Invesco Semiconductors ETF
Invesco
Tracks the Dynamic Semiconductor Intellidex Index — a quant/factor-selected basket of ~30 US semi stocks weighted by a multi-factor model rather than pure market cap, giving it an equipment-heavy, equal-weight tilt that reaches down into small-cap supply-chain names.
Top holdings
Top weight; equipment tilt vs cap-weighted peers
Memory/HBM
Process-control equipment
Etch/deposition equipment
CPUs and foundry
NVIDIA sits lower (~4%, #8) in this factor weighting
Recent moves
The factor/equipment tilt (AMAT, KLAC, LRCX top-weighted plus small-cap supply-chain names) outran cap-weighted peers in the equipment-led leg of the rally.
Our take
The contrarian pick of the semi group — its factor model under-weights NVIDIA and over-weights equipment + small-cap supply chain, which paid off in 2026's equipment-led leg but is the bet that lags when mega-cap chips lead. Higher expense ratio (0.56%) and far smaller AUM than SMH/SOXX.
The 2026 record is a factor bet that happened to align with the cycle's actual driver.
The AI capex wave paid the toolmakers and memory suppliers before it paid the fabless designers, and PSI's model — which systematically de-weights the most expensive megacap and up-weights momentum in cheaper mid-caps — was positioned there by construction, not by forecast. That is a real structural argument, and it is also the exposure: PSI wins when breadth is wide and equipment/memory leads, and lags when one megacap does the lifting, exactly the 2023-24 regime when SMH's Nvidia weight was the whole trade. Two second-order costs the marketing won't name. First, two of the five factors are momentum, and a quarterly reconstitution buys what just worked — after a +100% run the model tends to add to names that already moved, which is trend-following, not diversification. Second, 0.56% is roughly 1.6x SOXX/SMH's 0.35%, and four reconstitutions a year layer an unpublished turnover cost on top.
Thesis
PSI is not a semiconductor index fund — it is a quarterly-reconstituted quant factor sleeve wearing a sector label. The Dynamic Semiconductor Intellidex ranks ~30 US chip names on price momentum, earnings momentum, quality, management action and value rather than size, which put 5.39% in Nvidia (Jul 17 2026) vs SMH's 20.97% (Jul 18 2026) and pushed weight into equipment and memory — AMAT, KLAC and LRCX are its top three. You buy a factor model's opinion of the sector, not the sector.
Rules-based five-factor selection from a US semiconductor universe, 32 holdings, modified weighting with tiered caps, rebalanced and reconstituted quarterly in February, May, August and November — so it drifts down the cap ladder into supply-chain names (Ichor 3.51%, Ultra Clean 3.23%) that cap-weighted peers barely hold. Top 10 = 51.00% of assets vs SMH's 70.92%. Expense ratio 0.56%, AUM $2.40B (Jul 2026); inception Jun 2005.
- 2026-07Holdings at Jul 30 2026: AMAT 6.63%, MU 5.86% and AMD 5.79% are the top three, with LRCX 4th (5.54%), NVDA 5th (5.45%) and KLAC 6th (5.39%). The AMAT/KLAC/LRCX trio is 17.56%.
Assessment
- Nvidia weight of 5.39% vs SMH's 20.97% (Jul 2026) makes it the genuine single-stock-concentration alternative among large semi ETFs.
- Top-10 at 51.00% vs SMH's 70.92% — materially less concentrated than the cap-weighted majors, though not as flat as equal-weight XSD.
- Reaches the equipment and small-cap supply chain (Ichor, Ultra Clean) that cap-weighted indexes structurally exclude.
- 21-year live record since Jun 2005 — the methodology has run through multiple full semi cycles, not just this one.
- 0.56% expense ratio is ~1.6x SOXX/SMH's 0.35%, before the trading cost of four reconstitutions a year.
- Two of five factors are momentum; quarterly rebalancing buys strength after the move and can add to already-extended names.
- $2.40B AUM vs SMH's $67.42B means wider spreads and thinner liquidity in stressed tape.
- Small-cap supply-chain names import operating leverage — they fall harder than megacaps in a capex pause.
- The label says 'semiconductors' but the exposure is a factor model; tracking a mental SOX benchmark is not its goal.
Record
1-year total return +131.76% (Jul 21 2026) vs SMH's +93.09% (Jul 20 2026) — a ~39pt lead on a like-for-like comparison. Attribution is unusually clean: the gap is almost entirely the Nvidia underweight not costing anything this cycle plus the equipment/memory overweight paying, with AMAT, KLAC and LRCX alone at ~18.5% combined and low-base-weight supply-chain positions contributing outsized dollars. This is the mirror image of 2023-24, when the same underweight was why PSI trailed. Read the lead as regime alignment, not model superiority — one cycle of a tilt being right is not evidence the tilt is right. Since-inception average annual return of 17.79% is a far more honest anchor than the trailing year.
- 2026-07Re-marked at Jul 31 2026: PSI +120.97% over one year against SMH's +83.02%, and YTD +73.0% versus +138.2% at Jun 30. The ~38pt relative lead survives, but on a far lower base.
- 2026-07July 2026 inverted that attribution: NVDA +0.3% and AVGO +3.1% were the haven while equipment fell 28-30% and MU -28.7%. PSI -27.4% against SMH -17.6% — the Nvidia underweight cost about 10pt in a month.
Risks & fit
- Regime flip: if leadership narrows back to one or two megacap designers, the 5.39% Nvidia weight becomes the drag it was in 2023-24.
- Capex-cycle turn — WFE spending is the most cyclical line in semis; an equipment-heavy book is levered to a hyperscaler capex pause.
- Quarterly reconstitution (Feb/May/Aug/Nov) can lock in extended momentum weights immediately before a reversal.
- Small-cap supply-chain drawdowns run far deeper than the sector's; Ichor/UCTT-class names round-trip violently.
- Fee plus four-times-yearly turnover drag compounds against cheaper cap-weighted peers in flat or sideways sector years.
The core claim — that construction, not luck, produced the lead — fails if PSI underperforms SOXX over a full cycle including a downturn, or if attribution shows the outperformance concentrated in two or three sub-4% starting weights rather than the broad equipment/memory tilt. It also fails if a coming reconstitution materially raises Nvidia weight, which would show the model chasing megacap momentum rather than structurally diversifying away from it. Watch the post-rebalance holdings file after each Feb/May/Aug/Nov reconstitution and the relative line versus SOXX through the first real capex-guidance cut.
Someone who already wants semiconductor exposure, has looked at SMH's ~21% Nvidia weight and decided single-name concentration is what they want less of, and accepts 0.56% plus deeper small-cap drawdowns to get a broader, equipment-tilted book. It is a poor proxy for anyone whose mental model of 'semis' is the megacap AI designers — PSI diverges from that expectation by design, in both directions.
0.56% expense ratio vs 0.35% for SOXX and SMH — roughly 21bp/yr of headline drag, plus unquantified turnover cost from four reconstitutions a year.
- 2026-07Both conditions arrived in July 2026. SOX fell 20.6% in an AI-hardware-specific rout, its worst month since Oct 2008, while NVDA and AVGO held up and MSFT rose 24.6% after holding capex flat.