Artificial intelligence (AI) is driving one of the largest infrastructure investment cycles the technology sector has ever seen. Unsurprisingly, investors are treating it as a significant long-term theme, and not just a passing trend. In fact, according to The Motley Fool's 2026 AI Investor Outlook Report, 9 out of 10 existing AI investors plan to maintain or increase AI exposure in the next 12 months. Nearly 62% of the surveyed respondents believe companies investing heavily in AI will deliver strong long-term returns, despite increasing valuation concerns. The trends in enterprise AI spending further reinforce that optimism.
Image source: Getty Images.
Hyperscalers are investing heavily in GPUs, advanced logic and memory chips, as well as networking and other equipment for building and expanding AI data centers. According to Goldman Sachs, the consensus Wall Street estimate for global AI-related data center capital expenditures in 2026 is $527 billion.
Those capital expenditures primarily benefit a small set of semiconductor companies, including chip designers, foundries, equipment manufacturers, and memory suppliers, which are heavily concentrated within the VanEck Semiconductor ETF (SMH 3.22%).
Here's why SMH can prove to be a core AI investment in 2026.
Concentrated exposure to AI leaders
SMH's top five holdings are Nvidia, Taiwan Semiconductor Manufacturing, Broadcom, Micron Technology, and Advanced Micro Devices, which together account for close to 49.8% of its assets, while the top 10 holdings, which also include ASML, Lam Research, KLA, Texas Instruments, and Qualcomm account for over 73% of the total assets. However, this concentration is not a risk but an advantage, as SMH investors benefit from exposure to dominant AI players while avoiding reliance on a single business model.
Robust performance
SMH delivered a roughly 49% gain in 2025, comfortably outpacing the 16.4% return of the benchmark S&P 500 (^GSPC 1.52%) index.
NASDAQ: SMH
Key Data Points
The ETF also has a long-term track record of exceptional performance, with annualized returns of around 30.9% over the past decade (as of Jan. 8, 2026), which is significantly higher than the S&P 500's 12.9% annualized return in the same time frame. Hence, it is evident that the ETF has proved its mettle through multiple semiconductor cycles, and not just during the recent period of AI-driven investor enthusiasm.
Inference opportunity
Although the early AI infrastructure build-out was primarily focused on training large language models, the next phase is now increasingly driven by inference workloads, which involve running these AI models across real-world applications. According to Deloitte, inference is expected to account for two-thirds of total AI compute demand by 2026, up from one-third in 2023. Unlike training workloads, which are episodic and front-loaded, inference workloads are durable and can scale with usage and increasing AI adoption across use cases. This implies that the demand for GPUs, memory chips, networking components, and other power-efficient hardware is long-term and sustained, which will further drive the growth of SMH.
Valuation
Despite the rally, SMH is currently trading at nearly 33 times trailing-12-month earnings, which is in line with the price-to-earnings (PE) multiples of many large-cap tech stocks. Hence, SMH is a suitable choice for investors seeking exposure to AI without selecting individual stocks.
