SMH: Assessing the Semiconductor ETF's Position on the AI Infrastructure S-Curve

The semiconductor sector is no longer just about making faster processors for phones and laptops. It is undergoing a fundamental paradigm shift, moving from a niche application to becoming the essential infrastructure layer for the next computing era. This transformation is powered by the artificial intelligence boom, which has created an exponential growth engine for the entire industry. Global semiconductor revenue grew a robust 21 percent year-over-year to $793 billion last year, with AI-related chips alone accounting for nearly one-third of total sales. This isn't a cyclical upswing; it's a structural repositioning of the industry's core purpose.

The growth is hyper-concentrated, with NVIDIA emerging as the undisputed engine of this expansion. In 2025, the company's staggering semiconductor revenue grew 63.9 percent from 2024, and it accounted for 35 percent of all industry growth. This dominance has cemented NVIDIA as the first company to surpass $100 billion in annual chip revenue, a milestone that underscores its role as the backbone hardware provider for generative AI. The shift is clear: what was once a niche business designing graphics processors has become the fundamental rail for a new technological paradigm.

This creates a powerful but maturing investment thesis. The SMH ETF is a concentrated bet on the dominant players building these hardware rails. The initial phase of exponential adoption is now giving way to a period of intense competition and infrastructure build-out. As AI infrastructure spending is forecast to surpass $1.3 trillion in 2026, the focus is shifting from pure chip design to the entire stack-networking, memory, and custom silicon. This sets the stage for the next wave of winners, where companies like Broadcom are positioned as essential, non-cyclical enablers of the AI infrastructure era. The paradigm has shifted; the question now is who builds the rails for the next leg of the S-curve.

Infrastructure Layer Analysis: The Semiconductor Stack

The SMH ETF is a concentrated portfolio of the companies building the physical and logical rails for the AI infrastructure era. Its top holdings form a complete stack, from the foundational manufacturing layer to the specialized compute and networking components. The ETF's 26 holdings are dominated by five giants: NVIDIA at 19.25%, Taiwan Semiconductor Manufacturing Company at 10.54%, Broadcom at 7.91%, Micron Technology at 5.64%, and Lam Research at 5.64%. Together, they represent the core hardware layers required to power the next computing paradigm.

At the apex of this stack is the compute layer, where NVIDIA reigns supreme. The company has become the undisputed engine of the AI boom, with semiconductor revenue of $125.7 billion in 2025 and a staggering 63.9 percent year-over-year growth rate. This dominance is not just about selling chips; it's about controlling the ecosystem. NVIDIA's CUDA software platform and NVLink networking have created a powerful moat, making its GPUs the default choice for training foundational AI models. Its position is so central that it accounted for 35 percent of all industry growth last year.

Beneath this compute layer sits the critical manufacturing infrastructure, where TSMC operates as a non-partisan enabler. The company is the world's leading foundry, and its role is to produce the chips designed by others. This creates a unique advantage: TSMC wins no matter whether Nvidia's GPUs or Broadcom's AI ASICs become the main chips used to power AI workloads. As the industry scales, TSMC's capacity and process technology leadership become essential, making it a foundational, non-cyclical infrastructure play.

The stack is completed by specialized components. Broadcom is emerging as a key player in the networking and custom silicon layer, providing the interconnects and manufacturing support for AI chips. Micron Technology supplies the high-bandwidth memory that feeds the AI processors, while Lam Research provides the advanced equipment used to build the chips themselves. This diversified exposure across the stack reduces single-point risk and captures growth from multiple angles as the AI infrastructure build-out accelerates.

Position on the S-Curve: Adoption Rate and Competitive Inflection

The SMH ETF's performance charts a clear path along the technology adoption S-curve. Its 120-day return of 34.94% and 61.1% rolling annual return reflect the powerful momentum of the early-to-mid adoption phase. This isn't a speculative bubble; it's the market pricing in the exponential scaling of AI infrastructure demand. The ETF's concentrated portfolio of foundational players is capturing this growth wave, with its price hovering near its 52-week high of $396.10.

