Buying AI ETFs? You’re Probably Mostly Buying NVIDIA

The headline is blunt for a reason. Retail investors who buy AI exchange-traded funds often expect broad exposure to the whole machine-learning story. In practice, however, a handful of infrastructure winners — led by Nvidia — can make up a surprisingly large slice of those funds.

Why this matters

Funds that call themselves AI or robotics are assembled in different ways: some screen by AI-derived revenue, others rely on self-reported exposure, and some simply weight holdings by market cap. Across all those methods two things tend to repeat.

• Infrastructure concentrates value. GPUs, datacenter chips and cloud APIs require huge capital and, more often than not, one or two suppliers end up dominating training and inference. When that happens, their market caps balloon and the ETFs that track market-cap weights follow suit.
• Cap-weighting breeds concentration. Heavy weights aren’t a mistake; they’re a predictable consequence of indexing by market cap. So buying an AI ETF can feel a lot like buying one or two mega-caps.

What to look for, concretely

Top holdings you’ll see again and again include Nvidia (NVDA), Microsoft (MSFT), Alphabet (GOOG) and Amazon (AMZN). Many popular AI funds put single-stock weights in the low double digits. Expense ratio and the fund’s construction matter a great deal — is it a thematic bet, a revenue-screened basket, or just a tech-heavy index rebranded as AI? And don’t forget overlap risk: owning several AI funds or a general tech ETF can create hidden redundancy.

A little history helps

This pattern isn’t new. In the late 1990s tech-themed products stood in for internet exposure and ended up concentrated in a few names before the bubble burst. The 2020s are different in the specifics — chipmakers and cloud platforms instead of dial-up vendors — but the mechanics are disturbingly similar. The label can mask single-stock risk.

Why concentration might make sense

There’s a defensible case for the tilt. If Nvidia and a handful of cloud providers are literally building the plumbing for modern AI, their profit pools and pricing power could be both large and persistent. If you believe the market is pricing that dominance efficiently, a cap-weighted bias is just the market sending a signal. Still, believe what you will; in practice the story is messier than the textbook version.

A practical allocation checklist

• Look at the top 10 holdings and each weight. If one name is north of 10–15% and you don’t want that exposure, rethink the trade.
• Compare methodologies: revenue thresholds, screening rules, index construction. They change outcomes.
• If you want to avoid mega-cap bias, consider equal-weight or factor-adjusted AI ETFs.
• If you have conviction, owning the single stock directly can be cleaner — but size it carefully.
• Check turnover and tax efficiency. Thematic funds can churn more than broad indexes, which matters for after-tax returns.

Smarter alternatives

You can buy the leader explicitly. If you trust Nvidia’s moat, own NVDA and size the position consciously. Or use an equal-weight AI or sector fund to mute market-cap skew. Mix in active managers who actually dig into AI value chains instead of chasing headlines — some do useful work here.

Final thought

AI looks structural and long-lasting, but the wrapper matters. Don’t assume an AI ETF equals a diversified bet across a thousand startups. Peek under the hood: you might discover your exposure is really to a few industrial-scale firms that control the levers of future models — and that should change how you size the position.

Pedro Marini