A practical way to think about artificial intelligence (AI) investing is to break down the value chain and identify bottlenecks, where demand exceeds available capacity.
These imbalances matter because companies controlling the constrained part of the ecosystem often gain pricing power and experience surging demand.
If you’re an investor, you probably know by now about several bottlenecks emerging from the AI boom. Electricity demand, for example, has turned utility stocks into indirect AI plays because they generate and transmit the electricity that powers datacenters.
Sign up for Kiplinger’s Free Newsletters
Profit and prosper with the best of expert advice on investing, taxes, retirement, personal finance and more – straight to your e-mail.
Profit and prosper with the best of expert advice – straight to your e-mail.
Data center demand leads data center supply right now, too, a trade expressed through certain real estate investment trusts (REITs) that own and lease these facilities to hyperscalers and enterprise customers.
The most obvious bottleneck, however, remains semiconductors. Modern computing is built on them, and AI systems in particular require enormous quantities of advanced chips capable of handling parallel processing and large-scale data movement.
That said, the semiconductor industry itself is highly diverse. Some companies design chips, others manufacture them, while others focus on fabrication equipment, memory, networking or analog components.
Depending on which type of semiconductor company you own, you may be exposed to very different risks, ranging from geopolitical tensions and manufacturing bottlenecks to cyclical swings in consumer electronics demand.
If you’re simply looking to capitalize on the broad drivers behind AI infrastructure spending, semiconductor ETFs can simplify the process of establishing more concentrated but still efficient exposure to chipmakers.
The types of semiconductor stocks
The semiconductor industry is a broad category with big variations between business models, risks and sources of revenue.
One of the most recognizable groups are chip designers, which develop architecture and intellectual property but often outsource manufacturing to third-party foundries. Examples include Nvidia (NVDA), Advanced Micro Devices (AMD) and Broadcom (AVGO).
Fabrication companies, or “fabs,” manufacture chips at extraordinarily expensive and technology demanding foundries. Taiwan Semiconductor Manufacturing (TSM) is the dominant pure-play foundry, while Intel (INTC) both designs and manufactures chips through its integrated model.
Memory semiconductor companies specialize in products such as DRAM and NAND flash memory, critical for servers, AI accelerators, smartphones and data centers. Examples include SK Hynix and Micron Technology (MU).
Semiconductor equipment companies, which make the specialized tools and machinery needed to fabricate chips, sit even higher up the value chain. The standout company here is ASML Holding (ASML).
How we chose the best semiconductor ETFs
We began by excluding ETFs tied to single stocks, including leveraged long and inverse ETFs linked to companies such as Nvidia, along with covered call income variants advertising double-digit yields.
These products are ultimately tied to the fortunes of one company rather than the semiconductor industry, they cost more, they’re more volatile and they’re more complex, making them less suitable for the average long-term investor.
We also excluded leveraged, inverse and income-oriented semiconductor ETFs tied to the broader industry itself because of concerns around volatility, leverage decay and elevated fees.
That screening process narrowed the universe down to traditional semiconductor ETFs, either actively managed or index-tracking. From there, we imposed a maximum expense ratio of 0.35%.
We also applied several practical filters designed to improve the investor experience. First, we required ETFs to hold at least $100 million in assets under management.
Second, we screened for liquidity by focusing on each ETF’s 30-day median bid-ask spread. To qualify, spreads needed to remain at or below 0.15%. This helps reduce implicit transaction costs when buying or selling shares.
The end result was a smaller group of semiconductor ETFs that combined broad industry exposure, reasonable fees, sufficient scale and adequate trading liquidity.
VanEck Semiconductor ETF
(Image credit: Getty Images)
- Assets under management: $67.9 billion
- Expense ratio: 0.35%
- 30-day median bid-ask spread: 0.02%
- One-year annual return: 137.4%
By assets under management, the VanEck Semiconductor ETF (SMH) is the largest and most widely used semiconductor ETF on the market. Since debuting in December 2011, it’s also been one of the strongest-performing ETFs, including a 36.9% annualized total return over the trailing 10 years.
