Semiconductor ETFs: How to Invest in the Chips That Build Artificial Intelligence (VanEck SMH)

A semiconductor ETF lets you invest, through a single instrument, in the companies that make the chips artificial intelligence runs on. One of the best-known semiconductor ETFs is the VanEck Semiconductor ETF (SMH): it brings together the 25 largest and most liquid semiconductor companies listed in the U.S., spanning design, manufacturing, and equipment.

For AI to grow, three things have to grow together: more data centers, enough energy to power them (driven by AI, U.S. data center power demand is projected to more than double by 2027, from 31 to 66 gigawatts), and ever more powerful chips. If one falls behind, everything slows down.

It's worth pausing here, because understanding the "why" behind this wave matters. For decades, the fate of chips depended on how many PCs and phones were sold: a deeply cyclical business. Today the engine is different: sustained investment by Big Tech in AI infrastructure. In 2026 alone, Amazon, Microsoft, Alphabet, and Meta plan to invest close to $725 billion (most of it in AI data centers) and expect to spend even more in 2027. That spending pulled the trillion-dollar milestone forward by nearly four years: the industry is expected to reach around $975 billion in 2026, up 26% after 22% growth in 2025.

A key nuance: the growth is highly concentrated. AI chips account for roughly half of the industry's revenue but less than 0.2% of units shipped. The clearest sign of that tension is memory: the HBM that feeds AI accelerators is sold out for 2026, in what is already being described as a "supercycle" comparable to the 1990s, with DRAM prices rising 40-50% in the third quarter and no clear relief before 2028. The uncomfortable side effect: as capacity shifted toward HBM, conventional memory became scarce, making phones and computers more expensive.

There is an even deeper shift underway. Training a model happens only a few times; using it (inference) happens billions of times a day, and its cost could fall by up to two orders of magnitude. The counterintuitive part: making AI cheaper doesn't reduce chip demand - it multiplies it, because it makes profitable uses that aren't viable today. That is the engine behind VanEck's thesis of a structurally less cyclical sector, further supported by policies such as the CHIPS Act.

A semiconductor fund cannot sidestep politics. The U.S. authorized Nvidia to sell its H200 chip to approved Chinese customers, but with an unprecedented condition: 25% of the revenue goes to the U.S. Treasury (the newest chips, Blackwell and Rubin, remain restricted). And the most advanced chips are made essentially by a single company, Taiwan's TSMC, whose advanced packaging capacity (CoWoS) is running at its limit: much of the world's AI rests on one island. Between the CHIPS Act, tariffs, and rare earths, making the supply chain more agile is now the number-one priority for industry executives. The SMH is exposed to all of this at once: that is part of its appeal - and of its risk.

The underlying reason to look at the SMH is simple, and VanEck states it plainly: within the industrial system of AI, semiconductors come first (everything, ultimately, runs on chips). Because advanced manufacturing is in very few hands, that layer enjoys scarcity, pricing power, and high barriers to entry: the conditions for long-term demand. That is the logic of owning the entire layer in one instrument rather than guessing the winning company. As of publication, the SMH had gained more than 100% over the previous 12 months (total return); for reference, the sector index (SOX) was coming off its best quarter on record, with a gain of nearly 69% in two months.

The SMH is a thematic, concentrated investment in a single sector. Many investors combine it with broad, diversified funds, such as an S&P 500 ETF, to balance the risk. You can compare it using our Compare ETFs tool or our guide to the Best AI ETFs (2026). As always at El Fondo: we don't tell you what to do; we help you understand how it works.

How much does it cost, and can I buy it from Latin America? Its annual cost is 0.35%. It is domiciled in the U.S. and can be purchased through a broker or platform that provides access to the U.S. stock market; consider currency risk and U.S. dividend taxation.