VanEck Semiconductor ETF surges 82% as equipment makers lead sector rotation

The VanEck Semiconductor ETF's 82.1% first-half gain masks a sector rotation: semiconductor equipment makers outpaced chip designers like Nvidia, signaling a structural shift in AI spending toward inference infrastructure.

The VanEck Semiconductor ETF (SMH) returned 82.1% in the first half of 2026, according to S&P Global Market Intelligence, but the headline number hides a critical divergence. Semiconductor equipment makers — the companies that build the machines that fabricate chips — have massively outperformed chip designers like Nvidia, whose 10% capped weighting in the underlying MVIS index actually helped the ETF's performance.

"The nature of AI spending is shifting from training to inference, and that changes which parts of the semiconductor stack benefit," said a semiconductor supply chain analyst. "Equipment makers are seeing orders accelerate as foundries race to add capacity for AI-specific chips, while GPU demand growth is decelerating off a massive base."

The MVIS U.S. Listed Semiconductor 10% Capped Screened Index, which SMH tracks, caps any single holding at 10% and rebalances quarterly. That structure proved advantageous: Nvidia's market cap stood at 3.7 times Micron Technology's and 5.1 times Advanced Micro Devices' as of June 30, yet its ETF weighting was limited. Nvidia significantly underperformed the ETF in the first half, meaning the cap protected investors from single-stock concentration risk.

Two themes drove the rotation. First, AI capital spending has exceeded expectations, forcing semiconductor companies to ramp up their own equipment purchases. Second, spending is shifting toward inference — the process of running trained AI models — which favors central processing units over graphics processing units. Intel has been the primary beneficiary of that CPU demand surge, outperforming the broader sector.

Equipment Makers Capture the AI Capacity Buildout

The outperformance of semiconductor equipment names reflects a simple dynamic: hyperscalers like Microsoft, Amazon, and Google are committing tens of billions to AI data centers, and those facilities need chips. Building more chips requires more fabrication equipment, and companies like Applied Materials, ASML, and Lam Research are the primary suppliers.

Nvidia's own results underscore the scale. The company reported Q1 FY2027 revenue of $81.61 billion, up 85% year over year, with data center revenue of $75.25 billion and data center networking up 199%. Its non-GAAP gross margin expanded to 75%, and management guided Q2 revenue to $91 billion. Yet the stock trades at 23 times forward earnings — a discount to SanDisk, a commodity memory maker, at 27 times.

That valuation gap has drawn attention from prominent investors. Jim Cramer called Nvidia "the most proprietary chip company in the history of the world" on his July 9 broadcast, arguing the market is mispricing its software moat — CUDA, NVLink Fusion, and the Dynamo inference stack — relative to commodity chipmakers. Nvidia carries 58 buy ratings, 2 holds, and 1 sell, with an average analyst price target of $301.62, implying roughly 44% upside from its current $209.79 level.

For investors, the rotation raises a portfolio question. The SMH ETF's diversification across equipment makers and chip designers captured both sides of the AI trade in the first half. But if inference spending continues to gain share relative to training, the winners may shift further toward CPU-focused names like Intel and equipment suppliers — and away from pure GPU plays. Nvidia's 2.21 beta and history of sharp drawdowns suggest position sizing remains critical, even for believers in the long-term AI thesis.

This article is for informational purposes only and does not constitute investment advice.