Semiconductor ETFs: Which Ones Suit Your Portfolio The Best?

Most semiconductor ETFs own many of the same chip companies. Yet in 2026, choosing between them has meant choosing very different outcomes. One fund put more than 22% in Nvidia. Another kept Nvidia below 3%. One charged 0.19% a year. Another used a factor model, turned over its portfolio aggressively and charged 0.56%. The label says “semiconductors” in every case. What sits beneath the label can be a completely different bet.

Let’s break down the major US-listed semiconductor ETFs, the index rules driving their returns, where each one fits, and the numbers investors should check before buying one.

The Short Answer: There Is No Single “Best” Semiconductor ETF

There is a best ETF for a particular job.

• SOXX is our best all-round default for an investor who wants a liquid, reasonably balanced basket of large semiconductor leaders.
• SOXQ is our low-cost pick. Its 0.19% expense ratio is hard to ignore if the goal is broad chip exposure held for years.
• SMH is the high-conviction leaders fund. It is the clearest choice when an investor deliberately wants Nvidia and the largest global chip companies to drive the outcome.
• XSD is the breadth bet. Its modified equal-weight approach is built for investors who think the next leg of the cycle will spread beyond today's giants.
• PSI and FTXL are factor bets, not plain sector trackers. They can win when their models are in sync with the cycle, but their higher fees and portfolio turnover demand more scrutiny.

We use the word “default” carefully. A semiconductor ETF can be a core holding within a sector allocation. It is not a diversified replacement for the core of an investor's entire equity portfolio.

What Counts As A Major Semiconductor ETF In This Comparison?

We have placed six broad, unleveraged US-listed funds in the main comparison: SMH, SOXX, SOXQ, XSD, PSI and FTXL. Together, they cover the most useful construction choices available to investors and each has roughly $1 billion or more in assets.

Specialist funds such as SMHX, SHOC, CHPS and DRAM are covered separately. They can be useful, but they make narrower bets on fabless designers, lower Taiwan exposure, global breadth or memory chips.

Leveraged and inverse products such as SOXL and SOXS are excluded from the ranking. Their daily reset makes long-period returns path-dependent. They are trading instruments, not ordinary buy-and-hold semiconductor ETFs.

First Understand The Chip Industry's Four Economic Engines

A chip ETF is not simply a basket of companies making the same product. The supply chain has different businesses with different economics:

1. Designers such as Nvidia, AMD, Broadcom and Qualcomm decide what the chip should do. Many outsource manufacturing and can earn high margins, but their valuations depend heavily on future growth.
2. Foundries such as Taiwan Semiconductor Manufacturing Company, or TSMC, manufacture advanced chips for other companies. Their competitive moat is process technology, scale and execution.
3. Equipment suppliers such as ASML, Applied Materials, Lam Research and KLA sell the tools required to build chips. They are the picks-and-shovels suppliers of the industry, though orders can swing with fabrication-plant spending.
4. Memory, analogue and integrated manufacturers such as Micron, Texas Instruments and Intel own different combinations of design and manufacturing. Memory can be sharply cyclical because small changes in supply often create large changes in pricing.

Think of the industry as a restaurant. Designers write the menu. Foundries run the kitchen. Equipment companies sell the ovens. Memory companies supply a commodity-like ingredient whose price can change quickly. Every ETF chooses how much of each part it wants, even when the fund name does not tell you.

That distinction matters more than ever. The World Semiconductor Trade Statistics organisation forecasts the global chip market to grow about 90% to $1.51 trillion this year, with memory revenue expected to rise more than 250%. 

It is a forecast, not a guarantee, but it tells us that the current boom is not only an AI accelerator story. It is also a powerful memory and capacity cycle. Meanwhile, SEMI expects global 300mm fabrication equipment spending to rise 18% to $133 billion in 2026 and another 14% to $151 billion in 2027.

An ETF heavy in Nvidia, a fund tilted towards Micron and a portfolio rich in equipment makers can therefore report very different returns while all three correctly call themselves semiconductor ETFs.

The Six Main Semiconductor ETFs At A Glance

Sources: Fund issuers and fund holdings as of August 20-24, 2026. Official portfolio counts can differ slightly from third-party counts that include cash or derivatives. (VanEck, iShares, Invesco SOXQ fact sheet, State Street, Invesco PSI fact sheet, First Trust)

The table reveals the central point of this article. The index is the portfolio manager. It decides which companies qualify, how they are weighted, when they rebalance and how much one winner can dominate.