Yet, the trajectory is shifting. As adoption matures, the competitive landscape is becoming more complex. The dominant player, NVIDIA, is beginning to see its moat tested. While its GPUs remain the default for training, customers are actively seeking cheaper, more efficient alternatives for the ongoing task of inference. This is where competition is emerging most sharply. Broadcom is increasingly becoming the company others turn to in order to help create custom AI chips. This move into custom ASIC design represents a direct, albeit complementary, challenge to NVIDIA's ecosystem lock-in, particularly as companies like Alphabet and Anthropic place massive orders for Broadcom-manufactured chips.

This dynamic is reflected in the portfolio's active management. The ETF's turnover rate of 6.2% signals a market where positions are being actively rotated. In a maturing adoption phase, this isn't just about chasing the latest hype; it's about navigating the inflection point where the infrastructure stack is being built out. The high turnover suggests fund managers are adjusting to the evolving competitive dynamics, potentially trimming exposure to pure-play GPU leaders while adding to the enablers-like TSMC, which benefits regardless of whether GPUs or ASICs dominate, and the custom chip designers themselves.

The bottom line is that the SMH ETF is positioned at a critical juncture. It rides the powerful adoption wave, but the path forward requires navigating a more crowded field. The initial phase of exponential growth is giving way to a period of intense infrastructure build-out and competitive differentiation. The ETF's performance and turnover highlight this transition: the momentum is real, but the winners will be determined by who builds the next layer of the stack.

Risks, Catalysts, and What to Watch

The SMH ETF's thesis is built on a powerful, long-term trend. But even the steepest S-curves have friction. The key risks are not about a lack of demand, but about the physical and architectural constraints that could slow the adoption rate or redirect the growth engine.

A major near-term risk is the cyclical memory shortage. The AI infrastructure build-out is pulling an unprecedented share of global manufacturing capacity toward high-margin memory solutions for data centers, like high-bandwidth DRAM. This has created a supply/demand imbalance that is driving up prices across the board and could persist well into 2027. For the SMH portfolio, this is a double-edged sword. While Micron Technology benefits from the price surge, the shortage itself poses a tangible risk to the entire AI deployment timeline. If memory bottlenecks delay server rollouts, it could temporarily dampen demand for the GPUs and networking chips that form the core of the ETF's holdings.

The primary catalyst for continued momentum is straightforward: sustained capital spending on AI infrastructure. JPMorgan points to strong capital spending and earnings growth fueled by AI as a key driver for stocks in 2026. This is the fuel that keeps the entire stack running. As long as hyperscalers and enterprises maintain their multi-trillion-dollar investment plans, the demand for advanced chips, foundry capacity, and supporting components will remain robust. This spending is the bedrock of the ETF's growth trajectory.

The most strategic risk, however, is architectural. The entire portfolio is built on the assumption that NVIDIA's GPU architecture will remain the dominant, if not the only, path for AI compute. The rise of custom ASICs-like those being developed by Alphabet and others-is a direct challenge to this paradigm. As noted, Broadcom is increasingly becoming the company others turn to in order to help create custom AI chips. If a new chip architecture proves significantly more efficient or cost-effective for the massive, ongoing task of inference, it could erode NVIDIA's moat and shift the value chain. This isn't a threat to the entire semiconductor industry, but it would be a major inflection for the SMH's top holdings, particularly NVIDIA itself. The ETF's concentrated nature means it is highly exposed to this single point of architectural risk.

In practice, the path forward hinges on managing these tensions. The memory shortage is a supply-side friction that could cause short-term volatility. The capital spending catalyst is the macro engine. And the architectural risk is the long-term question mark. For the SMH ETF to continue its ascent, it needs the capital spending to outpace the supply constraints, while the industry navigates the competitive shift without a disruptive paradigm change.