Data Infrastructure: The Grid Behind Every AI Query

The AI boom has a physical problem. Every large language model query runs on servers that need reliable power, precise cooling, and ultra-low-latency switching. Vertiv Holdings sits at the intersection of all three — its thermal and power management systems are inside the world's busiest data centers. Arista Networks builds the high-speed switching fabric that moves data between those servers. Neither company makes a chip, but both are indispensable to every company that does. That's the defining characteristic of the data infrastructure theme: the picks-and-shovels layer below the headline hardware.

The 83 companies in this theme span power generation, network equipment, satellite connectivity, and telecom backbone — a wider perimeter than most investors draw around "data infrastructure." The common thread is physical scarcity. You can design a new chip in months. Building a substation, laying fiber, or launching a satellite constellation takes years.

Power Is the Binding Constraint

GE Vernova is the clearest expression of this. Data centers now compete directly with industrial users for grid capacity, and utilities struggle to add generation fast enough. GE Vernova's gas turbines, grid equipment, and electrification hardware are on every hyperscaler's procurement shortlist. NRG Energy operates from the generation side — a power producer whose contracted capacity becomes more valuable as data center load forecasts keep rising. The energy angle isn't peripheral to data infrastructure; it's the rate-limiting step.

Vertiv's role reinforces the point. Power enters a data center, but a meaningful fraction is lost as heat before it ever reaches a processor. Vertiv's liquid cooling and power distribution units recapture that efficiency. As rack densities climb with GPU clusters, the thermal problem scales faster than the compute problem. Vertiv is the direct beneficiary.

Network and Connectivity Extend the Perimeter

Data infrastructure doesn't stop at the data center wall. Traffic has to get there. Verizon, Ericsson, and TELUS represent the carrier and equipment layer that backhauls traffic from edge to core. SK Telecom and PLDT fill the same role across Asia-Pacific, where mobile data growth is accelerating faster than in mature markets. These aren't growth stocks in the traditional sense, but they own the physical infrastructure — spectrum, towers, subsea cables — that no software company can replicate.

AST SpaceMobile adds the satellite dimension. Its direct-to-device broadband constellation is designed to reach users where terrestrial fiber never will. If it executes, it expands the addressable base of connected devices dramatically — which flows directly back into demand for the core infrastructure that handles that traffic.

Planet Labs approaches the theme from the data-collection side. Its satellite imaging fleet generates geospatial data that infrastructure planners, utilities, and logistics operators use for site selection and network optimization. The value isn't the satellite itself but the continuous, machine-readable data stream it produces.

The Investment Case Is Structural, Not Cyclical

The demand driver here isn't a single product cycle. AI training, inference, cloud migration, and 5G densification are all running in parallel. Each one requires more rack space, more power, more fiber, more switching capacity. VEON serves emerging markets where baseline connectivity is still being built — a slower story, but one that adds a second growth vector independent of hyperscaler capex. Realty Income (O) — typically a retail REIT — has expanded into data center sale-leasebacks, a signal that even conservative capital allocators see the infrastructure demand as durable.

The theme rewards investors who look past the chip manufacturers and model companies. The grid, the cooling, the switching, and the fiber are what make every AI application possible. Explore the full company set in the data infrastructure theme and the broader ETF library for exposure options.