Nvidia (NVDA) vs Semiconductor ETF (SMH)

NVDA (Nvidia Corporation) is a single-stock position. Nvidia designs GPUs (graphics processing units) and related data center hardware, with its H100, H200, Blackwell B100/B200, and Grace Hopper platforms powering the majority of AI training and inference workloads globally. As of late 2025, Nvidia held approximately 86 percent market share in AI-specific GPUs. The company derives approximately 91 percent of revenue from data centers.

SMH is the VanEck Semiconductor ETF tracking the MVIS US Listed Semiconductor 25 Index, launched December 20, 2011. It holds 25 of the largest US-listed semiconductor companies including Nvidia, TSMC (ADR), Broadcom, AMD, Qualcomm, ASML, Intel, Applied Materials, Micron, and others. Nvidia typically weights 18-22 percent of SMH, TSMC 12-14 percent, Broadcom 8-9 percent, and AMD 5-6 percent. Expense ratio 0.35 percent, AUM approximately $28 billion.

Nvidia's rise to $4.4 trillion market cap (December 2025) is almost entirely a data center story. Data center revenue grew from approximately $3 billion annually in fiscal 2020 to $194 billion in fiscal 2026 (Nvidia's fiscal year ends in January). Q3 FY26 (calendar Q3 2025) alone posted $51.2 billion in data center revenue, up 66 percent year-over-year. The company's fiscal 2026 total revenue was $57 billion in Q3 alone.

The driver is enterprise and hyperscaler capex on AI training infrastructure. Microsoft, Google, Meta, Amazon, Oracle, and others have collectively committed hundreds of billions of dollars to data center buildout through 2027-2028. Nvidia's management estimates that data center capex will grow 40 percent annually through 2030, reaching $3-4 trillion in annual spending by the end of the decade. This is the single largest capex cycle in technology history.

Because Nvidia weights roughly 20 percent of SMH by market cap, a 10 percent move in NVDA translates to approximately 2 percent move in SMH holding other constituents constant. The other 80 percent of SMH tracks a broader semiconductor market that includes foundries (TSMC), memory (Micron, SK Hynix via ASML), analog (Texas Instruments, Analog Devices), equipment (Applied Materials, Lam Research, ASML), and legacy logic (Intel, Qualcomm).

When NVDA outperforms SMH, the spread reflects AI-specific demand dominance. When SMH outperforms NVDA, broader semi end-markets (PC/mobile recovery, automotive, industrial) are driving performance. The NVDA-SMH ratio in early 2023 was roughly 0.3 versus roughly 0.4 in early 2026, reflecting Nvidia's outsized AI-cycle gains compressing back partially as other semi names caught up in 2024-2025.

From the ChatGPT release in November 2022 through late 2025, Nvidia stock rose approximately 14x while SMH rose approximately 3x. This 11-percentage-point compound annual gap over three years is one of the largest single-stock vs sector-index divergences in modern market history. For comparison, the dot-com era saw Microsoft and Intel outperform semi benchmarks by 2-3x over similar windows, but never at Nvidia's magnitude.

The gap has narrowed somewhat in 2025-2026 as Nvidia faces competitive pressure from AMD's MI300 and MI350 series, Intel's Gaudi, custom hyperscaler chips (Google TPU, Amazon Trainium), and Broadcom's custom AI ASIC business. Nvidia Q3 FY26 revenue growth of 62 percent YoY is still extraordinary but below the triple-digit growth rates of FY25. The NVDA-SMH spread has compressed 15-20 percent from its peak in mid-2024 to early 2026.

The three most important names in SMH outside Nvidia are TSMC, Broadcom, and AMD. TSMC (Taiwan Semiconductor Manufacturing) fabricates approximately 90 percent of Nvidia's chips plus similar shares of Apple, AMD, and Qualcomm silicon, making it the single most important company in the global semiconductor supply chain. TSMC's capex and capacity decisions set the ceiling for Nvidia's shipments.

Broadcom has built a substantial AI custom silicon business through 2024-2025, partnering with Google, Meta, and ByteDance on ASIC designs that compete with Nvidia GPUs for specific workloads. Broadcom's stock performance has closely tracked its AI-specific revenue growth, reaching approximately $11 billion annual run rate in AI in late 2025. AMD has narrowed the GPU performance gap with Nvidia at the architectural level (MI300X in 2024, MI350 in 2025) but remains at approximately 5-8 percent market share versus Nvidia's 86 percent.

