Semiconductors (SMH)

SMH is the VanEck Semiconductor ETF, which tracks the MVIS US Listed Semiconductor 25 Index. The fund holds the 25 largest semiconductor names listed on US exchanges, with NVDA at 20-25% of fund weight (sometimes higher) and TSM, AVGO, AMD, ASML, AMAT, LRCX, MU rounding out the top exposures. SMH trades several billion dollars of average daily volume and is the most-traded thematic semis ETF.

Why it matters: SMH is the cleanest pure-play on the AI capex cycle and on global semiconductor demand. The semiconductor industry is mid-cycle in 2026 with AI training infrastructure as the dominant demand driver, while traditional segments (PC, smartphone, auto) are in various recovery phases. SMH leads broader tech (XLK, QQQ) by 1-3 months in cycle inflections because semis are the foundational layer of the AI value chain. SMH-versus-XLK relative performance is the cleanest read on whether AI capex or AI software is the dominant theme.

SMH trades near all-time highs in April 2026 driven by NVDA Blackwell shipments, the upcoming Rubin product cycle, and sustained hyperscaler capex (MSFT + GOOGL + AMZN + META combined ~$300 billion run-rate in 2026). NVDA at roughly 20-25% of SMH weight makes the fund essentially a "NVDA-plus-semi-supply-chain" vehicle.

The risk to SMH is concentration in NVDA: any deceleration in datacenter GPU revenue (current YoY growth running 50-80%) would compress SMH disproportionately versus broader tech. The bull case is sustained AI capex through 2026-2027 plus the inference-workload growth phase that follows training-workload growth. Watch hyperscaler capex guidance, NVDA datacenter revenue, and TSM advanced-node utilization (3nm, 2nm) for the foundational supply-chain signal.

SMH has compounded approximately 17% annualized since 2000 launch, with extreme cyclical volatility. Major drawdowns: -85% in dot-com (2000-2002), -55% in 2008-2009, -45% in 2018 (memory cycle plus crypto-mining wipeout), -38% in 2022 (PC and smartphone bust). The three drivers are AI training capex (NVDA-driven), memory cycles (MU, SK Hynix idiosyncratic), and global PC and smartphone demand (broader semi industry health).

First, hyperscaler capex guidance. Combined MSFT + GOOGL + AMZN + META + ORCL capex above $400 billion for 2026 is the bull threshold; below $300 billion would compress SMH 10-20%.

Second, NVDA Blackwell and Rubin shipment cadence. Any sign of price erosion in datacenter GPU ASPs would compress NVDA gross margin and re-rate the entire SMH constituent set.

Third, TSM 3nm and 2nm utilization. Foundational capacity utilization signals whether the AI capex flow translates into actual production volume.