Assessing AI ETFs: Infrastructure Bets on the Early S-Curve

The investment thesis for AI ETFs is shifting from chasing the latest models to backing the fundamental rails being laid today. This is a multi-year buildout, not a sprint. The structural demand is clear and massive, driven by the capital expenditure of hyperscalers. The consensus estimate for their 2026 capital spending is now $527 billion, a significant climb from the start of the third-quarter earnings season. Historically, analyst estimates have consistently underestimated this spending, suggesting the final figure could be even higher. This isn't just a one-time surge; it's the foundational investment required to power the next paradigm of computing.

Yet, this buildout is hitting a critical bottleneck. Hyperscale data centers are being constructed at a breakneck pace, but the supporting fiber infrastructure is struggling to keep up. The Fiber Broadband Association projects a 2.3x increase in total fiber miles by 2029 to meet the demand. This lag creates a fundamental constraint on the entire system. Without this expanded, high-capacity backbone, the compute power housed in data centers cannot be efficiently connected, limiting the speed and scale of AI adoption. The infrastructure layer is where the real engineering challenge-and opportunity-lies.

This bottleneck underscores the long runway for adoption. While the technology is advancing exponentially, public usage is still in its early stages. A survey from mid-2025 found that only 34% of U.S. adults had used ChatGPT. This figure illustrates that we are in the early phase of the adoption S-curve. The massive capex and fiber buildout we are witnessing today are the necessary investments to support the millions of users and billions of AI interactions that will come as the technology becomes more embedded in daily life and business. The infrastructure being built now will determine how steep that adoption curve becomes.

ETF Exposure: The 3 Key Infrastructure Bets

For investors, AI ETFs offer a way to ride the infrastructure buildout without picking individual winners. But not all funds are built the same. Their strategic differences-between broad diversification and focused tilt-will determine how directly they capture the S-curve of adoption.

The largest player, the Global X Artificial Intelligence & Technology ETF (AIQ), provides the broadest exposure. With a portfolio of 86 companies spanning tech, consumer, and communication sectors, it offers a diversified basket. This approach is appealing for its risk mitigation, but it also dilutes the core infrastructure thesis. A significant portion of its holdings are in software and consumer-facing companies, not the capital-intensive hardware and networking layers that are the true bottleneck. Its 2025 return of 32% and three-year average of 36.4% are strong, but they reflect a mix of winners across the entire AI value chain, not a pure play on the foundational rails.

A more targeted approach is seen in the Invesco AI and Next Gen Software ETF (IGPT). With a focus on software and infrastructure, it has a clearer strategic tilt. Its 2025 return of 31.7% and three-year average of 25.2% show it has captured the software-driven growth phase of the S-curve. This fund is built for investors who see the next paradigm as being defined by intelligent applications and platforms, not just the underlying compute.

The third key bet is the Roundhill's Generative AI ETF (CHAT). This fund is another concentrated, focused play. It holds around 40 stocks, with a portfolio weighted toward U.S. large caps. Its stellar 2025 performance aligns with the generative AI hype cycle. More importantly, its holdings include major infrastructure names like Nvidia, SK Hynix, Samsung, AMD, and Broadcom, alongside hyperscalers like Alphabet and Microsoft. This gives it a direct line to the semiconductor and memory supply chain that is critical for the AI compute explosion.

The bottom line is that these ETFs represent different points on the infrastructure S-curve. AIQ offers a wide net, CHAT is a focused spear, and IGPT is a software-centric tool. For an investor betting on the multi-year buildout of compute and connectivity, the funds with the deepest exposure to hardware and networking-like CHAT-may be better positioned to ride the exponential growth phase as adoption accelerates.

Catalysts, Risks, and What to Watch

The infrastructure thesis hinges on a few key forward-looking scenarios. The most critical is the resolution of the fiber bottleneck. The Fiber Broadband Association projects a 2.3x increase in total fiber miles by 2029 to meet AI demand. If this buildout accelerates, it will validate the multi-year capex cycle and remove a fundamental constraint on adoption. The lag in fiber infrastructure is already a known risk, so its resolution would be a major catalyst for the entire ecosystem.

At the same time, the market is maturing. The initial investor rotation has been toward pure infrastructure plays, but that trade is showing signs of fatigue. Goldman Sachs Research observes a rotation away from AI infrastructure companies where capex is debt-funded and operating earnings growth is under pressure. This signals a shift in focus toward the next phase: companies that can demonstrate a clear link between AI spending and revenue. The firm expects the next beneficiaries of the AI trade to be platform stocks and productivity beneficiaries. For investors, this means the thesis will evolve from betting on the rails to betting on the traffic they carry.

The primary risk to the infrastructure thesis is that massive capital expenditure does not translate into proportional revenue growth for the companies building it. This is especially true when capex is funded with debt. The market is already punishing infrastructure names that show this disconnect, as seen in the decline in stock price correlation among large public AI hyperscalers. The divergence in performance shows investors are being selective. The bottom line is that the infrastructure layer is a necessary but not sufficient condition for returns. The exponential growth in AI adoption must eventually flow through to the balance sheets of these capital-intensive builders, or their valuations face pressure.