Nvidia's quarterly earnings releases (four per fiscal year, typically in February, May, August, and November) are among the most-watched macro events in financial markets. Options implied move on earnings has ranged from 8 to 12 percent of stock price, implying $100-150 billion of market cap at risk on each report. SMH typically moves 2-3 percent on the same events given Nvidia's 20 percent weight.

Key metrics analysts watch in Nvidia earnings: data center revenue growth rate, hyperscaler commentary about capex plans, gross margin (has trended 75-78 percent due to AI premium pricing), guidance for the next quarter, and any commentary on specific product cycles (Blackwell ramp, next-generation Rubin platform, networking). Misses or cautious guidance produce sharp SMH moves despite broader sector strength. The August 2024 earnings miss drove SMH down 6 percent on a single day even though only Nvidia missed.

A critique of SMH is that Nvidia's 20 percent weight makes the ETF more concentrated on AI-specific risk than ideal for diversified semi exposure. Investors seeking equal-weight semi exposure use XSD (SPDR S&P Semiconductor ETF, equal-weighted at ~2-3 percent per holding) or PSI (Invesco Dynamic Semiconductors ETF, factor-weighted). XSD's tracking error to SMH has averaged 6-10 percentage points annually since 2023 because of Nvidia's outperformance, with SMH outperforming XSD during the AI rally.

For investors who want semi exposure without Nvidia dependence, XSD or SOXX (iShares Semiconductor ETF) provide meaningful differentiation. SOXX specifically caps Nvidia weight at 8 percent by index rules, making it a lower-Nvidia alternative. The choice between these ETFs is essentially a bet on whether AI-specific outperformance continues (favor SMH) or whether broader semi end-markets catch up (favor XSD or SOXX).

The NVDA-to-SMH price ratio is a useful gauge of where the AI cycle is in its maturity. When NVDA is sharply outperforming SMH (ratio rising), AI-specific capex is accelerating and Nvidia is capturing the lion's share. When the ratio compresses (SMH catching up), either AI capex is broadening to more players (bullish for sector), AI capex is moderating (bearish for NVDA specifically), or non-AI semi end-markets are recovering (cyclical signal).

The 2023-2024 period saw the ratio rise sharply as Nvidia locked in market share. The 2025-2026 period has seen modest compression as AMD, Broadcom custom chips, and hyperscaler in-house silicon gained share. A ratio staying flat or compressing through 2026 alongside rising AI capex would signal a competitive erosion story for Nvidia. A ratio expanding again on AI capex acceleration would mean Nvidia is maintaining dominance.

Nvidia and SMH both carry geopolitical risk centered on Taiwan. TSMC manufactures the majority of Nvidia's chips (and most of SMH's constituents' chips), making any disruption to Taiwan-based production catastrophic for the sector. US-China tensions over Taiwan have been elevated throughout 2024-2026 as China has pursued its own semiconductor independence through SMIC and other firms.

US export controls on advanced AI chips to China have limited Nvidia's addressable market. H100, H200, and Blackwell chips are restricted; Nvidia's China-specific H20 chip was introduced in 2024 to comply with regulations but faced its own restrictions through 2025. Approximately 10-15 percent of Nvidia revenue came from China pre-2023, down to 8-10 percent by 2025-2026. Any escalation (either Taiwan-related or further export controls) would hit Nvidia and TSMC disproportionately, and thus SMH disproportionately.

The primary signal is Nvidia's fiscal 2027 data center revenue trajectory. Consensus expects approximately $240-260 billion in fiscal 2027 data center revenue (up from $194B in fiscal 2026). Hyperscaler capex guidance from Microsoft, Google, Meta, Amazon, and Oracle sets the ceiling. If capex guidance stays elevated (above $500 billion combined for the top 4 hyperscalers), Nvidia outperformance of SMH likely continues.

Secondary signals: AMD MI350 market uptake (measures competitive pressure), Google TPU and Amazon Trainium deployment rates (hyperscaler insourcing), TSMC capacity expansion announcements (supply-side ceiling), and any escalation in US-China chip export controls. The April 2026 earnings season begins in late May for Nvidia; the May report will include Blackwell full-ramp revenue and the first Rubin-platform commentary. Key risk scenarios: AI capex peaks earlier than expected; Chinese chip independence accelerates; Taiwan tensions escalate; hyperscaler in-house chips capture more than 20 percent of workloads.