AI Dominated ETFs: Which Ones Suit Your Portfolio The Best?

An “AI ETF” can mean three completely different portfolios. One may own Nvidia, memory-chip manufacturers and data-centre suppliers. Another may be dominated by Microsoft, Amazon and Alphabet. A third could hold Japanese factory-automation companies. They all carry the AI label, but they are not betting on the same future. That distinction mattered enormously in 2026 as the best-performing dedicated AI ETF returned 68.7% through July, while another lost 2.7%.

Let’s break down the leading US-listed AI ETFs, what each fund really owns, why their returns have been so different, and which ones make the most sense for different AI investment theses.

First, What Qualifies As A Best “AI Dominated ETF”?

There are now dozens of funds with artificial intelligence, robotics, automation, innovation or next-generation technology somewhere in their names.

Including all of them would create more noise than insight. A broad technology ETF does not become an AI ETF simply because Nvidia and Microsoft are among its holdings. Similarly, a semiconductor ETF is an important AI benchmark, but it does not claim to select companies across the full AI value chain.

For this analysis, we selected the 10 largest US-listed ETFs whose stated mandate is substantially focused on AI, generative AI, AI infrastructure, AI software or AI-powered robotics.

Sources: Global X, iShares, Roundhill, Invesco, Wedbush Funds, VistaShares, First Trust, WisdomTree and ROBO Global.

How Is An AI ETF Actually Constructed?

An ETF provider does not simply ask, “Which companies are involved in AI?” and buy the answers.

The process usually has five stages.

First, the index provider defines the AI universe. This may involve revenue screens, company filings, business-segment classifications, patents, natural-language analysis or an internal committee’s judgment.

Second, companies are separated into categories. These can include AI enablers, infrastructure providers, software developers, AI users and robotics businesses.

Third comes weighting. A fund can weight companies by market capitalization, revenue exposure, an equal-weight formula or a proprietary score. An active fund lets a portfolio manager decide.

Fourth, concentration limits are applied. A cap prevents one or two stocks from swallowing the entire portfolio.

Finally, the fund rebalances, usually quarterly or semi-annually. Stocks whose businesses have become more relevant can enter, while those no longer meeting the rules can be removed.

Think of it like selecting a cricket team. Every selector may agree that the team should be built to win the match, but one may pick six batters, another may load up on fast bowlers, and a third may choose all-rounders. The label on the jersey is the same. The way the match plays out is not.

AI ETF Returns in 2026: The “AI” Label Alone Did Not Drive Performance

Here is where the comparison becomes revealing.

Returns are total returns through July 31, 2026. Most are market-price returns; issuer NAV returns are used where comparable market-price data was unavailable. Small differences between NAV and market-price returns can occur.

The 71.4 percentage-point gap between AIS and BOTZ was not primarily a debate about whether AI would grow. It was a debate about which part of AI would capture investor money first.

AIS owned memory manufacturers, semiconductor designers, foundries and data-centre infrastructure suppliers. BOTZ had much heavier exposure to industrial automation, Japanese robotics and factory equipment.

AI spending accelerated, but its stock-market rewards were not distributed evenly.

That leads to the most important lesson in this article:

“An AI ETF’s return is driven less by the word “AI” and more by its hidden factor exposures.”

A useful way to express this is:

AI ETF return ≈ technology-market return + semiconductor cycle + concentration effect + portfolio selection − fees

This is not a precise forecasting equation. It is a better way to think about performance attribution. When a fund rises, investors should ask which component produced the return before concluding that its AI strategy worked.

The Semiconductor Reality Check

Comparing AI ETFs with semiconductor funds exposes this issue clearly.

Sources: fund issuers for SOXX, SMH, QQQM and VOO.

AIS was the best-performing dedicated AI ETF, but the lower-cost SOXX almost matched it. This suggests that much of AIS’s extraordinary return came from correctly being exposed to the semiconductor and memory cycle, rather than from an isolated “AI ETF effect.”

