It’s quite understandable for some investors to feel fatigued from a seemingly never-ending buzz about AI. Dotcom era-like obsession of media with everything related to “artificial intelligence” has already led to hyped expectations in the tech sector. This April, however, we clearly see a hangover, with many AI-related stocks falling by -10-20% off highs.
The VanEck Semiconductor ETF (NASDAQ:SMH) represents a relatively risky and concentrated bet on semiconductor companies that are essential for the ongoing AI trend.
Even though the risk profile has become less favorable for the sector, I still think there are enough catalysts to keep pushing AI stocks up. The recent correction in semiconductor stocks should be used as a buying opportunity. Thus, I give the SMH a “Buy” rating.
SMH ETF Overview
According to the Fund description, the VanEck Semiconductor ETF is a passively managed ETF that seeks to replicate as closely as possible the MVIS US Listed Semiconductor 25 Index. The holdings structure strongly favors Nvidia (NVDA) with a weight of 20.04% of the overall SMH’s holdings. The key partner of Nvidia, Taiwan Semiconductor Manufacturing Company (TSM) comes second with a 12.53% weight.
Compared to key peer semiconductor ETFs, the SMH has the highest assets under management (AUM) of $17.51 billion, which ensures high liquidity for investors.
Compared to peers, the SMH ETF has the most concentrated holdings structure, with 26 companies included. In this regard, iShares Semiconductor ETF (SOXX) and SPDR S&P Semiconductor ETF (XSD) have a considerably higher diversification profile, with 35 and 40 holdings respectively.
When it comes to performance, the high portfolio concentration of the SMH ETF fully manifests itself. The 1-month performance of SMH is the worst compared to peers, while the 10-year performance is the best. Again, thanks to Nvidia, that is the major contributor to the SMH’s returns as well as volatility.
Overall, I’d characterize the SMH ETF as an aggressive bet on the semiconductor sector. The SOXX ETF has a more moderate approach to portfolio composition, with high-single-digit weights for top-10 holdings. At the same time, the XSD has a radically different holdings composition with an equal-weight structure, making it a safer long-term bet in terms of holdings diversification.
More AI-Related Catalysts To Feed The Hype
Over the last month, it could seem that the AI hype started to show signs of wear. Let me provide some of the key AI-related catalysts and trends:
1. OpenAI may release GPT-4.5 and its text-to-video model, Sora, later this year. In recent months, there have been a lot of speculations around a possible release date of GPT-5, the new generation of the large language model (LLM) from OpenAI. However, more realistic estimates suggest that the next release this year will be GPT-4.5, and not GPT-5. Nevertheless, any meaningful improvements over GPT-4 may be enough to spark another wave of AI hype. Moreover, later this year OpenAI is expected to release its text-to-video model, Sora, which may revolutionize video content creation.
2. Meta boosts investments in AI. The company noted that the capital expenditures related to AI could total $40 billion this year, up to $5 billion more than Meta estimated in October. Even though the company seems to struggle with proper monetization of its investments in AI, a higher allocation of funds is still beneficial for semiconductor companies.
3. A possible Google-Apple partnership may drive demand for AI chips even further. According to media reports, Apple (NASDAQ:AAPL) is considering partnering with Google to use Google’s model, Gemini, to power AI-generative features in Apple products. While these talks are preliminary, such a deal would ultimately mean that millions of Apple devices will create stable demand for AI-related computing power, which is bullish for semis stocks.
4. Training of new large language models (LLM) is getting costlier. Dario Amodel (CEO of AI company Anthropic) gave an interview recently where he shared his thoughts on further scaling of LLMs. According to Dario, current LLMs cost around $100 million to train, the models training right now cost $1 billion, and the ones training in 2025-2026 may cost up to $5-10 billion. This is bullish for the semiconductor sector, as demand for computing power for LLM training may increase dramatically.
Nvidia Earnings In Focus
The majority of the SMH ETF’s top holdings have already reported their Q1 earnings, and Nvidia’s report, the most important report for the sector, will be announced on May 22, 2024.
Just as it happened in the past year, Nvidia’s earnings may affect the market dynamic of the whole semiconductor sector, at least for another quarter. At the same time, so far, the results of the majority of semiconductor companies were pretty good, which gives some hope that Nvidia may exceed investors’ expectations once again.
The Bottom Line
Expect lots of volatility during this US stock market earnings season, with Nvidia in primary focus in late May to define the further trajectory of the whole semiconductor sector. When picking a semis ETF, investors should be aware the SMH ETF has the highest share of Nvidia in its holdings compared to other peer ETFs. Therefore, consider the SMH as a bet primarily on Nvidia, with limited diversification across the sector.