SMH tracks the MVIS US Listed Semiconductor 25 Index, which focuses on the largest and most liquid semiconductor companies. The portfolio uses a market cap-weighted methodology, meaning larger companies receive larger allocations.
Historically, this approach has worked well for SMH. By allowing dominant large-cap stocks to appreciate without aggressively capping their weights, the ETF has fully benefited from the outsized gains generated by the industry’s biggest winners.
That dynamic is most visible with Nvidia. The leader of the AI revolution accounts for more than 15% of SMH’s portfolio. While this concentration has materially boosted returns, it also creates a meaningful degree of single-stock risk for new investors entering today.
Risk is another important consideration. Relative to the S&P 500, SMH carries a three-year beta of 1.93 and a three-year standard deviation of 30.5%. In practical terms, that means the ETF has historically been substantially more volatile than the broader stock market.
Still, for investors primarily seeking exposure to the largest and most influential semiconductor companies, and who are comfortable with a sizable NVDA allocation, SMH remains one of the purest and most liquid ways to express that view.
Learn more about SMH at the VanEck provider site.
VanEck Fabless Semiconductor ETF
(Image credit: Getty Images)
- Assets under management: $280.5 million
- Expense ratio: 0.35%
- 30-day median bid-ask spread: 0.10%
- One-year annual return: 118.5%
The VanEck Fabless Semiconductor ETF (SMHX) exists because some semiconductor investors are uncomfortable with fabrication-related supply chain and geopolitical risk. Much of this concern is about Taiwan Semiconductor Manufacturing.
Taiwan’s political status remains disputed under China’s reunification policy, and investors periodically worry that geopolitical tensions in the region could disrupt semiconductor supply chains. Since many leading chip designers depend heavily on TSM fabrication capacity, this concentration risk has become an increasingly important consideration for the industry.
Fabrication businesses also tend to be highly asset intensive. Building and maintaining fabs requires enormous upfront capital expenditures, ongoing equipment upgrades and substantial energy and water usage. This can create margin pressure during downturns and increase operational risk.
To address investor demand for fabless exposure, SMHX focuses specifically on semiconductor companies that design chips but outsource manufacturing.
Compared to SMH, the portfolio composition looks materially different. According to ETF Research Center data, the overlap between SMHX and SMH is roughly 45% by weight across 10 holdings.
That means investors are either underweight or entirely excluding companies such as TSM, Intel, Micron, Texas Instruments (TXN), Analog Devices (ADI) and Lam Research (LRCX). In exchange, the ETF leans more heavily into designers and intellectual property-focused firms such as Broadcom, Arm Holdings (ARM) and Lattice Semiconductor (LSCC).
Since launch, SMHX has trailed SMH because fabrication-related companies have generally outperformed during the recent AI-driven semiconductor boom. Investors should still expect concentration risk, however. Nvidia remains roughly 13.9% of the ETF’s assets, followed by Broadcom at 10.8%.
Learn more about SMHX at the VanEck provider site.
iShares Semiconductor ETF
(Image credit: Getty Images)
- Assets under management: $39.0 billion
- Expense ratio: 0.35%
- 30-day median bid-ask spread: 0.02%
- One-year return: 164.6%
Most semiconductor ETFs are going to have substantial overlap because the industry itself is only so large. However, the weightings can differ materially depending on the benchmark index each ETF tracks.
That distinction becomes very apparent with the iShares Semiconductor ETF (SOXX), which competes directly against SMH while charging the exact same 0.35% expense ratio. SOXX is actually the older product, debuting in July 2001, though it remains somewhat smaller than SMH despite still managing tens of billions in assets.
The ETF tracks the NYSE Semiconductor Index, which holds 30 U.S.-listed semiconductor companies screened for liquidity and size. The major difference comes down to weighting methodology. Compared to SMH, SOXX spreads allocations more evenly across its largest holdings.
Instead of a very large overweight to Nvidia, the top positions are distributed more broadly among companies such as Micron Technology, Advanced Micro Devices, Broadcom, Intel and Nvidia itself. The portfolio is still concentrated overall, but the weighting structure reduces the risk of a single company completely dominating performance.
