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

The article argues that TSMC, Nvidia, Broadcom, and AMD are the key semiconductor winners from AI growth, with Light Street Capital holding 40% of its portfolio across these four names. TSMC is highlighted as a near-monopoly in advanced chip manufacturing with a 14.4% portfolio weight, while Nvidia (8.9%), Broadcom (8.7%), and AMD (8.4%) are positioned for growth in inference, agentic AI, and custom AI chips. Broadcom's custom chip revenue is projected to exceed $100 billion by fiscal 2027, and Citigroup sees AI revenue reaching $180 billion by fiscal 2028.

Analysis

The common denominator is not “AI exposure” but control points in the AI capex stack. TSM is the toll booth, while NVDA, AVGO, and AMD are competing for different layers of compute value capture; that makes the real question less about whether AI spending grows and more about how quickly hyperscalers shift from merchant GPUs to customized silicon. If that transition accelerates, AVGO and TSM should be more durable beneficiaries than NVDA on a multi-year basis because they monetize design wins and wafer starts across multiple customers, not just one architecture cycle.

Second-order, the article understates the margin and timing asymmetry across the group. NVDA and AMD can outperform on revenue growth, but their upside is more vulnerable to a slower inference ramp, pricing pressure, or a capex digestion quarter; AVGO and TSM have cleaner visibility because custom silicon and leading-edge foundry demand tend to be locked in earlier. The market is likely overpaying for near-term growth optionality in NVDA while underappreciating the structural scarcity of advanced packaging, leading-edge nodes, and custom ASIC expertise.

The key contrarian risk is that “winner-take-most” in AI inference is not guaranteed. If model efficiency improves faster than demand expands, hyperscalers may delay incremental chip purchases and shift toward lower-cost, mixed-vendor deployments, which would compress the upgrade cycle for NVDA and AMD over the next 2-4 quarters. Meanwhile, TSM and AVGO still win in a slower-spending world because they are embedded in the supply chain and custom chip roadmap regardless of which accelerator standard prevails.

Positioning-wise, this setup argues for owning the picks-and-shovels while fading consensus enthusiasm in the most crowded AI beta. The market is treating all four names as one trade, but the earnings durability and drawdown profile are meaningfully different, especially if capital spending growth decelerates from 2026 peaks. The best risk/reward is likely in relative value rather than outright longs.

Key Decisions for Investors

• Long TSM / short NVDA over 3-6 months: express the view that foundry scarcity and packaging bottlenecks will monetize more reliably than GPU multiple expansion; target a 10-15% relative outperformance if AI capex normalizes.
• Add AVGO on pullbacks over 1-2 quarters: custom silicon and networking are the cleanest way to own hyperscaler spend without single-product dependency; favorable if the market starts rewarding recurring design-win visibility over headline growth.
• Reduce or hedge AMD into strength for the next 1-2 earnings cycles: the stock is most exposed to a re-rating gap if inference adoption is slower than expected; use collars or put spreads to protect gains.
• Keep a core TSM position and pair it against a basket of AI hardware beta if volatility rises: TSM’s downside should be lower because it monetizes every architecture outcome, not a single winning design.
• If NVDA rallies into a valuation multiple expansion without upward capex revisions, consider a tactical short via call spreads; risk/reward improves if the market starts pricing in inference margin pressure before revenue inflection fully materializes.