Why The VanEck Semiconductor ETF Rallied Almost 50% in 2025

The VanEck Semiconductor ETF (SMH) rallied 48.7% in 2025 as the AI infrastructure boom drove outsized gains across major holdings—Nvidia +38.9% (capped at 20% of the ETF), Micron +240.2% (including dividends) becoming the fourth-largest holding, and TSMC, Broadcom and AMD up 55.9%, 50.7% and 77.3% respectively. Rapid DRAM and NAND price appreciation (DRAM projected +50% this quarter; NAND +30–40% q/q) and constrained new supply until ~2028 underpin near-term earnings upside, but significant downside risks remain tied to OpenAI’s spending trajectory (CEO cited $20B ARR in 2025 and ~$1.4T committed spending over eight years) and the potential for generative AI scaling to slow. Hedge funds should weigh continued demand-driven pricing power and elevated multiples against concentration risks, AI funding dependencies, and supply-cycle uncertainty.

Analysis

Market structure: AI-driven capex is concentrating cashflows into GPUs (NVDA, AMD), foundry services (TSM), networking/IP (AVGO) and memory (MU). Memory’s 240% stock move reflects a DRAM/NAND price shock (DRAM +50% q/q, NAND +30–40% q/q) with constrained new supply until ~2028, so near-term pricing power is strong while OEMs and non-AI legacy semiconductor vendors face margin pressure.

Risk assessment: Key tail risks are (1) OpenAI funding/competitiveness shock (publicized $20B ARR target + $1.4T capex commitments — failure or funding pullback would remove a large demand pillar), (2) algorithmic efficiency reducing chip demand, and (3) geopolitical disruption to TSMC. Immediate: earnings/price prints (days–weeks) will swing sentiment; medium: memory price momentum into 2026; long: capacity additions by 2028 flip surpluses.

Trade implications: Favor concentrated, conviction-weighted longs in MU (memory tightness) and TSM (manufacturing moat) with defined stop-losses, and volatility-aware exposure to AVGO/AMD for AI ASIC/IP exposure. Use pairs to hedge concentrated winner risk (e.g., long AMD vs short NVDA small notional) and buy 9–18 month call spreads or LEAPs rather than outright shares to control downside; size total semis exposure to 8–12% of risk budget.

Contrarian angles: Consensus underprices the risk of demand-side elasticity from model optimization and overprices perpetual capex growth implied by current multiples. Historical memory cycles (2016–18 NAND/DRAM supercycle) show capex chasing rents and later multi-year drawdowns — plan for a potential 40–60% downside in cyclical names if new supply and efficiency trends materialize. Hedge accordingly.

Key Decisions for Investors

• Establish a 2–3% portfolio long position in MU (Micron) now to capture DRAM/NAND tightness; target +50% upside in 6–12 months, place a stop-loss at -25% and trim at +60% or if DRAM q/q spot prices decelerate two quarters in a row.
• Initiate a 1.5–2% long position in TSM (Taiwan Semiconductor) via Jan 2028 LEAP call spreads (buy LEAP, sell nearer-term call) to express structural foundry pricing power; horizon 18–36 months, take profits on a 30% move or if Taiwan geopolitical risk premium increases >15bps in 10-year KOSPI/TW spread.
• Implement a relative trade: long 1.5% AMD (exposure to second-source AI GPUs) and short 1.0% NVDA to neutralize market beta and monetize valuation dispersion; hold through FY2026, rebalance if NVDA underperforms by >15% (close short) or AMD misses guide.
• Buy a directional hedge: allocate 0.5–1.0% of portfolio to SMH/sector 9–12 month 20–30% OTM puts (or buy-put spreads) to protect against an AI-spending shock tied to negative OpenAI funding/ARR headlines or two consecutive quarters of memory price declines.
• Reduce cyclical consumer/PC chipset exposure by 50% (shift 3–5% portfolio weight) into semis capex beneficiaries; redeploy into AVGO (1% tactical) for IP revenue upside and maintain cash buffer for a potential 2028 oversupply buying opportunity.