Another reason some investors may prefer SOXX relates to tax-loss harvesting. Because SOXX and SMH have similar historical performance, comparable fees and overlapping holdings, investors can potentially swap between them to realize capital losses without materially changing exposure.
Importantly, because the ETFs track different indices, an investor could theoretically sell one and immediately purchase the other without triggering the wash sale rule, avoiding the need to wait 31 days out of the market.
Learn more about SOXX at the iShares provider site.
State Street SPDR S&P Semiconductor ETF
(Image credit: Getty Images)
- Assets under management: $3.4 billion
- Expense ratio: 0.35%
- 30-day median bid-ask spread: 0.15%
- One-year total return: 154.2%
The preceding semiconductor ETFs all use some variation of market cap- weighting, sometimes with additional adjustments for liquidity and free float. That approach has historically worked well in semiconductors because the industry exhibits a very large degree of performance dispersion.
Being able to identify the winner and allow it to compound into a larger position can materially boost returns. The downside is concentration risk.
Investors entering near the later stages of a semiconductor bull market can find themselves heavily exposed to a small number of companies. During downturns, that same concentration can amplify losses. A more balanced approach can be expressed through the State Street SPDR S&P Semiconductor ETF (XSD).
This ETF tracks the S&P Semiconductor Select Industry Index, an equal-weighted benchmark currently composed of 44 holdings. Equal weighting means each company receives roughly the same allocation regardless of whether it’s a small-cap, mid-cap or large-cap semiconductor stock.
The portfolio is rebalanced quarterly, which systematically trims outperformers and reallocates capital toward laggards. In practice, this creates a “sell high, buy low” mechanism, and XSD’s largest holdings are often the hottest chip stocks.
Importantly, however, those weights are not permanent. At the next quarterly rebalance, stronger-performing holdings will be trimmed back toward equal allocations, while laggards will be added to.
This structure can appeal to investors seeking lower concentration risk and broader exposure across the semiconductor ecosystem. The trade-off is that it limits the ability of major winners to compound into outsized positions.
Over the trailing 10-year period, XSD still delivered an impressive 29.8% annualized total return, though it lagged more concentrated ETFs such as SMH because the largest winners weren’t allowed to run.
Still, past performance doesn’t predict future results. For investors more concerned about excessive dependence on a handful of mega-cap chipmakers, XSD may be a more balanced way to gain semiconductor exposure.
Learn more about XSD at the State Street Investment Management provider site.
Invesco PHLX Semiconductor ETF
(Image credit: Getty Images)
- Assets under management: $2.5 billion
- Expense ratio: 0.19%
- 30-day median bid-ask spread: 0.03%
- One-year total return: 157.4%
Investors may have noticed that most of the previous semiconductor ETFs charge a 0.35% expense ratio. That’s largely a function of how industry-specific ETFs have historically been priced.
A major reason for this is the influence of State Street Investment Management, which built one of the largest lineups of industry ETFs in the market. Over time, 0.35% effectively became the standard pricing level investors grew accustomed to paying for specialized industry exposure.
Some ETF providers, however, have attempted to compete more aggressively on fees. In semiconductors, one of the lowest-cost options currently available is the Invesco PHLX Semiconductor ETF (SOXQ), which charges a 0.19% expense ratio.
For a $10,000 investment, that translates into roughly $19 annually in fee drag versus approximately $35 for many competing semiconductor ETFs.
SOXQ tracks the PHLX Semiconductor Sector Index, which holds 30 of the largest U.S.-listed semiconductor companies. The index is rebalanced quarterly in March, June, September and December, with a full annual reconstitution each September.
Rebalancing and reconstitution are related but different processes. Rebalancing adjusts the weights of existing holdings back toward the index methodology, while reconstitution determines which companies are added to or removed from the index entirely based on eligibility rules.
Like SOXX, SOXQ is materially less concentrated than SMH, but it achieves this without relying on the equal-weighted structure used by XSD. The portfolio’s largest holdings include Nvidia, Micron Technology, Broadcom and Intel, with each remaining at or below roughly a 10% portfolio allocation.
Learn more about SOXQ at the Invesco provider site.