That does not make AIS a bad fund. It makes the benchmark more demanding.

For AIS to justify a 0.75% fee over a 0.35% semiconductor ETF, its investments outside traditional chipmakers must add value over time. These include data-centre cooling, electrical equipment, networking and other infrastructure businesses.

The same test applies to ARTY and IGPT. If their returns are primarily being generated by Nvidia, AMD, Micron and SK Hynix, investors must ask what they are paying the extra thematic fee to receive.

How to Compare AI ETFs: The STACK Framework

Before selecting an AI ETF, “STACK” it.

S: Source of AI economics

Where does the fund expect profits to be earned?

This could be compute hardware, data centres, cloud platforms, software applications or physical robots. A fund betting on memory demand has a completely different earnings cycle from one betting on enterprise-software subscriptions.

T: Theme purity

How directly is a company’s revenue connected with AI?

Nvidia earns substantial revenue from computing systems used to train and run AI models. A streaming company may use AI extensively but still earns its money from subscriptions and advertising.

Using AI is not the same as selling the tools required to build AI.

A: Allocation mechanics

Is the portfolio market-cap weighted, equal weighted, tiered or active?

Market-cap weighting generally keeps winners large, but can make the portfolio resemble a conventional technology index. Equal weighting offers smaller companies more influence, but may repeatedly trim successful businesses to buy weaker ones.

C: Concentration and duplication

How much sits in the 10 largest holdings? How much do those holdings overlap with investments already owned?

An investor who already holds the Nasdaq-100 may not gain much diversification by adding another fund dominated by Microsoft, Amazon, Alphabet, Nvidia and Meta.

K: Known costs

The expense ratio is only one cost.

Investors should also examine bid-ask spreads, liquidity, valuation, currency conversion, taxes and the behavioural cost of owning an unusually volatile fund.

The STACK framework prevents investors from asking the wrong question, “Which AI ETF has the best past return?” It replaces it with the more useful question, “Which portfolio best represents the AI economics I want to own?”

What Do The 10 Largest AI ETFs Actually Own?

AIQ: The largest and most liquid all-purpose AI ETF

The Global X Artificial Intelligence & Technology ETF is the category leader by assets, with approximately $10.18 billion under management.

Its 88 holdings include Palantir, Microsoft, Oracle, Amazon, SpaceX-linked exposure, Alphabet, Cisco, Broadcom and Tencent. The top 10 account for about 31.2% of the portfolio. Information technology represents roughly 71.5% of the fund, while the US contributes close to 69% of geographic exposure.

AIQ follows the Indxx Artificial Intelligence & Big Data Index. Its construction attempts to balance AI developers and AI-as-a-service companies with hardware, big-data and related technology businesses. Individual weight caps prevent a single mega-cap stock from dominating.

Our view: AIQ is the easiest dedicated AI ETF to use as a broad, liquid category allocation. Its size, number of holdings and relatively controlled concentration are genuine strengths.

Its weakness is theme purity. Holdings such as Netflix may use AI extensively, but investors are not buying Netflix primarily to own AI infrastructure or AI software revenue. AIQ also overlaps materially with the Nasdaq-100 while charging 0.68%, versus 0.15% for QQQM.

AIQ is therefore best for an investor who prioritises liquidity and broad exposure. It is less compelling for someone who already owns a large US technology allocation.

ARTY: The strongest passive full-stack AI portfolio

The iShares Future AI & Tech ETF holds about 49 companies and charges 0.47%, making it cheaper than most dedicated AI peers.

Its largest holdings include Nvidia, Broadcom, TSMC, AMD, Micron, CoreWeave, Super Micro Computer, Marvell and Palantir. The top 10 account for approximately 41.4% of assets.

This is a much more direct AI supply-chain portfolio than AIQ. Semiconductors, computing infrastructure and newer AI businesses have meaningful representation. Approximately 86.7% of the fund sits in information technology, with significant exposure to Taiwan and South Korea alongside the US.

