Billionaire Chase Coleman Just Increased His Stake in These 3 Top Semiconductor Stocks. Are They Still Buys?

The article argues that Nvidia, Broadcom, and Taiwan Semiconductor Manufacturing remain attractive AI infrastructure plays, highlighting Chase Coleman’s Q1 additions to all three. Nvidia is framed as the LLM training and broader AI infrastructure leader, Broadcom as a custom ASIC/TPU beneficiary with a potential $100B+ fiscal 2027 opportunity, and TSMC as the manufacturing bottleneck benefiting from rising demand across GPUs, ASICs, and CPUs. The piece is largely opinion-driven rather than newsy, so direct market impact is likely modest.

Analysis

The key second-order effect is not just that AI spending stays strong, but that the mix shifts further away from commoditized accelerator supply and toward the bottlenecks with pricing power: custom silicon design, advanced packaging, HBM-adjacent subsystems, and foundry capacity. That favors AVGO and TSM more durably than NVDA on a 12-24 month view if hyperscalers keep pushing inference economics lower, because every custom chip program still has to flow through a small set of manufacturing constraints. The market is likely still underappreciating how much of the AI capex stack becomes a toll road rather than a product cycle.

NVDA remains the highest-quality platform, but the marginal upside from training dominance is increasingly capped by customer concentration and the inevitable push for workload-specific alternatives. The bigger risk is not share loss in the near term; it is multiple compression if investors begin to treat AI infrastructure as a basket of infrastructure annuity streams rather than a hypergrowth software-like narrative. That said, any cooling in model-training spend would likely hit NVDA first, while AVGO and TSM have better downside insulation because their revenue pools diversify across multiple end-markets and customers.

The contrarian read is that the real beneficiary of AI proliferation may be the manufacturing chokepoint, not the visible brand-name chip designers. TSM is the cleanest way to own that bottleneck, but it also creates a hidden concentration trade: if capacity remains tight, pricing power improves; if customers over-order and then normalize, near-term sentiment can wobble even while long-term units keep growing. The risk window is months, not days — this is a capex cycle story, so the main reversal catalyst would be a delay in hyperscaler deployment budgets or evidence that custom inference chips are taking share faster than expected from general-purpose accelerators.

GOOGL is the sleeper strategic beneficiary because it can arbitrage its own internal TPU stack into cloud margin expansion and use that to pressure peers on price/performance. INTC and NFLX are effectively non-factors here; the former only matters if it can prove manufacturing relevance, and the latter is a distraction relative to the capital intensity shift in AI infrastructure. The broader signal is that investors should stop treating AI as one trade and start separating training, inference, networking, and manufacturing exposure.

Key Decisions for Investors

• Long TSM vs. short a basket of AI software/semicap names over 3-6 months; thesis is that manufacturing bottlenecks monetize better than application-layer enthusiasm. Risk/reward: 1.5-2.0x if AI capex stays elevated, with lower drawdown than single-name accelerator exposure.
• Overweight AVGO on any 5-8% pullback and hold 6-12 months; custom silicon plus networking gives the best blend of secular growth and pricing power. Risk: customer concentration, but upside is still less dependent on one product cycle than NVDA.
• Keep NVDA as a core long but hedge 20-30% of the position with medium-dated calls on a semi ETF or a modest short against AVGO/TSM if the portfolio is too concentrated in AI beta. This captures upside while reducing multiple-compression risk if the market re-rates the space.
• Add GOOGL selectively as a hidden AI infrastructure beneficiary for 6-12 months; it is the cheapest path to monetize TPU economics without paying peak AI narrative multiples. Best expressed as a relative long versus mega-cap peers with less direct infrastructure leverage.
• Avoid chasing INTC on the AI infrastructure theme until there is proof of advanced manufacturing or meaningful foundry traction; current optionality is real but binary. Better used as a small, tactical call option than a directional equity long.