Our CHIP Framework For Analysing Any Semiconductor ETF

Investors can screen a fund using four questions. We call it CHIP.

C: Concentration load

Check the largest holding, top five and top 10. A 25-stock fund is not automatically concentrated, and a 50-stock fund is not automatically diversified. What matters is the distribution of weights.

H: Holdings map

Classify the holdings by their role in the value chain. How much belongs to design, manufacturing, equipment, memory, analogue and electronic-design software? Then compare that map with the thesis you actually hold.

I: Index engine

Ask whether the fund weights by market value, equal weight, cash flow or a factor score. Also check rebalance frequency and turnover. These rules explain why yesterday's winner can become tomorrow's largest holding, or why an equal-weight fund repeatedly trims winners.

P: Price paid

Expense ratio is only the first price. Add bid-ask spread, tracking difference, valuation and any currency or tax friction relevant to you. A cheap fund traded carelessly can still be expensive.

In one line:

ETF return = semiconductor cycle + weighting effect + rebalancing effect - fees - trading friction

This mental model is more useful than choosing whichever ticker topped a one-year performance table.

How The Leading Funds Have Performed

Data through July 31, 2026. Returns include distributions and periods longer than one year are annualised. Volatility is three-year standard deviation, so a higher number means returns varied more widely. Sources: issuer reporting and AAII fund data. (SMH, SOXX, SOXQ, XSD, PSI, FTXL)

PSI and FTXL have led in 2026 because their factor rules happened to capture several strong areas of the current cycle. SMH has been the superior five-year compounder in this group, helped by its concentration in the largest winners.

Neither fact tells us which will lead next. Performance tables are rear-view mirrors. They show the consequence of an index design in one market regime, not a durable promise.

There is another surprise. XSD has the lowest top-10 concentration by far, yet it has the highest three-year volatility in this table. Why? It owns far more small and mid-sized chip companies. Diversification across company names does not guarantee diversification across business risk.

1. VanEck Semiconductor ETF (SMH): The Leaders Fund

SMH tracks the MVIS US Listed Semiconductor 25 Index. Despite the name “US Listed”, the portfolio can own foreign companies with US listings. That allows it to hold TSMC and ASML alongside US designers and equipment suppliers.

Its defining feature is concentration. Nvidia was 22.18% of the fund on August 20, while TSMC was 9.46%. Broadcom, Micron, AMD, ASML, Applied Materials, Marvell and Lam Research followed. The top 10 represented roughly 72% of assets.

That concentration has been a feature, not a bug, while the largest winners kept winning. It also makes the fund unusually sensitive to one company's earnings expectations, product execution and valuation.

SMH is the largest fund here with $66.84 billion in assets and a 0.35% expense ratio. It has a long history, launched in December 2011, and its scale supports active trading.

Our view: SMH is best for an investor who has a positive view on the current global leaders and is comfortable letting market capitalisation express that view. We would not call it the most diversified semiconductor ETF. We would call it the most direct large-winner portfolio. If Nvidia already has a large weight in your Nasdaq-100, S&P 500 or technology holdings, calculate the combined exposure before adding SMH.

2. iShares Semiconductor ETF (SOXX): The Balanced All-Rounder

SOXX follows the NYSE Semiconductor Index and holds about 30 companies. Its modified market-cap method prevents one company from reaching SMH-like dominance. Nvidia was 9.05% of the portfolio on August 20, followed by Micron at 8.84%, AMD at 8.02% and Broadcom at 7.18%. Equipment makers and TSMC also had meaningful weights.

The fund had about $41.76 billion in assets, a 0.33% expense ratio and an average bid-ask spread of 0.03% on the issuer's reported 30-day measure. That combination of scale, trading depth and more balanced weights is hard to beat.

The compromise is simple. SOXX will usually capture less upside than a more concentrated fund when one giant races far ahead. It can also fall less because of that one stock, all else equal. Its return still depends heavily on the semiconductor cycle, so “balanced” should not be mistaken for “low risk”.

Our view: SOXX is the best all-round default for investors who want a single, broad large-cap semiconductor sector fund. It does not win on fee, concentration or factor sophistication individually. It wins because the overall package has few obvious weaknesses.

3. Invesco PHLX Semiconductor ETF (SOXQ): The Low-Cost Core

SOXQ tracks the PHLX Semiconductor Sector Index and offers broad large-cap chip exposure for a 0.19% annual expense ratio, the lowest among the six main funds. Its largest positions were Nvidia at 13.22%, Broadcom at 9.96% and Micron at 7.65%, followed by Applied Materials, ASML, KLA, Marvell, TSMC and Lam Research.