Its rules-based Morningstar Global AI Select Index gives it discipline without turning the portfolio into a watered-down collection of every company mentioning AI.

Our view: If we had to choose one dedicated passive AI ETF as a long-term core allocation, ARTY would get our vote.

It has a credible full-stack portfolio, a reasonable 0.47% fee and enough concentration for winners to matter without becoming a disguised single-stock bet. The trade-off is clear: it is highly sensitive to semiconductor valuations and AI capital expenditure.

ARTY is not a low-volatility fund. It is simply one of the cleaner implementations of the AI thesis.

BOTZ: A robotics fund wearing an AI badge

BOTZ is commonly included in AI ETF lists, but its portfolio tells a more specific story.

Its largest holdings include Keyence, ABB, Nvidia, Fanuc, Intuitive Surgical, SMC and Shenzhen Inovance. The top 10 represent 59.8% of assets, the highest concentration in our comparison.

Industrials make up approximately 45% of the portfolio, compared with about 37% for information technology. Only around one-third of assets are invested in the US. Japan represents roughly 31%, while China and Switzerland are also significant.

The fund’s index generally requires companies to derive at least half their revenue from robotics, automation or AI-related activities, or to have a primary business focus in these areas.

Our view: BOTZ should not be purchased as a general generative-AI or cloud-computing fund.

It is a bet on physical AI: factory automation, robotics, autonomous systems, medical devices and industrial productivity. That can become a powerful long-term theme as labour shortages, reshoring and manufacturing automation grow, but its earnings cycle differs from Nvidia or Microsoft.

BOTZ’s negative 2.7% return in 2026, during a strong year for several AI ETFs, demonstrates the difference. It is a useful satellite for investors specifically seeking robotics. It is not our preferred standalone AI ETF.

CHAT: The best active AI portfolio

The Roundhill Generative AI & Technology ETF is actively managed and charges 0.75%.

Its holdings include Nvidia, Alphabet, Broadcom, SK Hynix, AMD, Micron, Microsoft, Samsung, CoreWeave and newer AI infrastructure businesses such as Nebius. Its top 10 account for about 38% of the portfolio.

The active mandate allows CHAT to move across the AI stack. It can own established platforms, semiconductor suppliers, data-centre businesses and emerging companies that have not yet become large enough to dominate traditional indices.

That flexibility is the fund’s biggest advantage and its biggest test. Investors are paying for judgment, not merely index access.

Our view: CHAT is the strongest option for investors who specifically want active AI management.

Its portfolio has done a good job of extending beyond the obvious mega-cap names into second-order beneficiaries. However, its 0.75% fee is only justified if that selection continues to produce value after costs.

CHAT and ARTY also share considerable exposure. Owning both may feel diversified, but in practice it can mean paying two funds to own many of the same semiconductor and platform companies.

IGPT: High semiconductor torque behind a software name

The Invesco AI and Next Gen Software ETF holds roughly 100 companies and charges 0.56%.

Its current portfolio has been highly concentrated at the top, with Nvidia, Meta, Alphabet, AMD and Micron among its largest positions. The 10 largest holdings account for close to 58.7% of assets.

This creates an important mismatch between expectation and reality. A reader seeing “Next Gen Software” may imagine enterprise applications and recurring subscription revenue. In practice, recent performance has been heavily influenced by semiconductors, memory and mega-cap technology.

There is another detail investors should know. The fund traces its inception to 2005, but it did not operate under its current AI mandate for that full period. It was previously the Invesco Dynamic Software ETF under ticker PSJ. Its name, ticker, benchmark and principal strategy changed in August 2023, according to the fund’s SEC filing.

Our view: IGPT’s long fund history should not be interpreted as a 21-year track record for today’s AI portfolio.

Its concentration creates powerful upside when the semiconductor cycle works, as seen in its 45.8% 2026 return, but it also increases downside risk. IGPT suits investors seeking high sensitivity to AI compute and next-generation technology. It is not as balanced as its large number of holdings initially suggests.

