At Computex, Nvidia and Taiwan’s expanding role in AI infrastructure set to take centre stage

Taiwan’s role in AI infrastructure is taking center stage ahead of Computex, with Nvidia CEO Jensen Huang saying the company could spend as much as $150 billion a year in Taiwan and AMD pledging more than $10 billion for the island’s AI sector. The article highlights strong demand for AI servers and related supply-chain capacity, with Taiwan’s server exports rising to $60 billion last year from $571 million in 2017. Geopolitical risks around Taiwan remain elevated, but business momentum and AI-linked investment are driving a constructive near-term backdrop for the sector.

Analysis

The setup is less about a one-off AI rally and more about a capital-spending ratchet that widens the moat of the Taiwan hardware stack. NVDA is the clearest beneficiary, but the second-order winner is TSM: the more the ecosystem moves from chip design to full-stack deployment, the more bottlenecks migrate into advanced packaging, substrate capacity, and high-end manufacturing relationships that are structurally sticky. AMD benefits too, but with a lag because its upside depends on proving it can capture incremental board-level and datacenter share rather than just riding the AI demand wave.

The underappreciated read-through is for the “picks-and-shovels” layer: server ODMs, power-management, and network/interconnect suppliers should see the cleanest near-term demand surprise because AI cluster builds are constrained by power, cooling, and integration, not just silicon. That makes TSM/AI-adjacent Taiwan supply chain names the higher-quality expression versus pure-beta semiconductor longs. QCOM, ARM, and NXPI are more mixed: they participate if the show shifts sentiment toward edge AI, robotics, and client-device inference, but they are not the first-order beneficiaries of hyperscaler capex and could lag if compute-heavy AI remains the dominant narrative.

The key risk is that the market is extrapolating enthusiasm into 2026 capacity too aggressively. If order visibility slips, or if memory/power constraints start forcing customers to delay rack deployments, the current move can fade quickly over 1-3 months even if the long-term AI thesis stays intact. Geopolitical noise around Taiwan is a volatility amplifier, but not necessarily a trend-breaker unless it translates into export controls, shipping disruption, or a sudden repricing of supply-chain insurance and delivery schedules.

Key Decisions for Investors

• Long TSM vs short a basket of higher-beta AI semis (e.g., NVDA/AMD relative underweight) for the next 4-8 weeks; thesis is that infrastructure scarcity and packaging leverage will out-earn headline AI enthusiasm.
• Buy NVDA on any post-Comptx pullback only if management commentary confirms 2026 supply expansion; otherwise prefer a call spread rather than outright stock to limit downside if expectations are already crowded.
• Initiate a tactical long in AMD for 2-3 months into AI-capacity commentary, but size smaller than NVDA/TSM; risk/reward improves only if partner-investment and capacity language translates into booked demand.
• Pair long TSM / short QCOM or ARM into the event if the market stays focused on datacenter AI rather than edge inference; both names have lower direct leverage to the capex wave and can underperform if sentiment rotates back to infrastructure.
• If memory and server inventory commentary worsens, hedge semiconductor exposure via short-term put spreads on SMH or SOXX for 1-2 months; this is the cleanest way to express a reversal in AI capex enthusiasm.