SOXQ launched only in June 2021, so it has a shorter live history than SOXX, SMH, XSD or PSI. It is also much smaller at roughly $2.99 billion. That is not a sign of an unsafe portfolio. ETF assets are held separately from the sponsor. It does mean trading liquidity and the fund's own record deserve attention.

For a long holding period, the fee difference compounds. On $10,000, SOXQ's fee is about $19 in the first year versus $35 for SMH. If both portfolios earned a hypothetical 10% gross annual return for 20 years and fees were the only difference, a simple gross-return-minus-fee calculation leaves roughly $1,868 more in the lower-cost fund. Real returns, tracking differences and holdings will matter much more, but the arithmetic shows why small annual charges should not be dismissed.

Our view: SOXQ is the best value choice for a patient investor who wants conventional, broad semiconductor exposure and can use limit orders. We prefer SOXX when trading depth is important and SOXQ when long-term cost has greater weight in the decision.

4. State Street SPDR S&P Semiconductor ETF (XSD): The Breadth Bet

XSD tracks the S&P Semiconductor Select Industry Index using a modified equal-weight method. Instead of letting Nvidia and Broadcom consume the portfolio, it repeatedly pulls positions towards similar weights. Impinj was its largest holding at 3.51% on August 20, while Ambarella, Micron, Nvidia, AMD, Qorvo, Silicon Labs and Marvell all sat in a relatively narrow band.

Its 28.7% top-10 concentration was less than half that of SOXX and SOXQ, and far below SMH. XSD also held 47 companies, the most among the six core comparisons. State Street reported a 0.35% expense ratio, $2.69 billion in assets and a 0.16% 30-day bid-ask spread.

Equal weighting is like resetting a race so every runner gets another meaningful chance. It sells part of the leaders and adds to laggards during rebalancing. That can work beautifully when gains broaden. It can trail badly when a few giants keep pulling away.

Investors should also know that the index's industry classification makes XSD different from a full chip supply-chain basket. Its portfolio can give less direct exposure to some semiconductor equipment names than funds whose rules explicitly include that group.

Our view: XSD is the best fund for a “next winners” or market-broadening thesis. It is not our safety pick. Its smaller companies have produced higher volatility even though position-level concentration is low. Use it because you want breadth, not because equal weight sounds automatically conservative.

5. Invesco Semiconductors ETF (PSI): The Active-Looking Index

PSI tracks the Dynamic Semiconductor Intellidex Index. The rules evaluate US semiconductor companies using price momentum, earnings momentum, quality, management action and value. The portfolio is reconstituted and rebalanced quarterly.

That makes PSI look more like an active strategy delivered through an index. Applied Materials was its largest holding at 6.50%, followed by Nvidia, Micron, AMD, KLA, Lam Research, Analog Devices and Intel. The top 10 made up about half of the portfolio, less concentrated than the three cap-led funds but more concentrated than XSD.

PSI's 73.0% year-to-date and 129.4% one-year return through July were the strongest in the main group. Its reported portfolio turnover was also about 105% in the latest data, meaning the value traded over a year was roughly comparable with the whole portfolio. Turnover is not automatically bad inside an ETF, but it shows that the factor engine is doing substantial work.

The fund charges 0.56%, or $56 for each $10,000 before compounding and trading effects.

Our view: PSI suits an investor who deliberately wants systematic factor timing within semiconductors. We would not buy it merely because it topped the latest table. At 0.56%, the index model needs to keep earning its keep across cycles. Its 2026 performance proves that the method can look excellent, not that it will always do so.

6. First Trust Nasdaq Semiconductor ETF (FTXL): The Cash-Flow and Momentum Hybrid

FTXL tracks the Nasdaq US Smart Semiconductor Index. Companies are selected using gross income, return on assets and momentum, then weighted using cash flow, subject to caps and floors. The portfolio is rebalanced semi-annually.

The rules created a distinctive portfolio in August. Micron was 12.80%, Intel 10.95%, Marvell 8.33%, Nvidia 6.63%, AMD 6.35%, Broadcom 6.17% and Qualcomm 6.12%. The top 10 represented about 68% of assets. In other words, FTXL was concentrated, but in a very different set of names from SMH.

This is why ticker counts can mislead. FTXL had 34 holdings, more than SMH, yet its top 10 weight was only a few percentage points lower. Its factor engine had also built a large memory and Intel tilt at a time when those areas were strong.