IVES: A concentrated AI consensus portfolio with a new question

The Dan IVES Wedbush AI Revolution ETF owns only around 32 stocks and charges 0.75%.

Microsoft, Amazon, Broadcom, Nvidia, Apple, TSMC, Alphabet, Palantir, Meta and Micron are among its largest positions. Together, the top 10 represent approximately 46.9% of assets.

The portfolio is easy to understand. It owns many of the companies most frequently identified as leaders of the AI investment cycle. This provides high-quality exposure but also substantial duplication with the Nasdaq-100 and other AI funds.

A new issue emerged in July 2026. Dan Ives left Wedbush to establish a new venture. Wedbush stated that its fund-advisory business would continue managing IVES and IVEP without interruption. Operational continuity is therefore not the immediate concern. The more interesting question is intellectual continuity: how will the “AI Revolution” thesis evolve when the person most closely associated with the fund’s brand has left the firm? 

Our view: IVES is currently difficult to prefer over ARTY, CHAT or even a cheaper Nasdaq-100 fund.

Its holdings are credible, but a 0.75% fee is demanding for a portfolio dominated by well-known mega-cap AI stocks. Investors should monitor whether its methodology, branding or portfolio process changes following Ives’ departure.

AIS: The highest-conviction AI infrastructure ETF

AIS is an actively managed fund built around the AI infrastructure supercycle.

Its largest positions include Micron, SK Hynix, AMD, Vertiv, TSMC, Marvell, Intel, Silicon Motion, Nvidia and Foxconn Industrial Internet. Semiconductors represented close to half the portfolio in its latest detailed sector disclosure.

This positioning explains its extraordinary 68.7% return through July 2026 and its 114.8% one-year return. It also explains its risk. The fund’s one-year volatility was close to 48%, and its NAV fell approximately 23.8% in July alone, according to VistaShares’ performance data.

That is the nature of a high-beta infrastructure portfolio. AI investment may rise steadily over several years, but the stocks supplying it can move violently as valuations, memory prices and capital-expenditure expectations change.

Our view: AIS is the strongest high-conviction “picks and shovels” AI fund in the group, but it should be viewed as a satellite, not a complete portfolio.

Its 0.75% expense ratio must be judged against SOXX, which nearly matched its one-year return at less than half the fee. AIS earns its place only if its broader selection of memory, networking, cooling and data-centre suppliers can outperform a conventional semiconductor basket over a full cycle.

ROBT: Real diversification, but at a performance cost

The First Trust Nasdaq Artificial Intelligence and Robotics ETF is the most diversified portfolio in this comparison.

It holds approximately 114 companies. Appian, CCC Intelligent Solutions, Palo Alto Networks, SentinelOne, Oceaneering, UiPath, Cloudflare, Workday and Dynatrace are among its larger positions. The top 10 represent only about 18.8% of assets.

ROBT divides companies into three groups: AI and robotics “engagers,” “enablers” and “enhancers.” It assigns 60%, 25% and 15% of the portfolio to these categories respectively, with companies weighted relatively evenly inside each group.

This gives smaller application and automation businesses more influence. Nvidia, Microsoft and Amazon do not dominate the fund.

The downside is visible in its returns. ROBT gained only 6.8% in 2026 and 10.9% over one year. Its five-year annualised return through July was about 0.8%, materially behind the broader US market.

Our view: ROBT is one of the few AI ETFs that genuinely reduces mega-cap duplication, but diversification alone does not guarantee quality.

It can complement a portfolio already heavy in Nvidia and Microsoft. We would not select it as the primary AI allocation until its broader collection of AI-adjacent businesses demonstrates stronger earnings and share-price participation.

WTAI: The best value-for-money AI ETF

The WisdomTree Artificial Intelligence and Innovation Fund charges 0.45%, the lowest expense ratio among the 10 dedicated AI ETFs analysed.

