VanEck Semiconductor ETF
VanEck
Tracks the MarketVector US Listed Semiconductor 25 Index — the 25 largest US-listed chip design, equipment, and foundry names. Cap-weighted and highly concentrated (top 10 ~68%), making it the most NVIDIA-dominant of the big semi ETFs.
Top holdings
Dominant single position; AI-GPU leader drives the fund
World's leading foundry
Memory/HBM beneficiary of the AI build-out
Custom AI silicon and networking
Semiconductor equipment
Data-center CPUs/GPUs
NAND/memory — AI-driven storage demand (post-WD spinoff)
AI-cluster connectivity (active electrical cables / SerDes)
Recent moves
Riding the AI-chip rally hard, with sharp single-day swings (e.g. ~-4% on Jun 26, 2026) amid sector volatility.
Our take
The most concentrated mainstream semi ETF — a ~17-18% NVIDIA weight makes SMH effectively a leveraged bet on the AI-GPU leader, so it outruns peers on the way up and will fall hardest if NVIDIA de-rates. Iconic and liquid, but single-name risk is the trade-off for the upside.
SMH's concentration is a byproduct, not a design choice.
The index picks 25 names, lets market cap sort them, and reviews twice a year — so NVIDIA's weight is whatever the tape made it. An investor buying SMH is not really choosing 'semiconductors'; they are choosing whichever two or three names the cycle most rewarded, at whatever weight, refreshed semi-annually. That has been enormously profitable, and that is a description rather than a criticism. The point marketing will not make is the second one: NVIDIA-maximalism was not the winning trade in 2026. On trailing one-year returns, SOXX — the broader, position-capped peer — was up 113.43% against SMH's 93.09% (stockanalysis.com, 21 Jul 2026). The most concentrated fund lost to the less concentrated one during the biggest AI-chip run on record, because equipment and memory names outran the GPU designer. Concentration is not a synonym for upside; it is a synonym for path-dependence.
- 2026-07July 2026 ran the mechanism backwards: NVDA +0.3% and AVGO +3.1% were the haven while memory cracked (MU -28.7%, SanDisk -46.6%). SMH -17.6% vs SOXX -21.2% — the NVIDIA tilt cushioned rather than cost.
Thesis
SMH is a free-float cap-weighted slice of the 25 largest US-listed semiconductor names (26 fund line items vs 25 index components), reviewed only semi-annually. That last detail is the whole fund: between reviews winners are allowed to run, so NVIDIA sat at 20.97% and the top 10 at 70.92% of assets (stockanalysis.com, 21 Jul 2026). The pitch is "own the chip complex." The reality is a concentrated bet on a few AI-cycle winners, in an index that will not trim them for you.
Tracks the MarketVector US Listed Semiconductor 25 Index — the largest, most liquid US-listed firms deriving 50%+ of revenue from semis or semi equipment, free-float cap-weighted, reviewed semi-annually. ADRs count as US-listed, which is how TSM (9.16%) and ASML (5.22%) enter. 26 holdings, 0.35% net expense ratio, inception 20 Dec 2011 (stockanalysis.com, 21 Jul 2026). No leverage, no daily reset.
Assessment
- Genuine scale and liquidity — $67.42B net assets (stockanalysis.com, 21 Jul 2026) makes it a cheap, tight-spread instrument for a sector view.
- Revenue-purity screen (50%+ from semis/semi equipment) keeps out the conglomerate dilution that creeps into broader tech baskets.
- ADR eligibility captures TSM (9.16%) and ASML (5.22%) — foundry and lithography chokepoints a US-domiciled-only screen would miss.
- 0.35% is competitive with SOXX (0.34%) and XSD (0.34%); fee is not where this fund differentiates or loses.
- 70.92% in ten names with a single 20.97% position (21 Jul 2026). A concentrated portfolio priced and marketed as a sector fund.
- Semi-annual review lets weights drift far between resets, so concentration peaks exactly when it is most dangerous — after a name has already run.
- Cap-weighting embeds momentum: it buys what has appreciated and cannot rebalance away from crowding it structurally creates.
- Correlation inside the basket is high — shared end-demand, capex cycles, Taiwan and export-control exposure. 26 holdings deliver far less than 26 holdings' diversification.
- Provider figures diverge materially — AUM ($67B vs $77-78B) and NVDA weight (20.97% vs ETF.com's ~15%+). Treat any single quoted figure, including the weight the concentration case rests on, as approximate.
Record
The record is cycle-driven, not skill-driven — there is no manager, only an index rule that concentrated into the AI buildout. Trailing 1-year +93.09% and since-inception +28.79% annualized (stockanalysis.com, 21 Jul 2026). Attribution is straightforward: NVIDIA's re-rating plus the memory and equipment complex, amplified by weights that were never trimmed. The instructive detail is the 2026 relative result — SMH trailed SOXX by roughly 20 points over one year (93.09% vs 113.43%, 21 Jul 2026) despite carrying more NVIDIA; an earlier snapshot showed a similar YTD gap (64.11% vs 86.78%, ETF.com through 9 Jun 2026), since which the group has fallen into bear-market territory. When the rally broadened into equipment and memory, the single-name tilt became a headwind. A record built on one name compounding is not evidence the structure is superior; it is evidence the cycle favored that name.
- 2026-07Restated to 31 Jul 2026: 1-year +83.02% against SOXX's +104.58%. The YTD gap narrowed from 30.9pp at 30 Jun to 17.7pp — in the drawdown leg the equipment and memory tilt cost SOXX rather than paying it.
Risks & fit
- Single-name de-rating: at ~21% weight, a 30% NVIDIA drawdown alone costs ~6% before any other holding moves.
- Geopolitical chokepoint — TSM plus ASML is ~14% of assets, concentrating Taiwan-strait and export-control risk in two positions.
- AI capex is the demand driver; a hyperscaler spending pause transmits to nearly every holding at once.
- Semis are the most cyclical major equity sector — 40-50% drawdowns are normal, not tail events; the group is already ~22% off June 2026 highs.
- Crowding: SMH is itself a large marginal buyer of its top names, so flows and prices reinforce each other in both directions.
The concentration critique is falsified if top-heavy cap weighting delivers superior risk-adjusted returns across a full cycle rather than one leg of one. Concretely: if over a complete peak-to-trough-to-recovery semiconductor cycle SMH's Sharpe ratio matches or exceeds SOXX and equal-weighted XSD, and its maximum drawdown is not materially worse, then the concentration is being compensated and the criticism is wrong. The 2026 result — SMH trailing SOXX by ~20 points over one year while carrying more single-name risk — is one data point in the other direction, and one data point is not a cycle.
Fits an investor who wants deliberate, high-beta exposure to the largest AI-chip names and positions that way knowingly — treating SMH as a concentrated position, not a diversified sector sleeve. Poorly matched to someone seeking broad semiconductor diversification: a 70.92% top-ten weight means risk is dominated by two or three names. For anyone already holding NVIDIA, TSM, or Broadcom directly, look-through overlap is material — SMH likely doubles that exposure.
0.35% net expense ratio (stockanalysis.com, 21 Jul 2026), in line with SOXX (0.34%) and XSD (0.34%). Fee is not the differentiator; weighting methodology is. Dividend yield 0.20%, so returns come from price, not income.
- 2026-07July 2026 was a -17.59% month for SMH, with SOX -20.6% — its worst since Oct 2008 — against a flat S&P 500. SMH closed 31 Jul 19.2% below its 22 Jun high, after -24.6% at the 29 Jul low.