Nvidia’s Huang pushes suppliers on Taiwan visit

Nvidia CEO Jensen Huang pressed major Taiwanese suppliers to ramp production to meet surging AI demand, praising TSMC and saying he will need “a lot of wafers” as Nvidia’s AI-driven growth continues; he also warned of memory supply constraints this year. TSMC last month indicated capital spending could rise as much as 37% to about US$56 billion in 2024 and signaled further significant increases in 2028–29, with Huang forecasting TSMC capacity could more than double over the next decade. The comments underscore a robust demand backdrop for Nvidia and Taiwan semiconductor suppliers, support elevated capex for foundries and equipment vendors, but flag near-term memory tightness as a risk to production schedules.

Analysis

Market structure: Nvidia (NVDA) and TSMC (TSM) are the immediate winners — NVDA from demand pull-through and TSM from multi-year capex tailwinds (TSMC guiding to +37% capex to $56B in 2024 and management hinting >100% capacity growth over a decade). Memory suppliers (Micron/MU, SK Hynix, Samsung) are secondary beneficiaries as DRAM/NAND tightness should push spot and contract prices materially higher in H1–H2 2024. Short-term pricing power rests with foundries and memory; OEMs/consumer electronics face margin pressure and potential inventory discipline.

Risk assessment: Key tail risks are geopolitical escalation around Taiwan, new US/China export controls, and a capex-driven oversupply in 2027–2029 if fabs are overbuilt; any of these could cause >30% drawdowns in TSM/NVDA in stressed scenarios. Time horizons: immediate (days) — volatility spikes and sentiment trades; short-term (weeks–months) — order-book and ASP signals; long-term (years) — structural throughput and gross-margin changes from node mix. Hidden dependencies include OSAT/substrate capacity, power/water limits in Taiwan, and memory supplier cadence; catalysts include NVDA earnings, TSMC quarterly capex updates, and DRAM spot-price prints.

Trade implications: Primary trade is tactical long NVDA (3–6 month directional via call-spreads) and strategic long TSM (12–36 month core) sized to conviction. Pair trades: long TSM / short INTC for 6–12 months to capture foundry share shift; overweight semiconductor equipment names (ASML/AMAT) as indirect plays on sustained capex. Options: sell short-dated premium into post-earnings rallies, buy 3–6 month call spreads on NVDA to limit cost, and buy 9–12 month 10–20% OTM puts on TSM as geopolitical insurance.

Contrarian angles: Consensus underprices memory cyclicity and overprices perpetual multiple expansion for NVDA — a miss in AI enterprise spend could compress NVDA multiples >20% quickly. Conversely, markets may under-appreciate TSMC’s pricing leverage if wafer tightness persists; capex growth could paradoxically increase near-term margins before long-term ASP normalization. Historical parallel: 2000s semiconductor capex cycles show boom → short-term supply tightness → late-cycle oversupply; appropriate hedges and staged entries avoid being caught on the wrong side.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA over the next 5–10 trading days via staggered buys; complement with 3–6 month call spreads (buy calls / sell higher strike calls) to cap premium. Trim on a +25–40% move or if implied volatility compresses below 40%.
• Build a 2–4% core position in TSM for 12–36 months, add on pullbacks >10% from entry; purchase 9–12 month 10–20% OTM puts sized 0.5% of portfolio as geopolitical insurance (protects against >15% downside).
• Initiate a relative-value pair: long TSM (2%) / short INTC (2%) for 6–12 months to express foundry share gains; rebalance or close if the pair moves >15% against you or if TSM issues materially reduced capex guidance.
• Allocate 1–2% to DRAM exposure via MU (Micron) using 3–6 month call spreads to play near-term memory tightness; exit if DRAM spot prices fail to rise >15% within 3 months or MU rallies >40%.