Its largest holdings include Nvidia, Micron, Amazon, Samsung, Meta, Palo Alto Networks, Alphabet, Broadcom, TSMC and Oracle. The top 10 account for approximately 39.1% of assets.

WTAI provides exposure across semiconductors, cloud platforms, cybersecurity, enterprise software and AI applications. It is less concentrated than BOTZ or IGPT and cheaper than AIQ, CHAT, IVES, AIS and THNQ.

Its 33.5% return in 2026 and 55.7% one-year gain show that the lower fee did not require sacrificing participation in the AI trade.

The principal weakness is trading liquidity. Its bid-ask spread can be wider than those of AIQ or major broad-market ETFs. Investors using market orders can therefore lose part of the apparent fee advantage at the point of purchase.

Our view: WTAI is the strongest fee-conscious alternative to ARTY.

ARTY offers somewhat cleaner full-stack AI exposure. WTAI offers the better price. For patient investors who use limit orders and do not require very high daily liquidity, the balance is attractive.

THNQ: A balanced portfolio held back by cost and liquidity

The ROBO Global Artificial Intelligence ETF aims to cover both AI infrastructure and AI applications.

Its portfolio has included companies such as Palo Alto Networks, Cloudflare, CrowdStrike, Snowflake, Amazon, Shopify, Arista Networks, Ambarella and Alibaba. The top of the portfolio is more evenly distributed than BOTZ, IGPT or IVES.

That makes THNQ less dependent on one mega-cap stock. Its 32.9% return in 2026 and 51.2% one-year return show that a more balanced structure can still participate meaningfully in the theme.

However, it charges 0.68%, while assets and trading volume are considerably lower than those of AIQ or ARTY. Its bid-ask spread has at times been much wider than those of the larger funds.

Our view: THNQ’s portfolio is more thoughtful than its size suggests, but WTAI and ARTY make the buying decision difficult.

For long-term investors, the higher fee is manageable. For investors trading smaller or less liquid sessions, spreads can matter more than the annual expense ratio. Always use a limit order.

Which AI ETF Is Best for Your Investment Strategy?

There is no universally best AI ETF because these funds are designed to perform different jobs.

Our overall verdict is straightforward.

ARTY is the strongest one-fund choice for investors who specifically want a dedicated AI ETF. It is not the cheapest and not the least volatile, but it offers the best combination of theme relevance, construction, cost and breadth.

WTAI is the best value choice. Its 0.45% fee, respectable performance and balanced portfolio make it difficult to ignore.

CHAT is the best active choice. It earns consideration because its mandate can move toward emerging AI beneficiaries that rules-based indices may include only after their valuations have risen.

AIS and BOTZ are specialist satellites. AIS is for AI infrastructure and semiconductor conviction. BOTZ is for physical automation and robotics. Neither should be confused with a neutral, all-purpose AI basket.

How Much Do ETF Fees Really Matter?

A difference of 0.30 percentage points sounds trivial. Over time, it compounds.

Assume an investor puts $10,000 into an ETF, the portfolio generates a 10% gross annual return, and the expense ratio is deducted every year.

The difference between a 0.45% fund and a 0.75% fund is approximately $674 over 10 years on an initial $10,000 investment under these assumptions.

That gap is not large enough to choose an inferior portfolio solely because it is cheaper. But the higher-cost fund must provide something valuable: better stock selection, more precise exposure, superior risk management or access to companies the cheaper fund misses.

The illustration excludes taxes, currency movements, trading costs and future contributions. Actual ETF expenses are accrued through NAV rather than deducted through a simple annual calculation.

The Overlap Problem: Are You Buying AI Twice?

Suppose an investor already owns QQQM. They already have substantial exposure to Nvidia, Microsoft, Amazon, Alphabet, Broadcom and Meta.

Buying AIQ, IVES, WTAI or CHAT may increase the weights of those businesses, but it does not necessarily create a new source of diversification.

