Semiconductor ETFs are extending their AI-driven rally as investors broaden their exposure beyond graphics processors and into the memory infrastructure powering next-generation AI systems.
The VanEck Semiconductor ETF (NASDAQ:SMH) climbed to new highs on Tuesday, and is on track for the strongest quarterly performance on record, fueled by continued enthusiasm around AI infrastructure spending.
The fund has climbed roughly 57% in the second quarter and is up more than 65% year-to-date, extending a multiyear rally that accelerated after the launch of ChatGPT in late 2022 ignited the generative AI boom. This advance was supported by strong inflows, with the fund taking in around $5.6 billion this year so far, per ETFDb.
The BlackRock-backed iShares Semiconductor ETF (NASDAQ:SOXX) has also posted outsized gains as investors continue pouring money into chip-focused funds tied to artificial intelligence, data centers and hyperscaler spending. Year-to-date, SOXX saw about $3.4 billion in inflows. The fund prices also surged more than 6%, hitting an all time high on Tuesday.
Recent institutional positioning has reinforced that trend. Latest 13F filings showed billionaire investor Dan Loeb initiated a sizable new position in the SMH through his hedge fund Third Point, signaling continued conviction in the broader AI semiconductor trade rather than just individual chip names.
While NVDA remains the dominant driver across many semiconductor ETFs, investors are increasingly rotating toward other parts of the AI hardware supply chain, especially memory-chip makers benefiting from surging demand for high-bandwidth memory, or HBM, used in AI servers and accelerators.
The Micron Turning Point
That shift came into sharper focus Tuesday after shares of Micron Technology Inc (NASDAQ:MU) surged more than 16% after UBS analyst Timothy Arcuri more than tripled his price target on the memory-chip maker to $1,625 from $535, arguing that the AI memory market is undergoing a structural transformation rather than another traditional semiconductor cycle.
The call is notable because Micron has historically traded as a deeply cyclical DRAM and NAND producer, vulnerable to sharp swings in pricing and demand. UBS now argues that the rise of high-bandwidth memory, or HBM, used in AI accelerators and data-center infrastructure is fundamentally changing that dynamic.
According to UBS, newer long-term agreements in the memory market now include multi-year durations, fixed-volume commitments and partially fixed pricing structures, which is a major departure from older volume-based arrangements that often amplified volatility across the chip cycle.
The firm now expects Micron to generate earnings per share of $155 in 2027, $167 in 2028 and $117 in 2029, while projecting EPS to remain above $100 through the period. UBS said the market could begin assigning Micron a more "normal" valuation multiple as investors increasingly view the company as a structural AI infrastructure beneficiary rather than a commodity memory supplier.
Semiconductor ETFs Could See Broader AI Participation
That shift is also reverberating across semiconductor and AI-focused ETFs.
One of the biggest potential beneficiaries is the VanEck Semiconductor ETF (NASDAQ:SMH), which holds significant exposure to Micron alongside AI chip leaders including Nvidia, Broadcom, Inc (NASDAQ:AVGO) and Taiwan Semiconductor Manufacturing (NYSE:TSM). The fund has been one of the biggest winners of the AI trade over the past two years, but Micron's latest surge suggests performance leadership inside semiconductor ETFs may now be broadening beyond GPUs.
The BlackRock-backed iShares Semiconductor ETF (NASDAQ:SOXX) is also positioned to benefit from the memory-driven AI boom. While Nvidia remains a dominant weighting across semiconductor ETFs, Micron's nearly 180% year-to-date rally is increasingly making the company a meaningful contributor to returns.
Investors are also increasingly watching ETFs tied to AI infrastructure, hyperscaler spending and data-center supply chains as memory demand accelerates. High-bandwidth memory chips have become essential for training and deploying advanced AI models, creating a powerful tailwind for suppliers across the semiconductor ecosystem.
The broader implication for ETF investors is that AI infrastructure spending is becoming far more diversified. Many semiconductor ETFs became heavily dependent on Nvidia's outsized gains during the early stages of the AI rally. Micron's breakout could signal the beginning of a broader participation phase across the AI hardware ecosystem.
Goldman's $800 Billion AI Spending Forecast Adds Fuel
A new research note from Goldman Sachs estimates AI-related spending reached an annualized $650 billion in the first quarter and could exceed $800 billion by year-end. The bank raised its 2026 U.S. business investment growth forecast to 7.8% from 6.5%, citing continued investment across semiconductors, servers, memory storage, data centers, power infrastructure and software.
Goldman estimates AI spending alone could boost true capital expenditure growth by roughly 3.3 percentage points in 2026.
That trend could further support ETFs tied not only to AI compute leaders like Nvidia, but also to companies powering the memory backbone of generative AI systems.
Funds focused on semiconductor manufacturing, AI infrastructure and data-center supply chains could increasingly see Micron emerge as a core holding if HBM demand continues accelerating. Investors are also beginning to view memory chips as critical "picks-and-shovels" assets for AI expansion, especially as hyperscalers race to deploy larger AI clusters requiring massive memory capacity.
Still, the trade carries risks.
UBS warned that weakening HBM demand could send Micron shares down to $250, implying roughly 66% downside from Friday's close. That volatility risk could also spill over into semiconductor ETFs with elevated exposure to memory suppliers.
For now, however, Wall Street appears increasingly convinced that AI's next major investment wave may not be centered solely on processing power — but also on the memory infrastructure needed to sustain it.
Image: Shutterstock
Login to comment