AI ETFs: Fees, Rebalances, and Why Tracking Isn't Exact
The Global X Artificial Intelligence & Technology ETF (AIQ) reported $10.46 billion in net assets and a NAV of $66.07 on September 21, 2026. NAV, or net asset value, is the per-share value of the fund. The fund launched on May 11, 2018 and charges a total expense ratio of 0.68%. That is $6.80 per year for every $1,000 invested. Global X
The Invesco AI and Next Gen Software ETF held 101 stocks on September 18, 2026. Its management fee is 0.50%. The ETF and its index are rebalanced in March, June, September and December. That reset happens four times per year. Invesco
The Amplify Bloomberg AI Equal Weight ETF (AIVC) seeks investment results that generally correlate, before fees and expenses, to the total return performance of its benchmark index. Amplify The WisdomTree Artificial Intelligence and Innovation Fund seeks to track the price and yield performance, before fees and expenses, of its benchmark index. WisdomTree
Systematic stock picking by AI has also been tested outside ETFs. MarketWatch reported on June 17, 2023 that an AI-powered stock portfolio beat the S&P 500 and left market professionals behind. MarketWatch On January 31, 2024, MarketWatch profiled a hedge fund trying to learn whether AI and a supercomputer can beat the markets. MarketWatch
The broader context here is maintenance. What counts as AI changes fast. A March, June, September and December rebalance forces discipline. The portfolio must recheck what still qualifies as AI and next-gen software. That limits drift away from the theme. It also creates turnover, or frequent buying and selling. Turnover brings trading costs and small gaps versus the index, called tracking difference. For taxable accounts, it can bring payouts that trigger taxes. The fund you own in December can look different from January. Size matters too. A multi-billion dollar fund must create and redeem shares efficiently while trading around those changes.
Looking at what this means for implementation, index language matters. Promises to correlate to total return or track price and yield, both measured before fees and expenses, are not promises to match exactly. Fees build up daily. Company actions, idle cash, sampling and the timing of rebalances all feed tracking difference. Equal weighting works differently from weighting by company size. It cuts the sway of the largest stocks. It gives more weight to smaller members and needs more trading to reset those equal sizes. A 101-stock list spreads single-stock risk better than a concentrated portfolio. It still leaves shared risk. If model training, cloud, chips, software and services move together, owning many tickers gives less shelter than the count suggests.
In my view, the test that matters is net persistence. Any gross edge, from a supercomputer, from analysts aided by machine learning, or from a rules-based index, must survive fees, turnover and index rules before it reaches you. A 0.68% total expense ratio and a 0.50% management fee are different disclosures. The first is all-in. The second leaves out other fund costs. Neither includes trading costs or the cost of quarterly changes. Past periods of beating the market do not settle that math. They leave the question to watch: how much of any AI-driven edge remains after the work needed to stay on theme.