This is not automatically wrong. An investor may intentionally want to overweight AI. The mistake is believing that owning two differently named ETFs means owning two independent strategies.

Before adding an AI ETF, compare:

• Its top 10 holdings with existing ETFs
• Total exposure to Nvidia and other repeated stocks
• Semiconductor and information-technology sector weights
• US versus international allocation
• The combined portfolio’s valuation and volatility

A broad index plus a specialist satellite often makes more sense than owning three overlapping AI ETFs.

For example, a Nasdaq-100 holding combined with AIS creates a clearer barbell between large technology platforms and AI infrastructure. Pairing the Nasdaq-100 with BOTZ adds physical automation and international industrial exposure. Adding AIQ may simply produce more of what the investor already owns.

The Five Risks That AI ETF Investors Can Easily Underestimate

1. AI adoption can rise while AI stocks fall: A powerful technology can still be a poor investment if expectations become too high.

If a stock is priced for years of exceptional growth, merely good earnings can disappoint the market. The technology may continue winning while the shareholder experiences weak returns.

2. Infrastructure spending is cyclical: Semiconductors, memory, servers and data-centre equipment can face shortages followed by overcapacity. AIS, IGPT and ARTY are particularly sensitive to this cycle.

The long-term AI trend can remain intact even as these funds experience deep corrections.

3. Theme definitions can drift: An index provider can include companies that use AI but do not earn meaningful AI revenue. This increases diversification but reduces purity.

Investors should examine holdings, not marketing descriptions.

4. Concentration works in both directions: BOTZ and IGPT have close to 60% of assets in their top 10 positions. That helped high-performing stocks matter, but it also increases company-specific and industry-specific risk.

ROBT sits at the other extreme. Its broad diversification reduces single-stock exposure, but can dilute successful companies.

5. Liquidity can quietly reduce returns: The expense ratio is visible. The bid-ask spread is less visible. A 0.50% spread paid when buying and again when selling can outweigh several years of a small expense-ratio advantage. This is especially relevant for smaller funds such as THNQ.

Limit orders are generally more sensible than market orders when trading thematic ETFs with lower volumes.

What to Monitor After Investing in an AI ETF

The correct monitoring checklist is not “Did the ETF beat the market this month?” Instead, review the following at least every six months:

1. Has the portfolio’s source of AI exposure changed? A software fund can gradually become a semiconductor fund after price movements or rebalancing.
2. Has top-10 concentration increased sharply? The portfolio may now carry more single-stock risk than when it was purchased.
3. Are revenue and earnings catching up with valuations? Rising prices without rising business fundamentals increase fragility.
4. Is the fund outperforming the correct benchmark? Compare AIS with semiconductor ETFs, BOTZ with industrial-automation funds and broad AI ETFs with QQQM. The S&P 500 is not always the most informative benchmark.
5. Has the fund’s expense ratio or bid-ask spread changed? A small fund may become more efficient as assets grow, or more difficult to trade if interest disappears.
6. Is the ETF still adding something to the total portfolio? If the investor’s other funds now contain the same companies, the AI ETF may have become redundant.

Bottom Line: Which AI ETFs Stand Out?

The AI ETF market is not one category. It is at least four categories hiding behind the same label:

• Semiconductor and AI infrastructure funds
• Mega-cap platform funds
• AI software and application funds
• Robotics and physical-automation funds

The largest fund, AIQ, is not necessarily the purest. The best recent performer, AIS, is not necessarily the best core holding. The most diversified fund, ROBT, has not delivered the strongest returns. The most concentrated portfolios can produce spectacular gains, but they can also subject investors to unusually deep corrections.

ARTY stands out as the best-balanced passive AI ETF. WTAI stands out on value. CHAT stands out for active management. AIS and BOTZ are the strongest specialist tools for investors with a specific infrastructure or robotics thesis.

The real edge is not finding an ETF with “AI” in its name. It is understanding which part of the AI economy the fund owns, what you are paying for that exposure, and whether you already own the same trade somewhere else.