FTXL charges 0.60%, the highest fee in the main group. Compared with SOXQ, the annual gap is 0.41 percentage point, or $41 on an initial $10,000 in year one. Under the same hypothetical 10% gross-return assumption and simple fee approximation used earlier, that gap alone compounds to about $4,685 over 20 years.

Our view: FTXL is a useful factor fund for investors who understand and want its quality, cash-flow and momentum rules. It is not the efficient default. The strategy needs to outperform a cheaper broad portfolio by more than 0.41 percentage point a year before trading differences just to overcome the fee gap.

What About Specialist Semiconductor ETFs?

SMHX: fabless is a business-model bet

The VanEck Fabless Semiconductor ETF excludes the manufacturing-heavy parts of the chain and focuses on designers and related intellectual property. Nvidia was 22.52% and Broadcom 16.84% in mid-August. That gives investors purer exposure to design economics, but removes foundry and equipment holdings that might behave differently.

Our view is positive only as a satellite. Investors buying it alongside SMH or a technology index should check whether they are simply adding another layer of Nvidia and Broadcom.

SHOC: reducing one geopolitical exposure, not all of it

The Strive U.S. Semiconductor ETF explicitly excludes TSMC. Its June 2026 fact sheet showed 89.12% US country exposure, with Nvidia, Micron, Broadcom and Applied Materials as major positions.

That is useful if an investor's thesis specifically concerns Taiwan. It does not make the supply chain domestic. US chip companies still depend on overseas manufacturing, customers and suppliers. Excluding the world's leading advanced foundry can also create opportunity cost. Geography in a fund factsheet is not the same as geography in a company's operations.

CHPS: the overlooked global option

The Xtrackers Semiconductor Select Equity ETF held companies across the US, Taiwan, South Korea, Japan and Europe. Recent holdings included Nvidia and Broadcom, but also TSMC, SK Hynix, MediaTek, Tokyo Electron and Advantest. Its 0.15% expense ratio is even lower than SOXQ's.

Why is it not our low-cost winner? It remains small, and its underlying index changed in July 2026 when the ESG screen was removed and the benchmark was renamed. We like the breadth and fee, but would watch live spreads, tracking and how the new methodology behaves before using it for a large core allocation.

DRAM: memory prices with an ETF wrapper

The Roundhill Memory ETF launched in April 2026 as an actively managed portfolio focused on memory and storage. Its main exposures include Micron, Samsung Electronics, SK Hynix, SanDisk and Kioxia. The fund may use swaps to obtain some foreign exposure.

DRAM is a sharp tool. Memory profits can rise rapidly when demand outruns supply, then reverse when new capacity arrives or inventories rebuild. We see it as a small satellite for investors with a specific memory-cycle view, not as an alternative to SOXX, SOXQ or SMH.

Semiconductor ETFs Concentration Stress Test

Here is a simple way to translate holding weights into portfolio risk. Assume Nvidia falls 30% overnight and every other holding is unchanged. This is not a forecast. It isolates the mechanical effect of position size.

The math is weight multiplied by price move. For SMH: 22.18% × -30% = -6.65%.

In dollar terms, a $10,000 SMH position placed roughly $2,218 in Nvidia at the measurement date. SOXX placed about $905 there; XSD placed $292. That is a more intuitive description of concentration than saying one fund has 26 holdings and another has 47.

The same test should be run for Micron if the thesis is memory, TSMC if the concern is Taiwan, or equipment companies if fabrication spending is the key driver.

Owning Three Chip ETFs May Give the Illusion of Diversification

SMH, SOXX and SOXQ overlap heavily. Combining them spreads money across different weights more than different businesses. An investor can end up paying three fees to own many of the same companies.

Use this portfolio-level formula:

Total stock exposure = sum of each fund's portfolio weight × that stock's weight inside the fund

Suppose a broad US equity ETF is 60% of a portfolio and Nvidia is 8% of that fund. Add a 10% position in SMH with Nvidia at 22.18%. Nvidia then represents about 7.02% of the entire portfolio:

• Broad ETF contribution: 60% × 8% = 4.80%
• SMH contribution: 10% × 22.18% = 2.22%
• Total: 4.80% + 2.22% = 7.02%

The example is illustrative, and live weights should always be used. The point is that sector allocation must be checked against what you already own.

