6-12 month horizon and target upside from custom silicon adoption while acknowledging that any hyperscaler capex pause can compress the multiple quickly
• Buy AMD only as a staged position over 6-18 months; treat it as a relative-value beneficiary of inference/CPU mix shift, not a straight Nvidia replacement, and prefer call spreads to cap downside if execution slips
• For NVDA, favor defined-risk bullish structures rather than outright common stock; the stock can still work over 6-12 months, but the trade is increasingly about maintaining leadership, so upside remains good while drawdown risk on any roadmap hiccup is material