EXCLUSIVE: Nvidia CEO Jensen Huang Is Flying To Taiwan Nearly Every Month Asking For More GPUs – The Answer Remains Not Yet

Nvidia CEO Jensen Huang said customers may spend $3 trillion to $4 trillion on AI infrastructure by the end of the decade, but former Goldman Sachs executive Michael Parekh argued that TSMC is the key bottleneck and may limit upside if capacity does not expand. Parekh said Huang is flying to Taiwan nearly every month to request more production, while TSMC builds leading-edge fabs over 3-5 years at a cost of tens of billions of dollars. The article is primarily a supply-chain and capacity discussion, with modest sentiment implications for NVDA and TSMC rather than a direct earnings update.

Analysis

The key market implication is not simply that AI demand is strong, but that supply is becoming the binding constraint on monetization. When one foundry effectively ration-controls leading-edge capacity, every incremental GPU dollar gets translated into a multi-year backlog rather than near-term revenue acceleration, which can cap upside for designers while extending the durability of pricing power across the whole AI stack.

Second-order winners are likely to be the adjacent bottlenecks, not the headline semiconductor names: advanced packaging, HBM memory, substrate vendors, and equipment suppliers with exposure to new fab builds. If TSMC refuses to overbuild, the market may rotate from “GPU volume expansion” to “content per GPU” beneficiaries, because customers will spend more per deployed accelerator to squeeze utilization out of scarce silicon.

For NVDA, this is a double-edged setup. Near term, scarcity supports gross margin and keeps the narrative intact, but over 6-18 months it raises the risk that hyperscalers diversify architectures faster than expected, especially if they cannot get enough units to match AI capex plans. The real fragility is not demand collapse; it is deferred demand leaking into custom silicon, smaller model inference optimization, and non-Nvidia compute stacks.

The contrarian read is that the market may be underestimating how much of the AI buildout is now a supply-chain finance problem rather than a demand problem. If TSMC eventually greenlights more capacity, the earnings impulse could broaden to foundry, equipment, and packaging names; if not, the AI capex market may still grow, but with lower throughput and more volatility in who captures the economics.

Key Decisions for Investors

• Long TSM vs short NVDA on a 3-6 month horizon: own the capacity gatekeeper while hedging the designer whose shipment growth is increasingly constrained; favorable if the market keeps re-rating scarcity economics over unit growth.
• Add exposure to semiconductor equipment and advanced packaging beneficiaries on weakness over the next 1-3 months: the best risk/reward is in names levered to fab expansion and bottleneck relief, not pure GPU demand beta.
• Use NVDA call spreads rather than outright longs for the next earnings cycle: the implied upside from AI demand remains real, but supply rationing makes multiple expansion more volatile and caps near-term shipment surprise.
• Pair long TSM with short a basket of smaller AI compute beneficiaries that depend on abundant accelerator supply, as a hedge against demand leakage into custom silicon and diversified architectures over 6-18 months.