Our preferred structure is usually one broad core semiconductor ETF plus, at most, one genuinely different satellite. For example, SOXQ plus a small DRAM allocation expresses broad chips plus a memory view. Owning SOXX, SOXQ and SMH together mostly averages three weighting rules.

Which Semiconductor ETF Suits Which Thesis?

The Three Clocks Investors Should Monitor After Buying

Buying is only the first decision. A semiconductor position should be monitored through three clocks.

1. The earnings clock

Track revenue growth, margins, order books and earnings revisions for the ETF's largest holdings. In a concentrated fund, two or three companies can explain a large part of the result. The Semiconductor Industry Association reported that global chip sales rose 35.1% from the first to the second quarter of 2026, showing how powerful the current demand acceleration has been. Such growth can also raise the hurdle for future comparisons.

2. The capital-cycle clock

Watch inventory, memory pricing, foundry utilisation and fabrication-equipment spending. High prices attract new capacity. New capacity eventually changes pricing power. The best part of a semiconductor business cycle can contain the seeds of its reversal.

3. The index clock

Check the fund after rebalances. Equal-weight funds trim winners. Factor funds can replace holdings. Market-cap funds allow winners to grow. If the ETF no longer expresses the thesis that led you to buy it, the ticker has stayed the same but the investment has changed.

A useful review schedule is quarterly for holdings and thesis, annually for fees and tracking, and immediately after a major methodology change.

What Are the Risks of Investing in Semiconductor ETFs?

1. Semiconductor cyclicality: Demand for AI computing is structural, but semiconductor revenue and profits still move in cycles. Customer inventory, pricing, capacity additions and capital spending can amplify both the rise and the fall.

2. Valuation risk: The best business can still be a poor investment if expectations are too high. Issuers may report very different price-to-earnings ratios because of forward versus trailing earnings, treatment of loss-making companies and averaging methods. For example, XSD's issuer reported a forward P/E near 27 while iShares displayed a portfolio P/E above 60 for SOXX in August. Those figures are not directly comparable. Investors should compare valuations using the same data provider and methodology.

3. Concentration and overlap: A sector ETF removes single-company selection work but does not remove concentration. SMH's Nvidia position proves the point. Overlap with technology ETFs and broad US indices can make the true exposure much larger than the new allocation suggests.

4. Geopolitics and supply-chain dependence: Advanced chips cross borders several times through design, equipment, manufacturing, packaging and assembly. Taiwan is important, but simply excluding a Taiwan-listed company does not make the remaining revenue or supply chain domestic.

5. Policy and export controls: Chips are strategic goods. Export restrictions, subsidies, tariffs and licensing rules can change addressable markets and costs. These effects are not uniform across designers, equipment suppliers and foundries.

6. Liquidity and execution: The ETF's share price can move around its net asset value during volatile sessions. A 0.03% spread is about $3 on a $10,000 trade; 0.16% is about $16. Use limit orders, especially in smaller funds, and prefer trading when both US markets and the underlying securities' home markets are open where practical.

7. Leveraged ETF decay: Daily leveraged products target a multiple of each day's return, not a multiple of the long-period return. Volatility and daily compounding can create large gaps from what a buy-and-hold investor expects. That is why they are excluded from our core list.

Our Final Ranking, By Job

If we had to narrow the full list to decisions that are easy to defend:

1. SOXX for the best all-round construction. Its scale, liquidity, supply-chain coverage and moderate position caps make it the cleanest single-fund answer for many investors.
2. SOXQ for long-term cost efficiency. It offers a similar broad economic exposure at 0.19%, with the trade-off of a shorter history and lower trading depth.
3. SMH for deliberate concentration. Choose it when you specifically want the current mega-cap leaders to dominate. Do not choose it while pretending it is evenly diversified.
4. XSD for a broadening cycle. Its equal-weight rules provide the most different portfolio among the broad funds, but its smaller companies can make the ride rougher.
5. PSI over FTXL among factor funds, narrowly. PSI's broader factor set and less extreme current top-10 concentration earn our preference, though neither is a default at its fee.

The specialist funds come after the core decision. CHPS is the most interesting watchlist candidate for low-cost global breadth. SMHX is a focused fabless bet. SHOC is a geopolitical preference. DRAM is a memory-cycle trade. Each can add something distinct, but only when the investor can state the thesis in one sentence.

The wrong way to select a semiconductor ETF is to ask which ticker went up the most. The better question is: which index rule will make money if my industry thesis is right, and what will hurt me if it is wrong? Once that is clear, the choice becomes far less mysterious.