Data Infrastructure: The Physical Layer AI Can't Skip

Artificial intelligence gets the headlines, but Micron Technology gets the orders. Memory and storage are the first bottleneck in any AI pipeline — training runs consume enormous amounts of high-bandwidth memory, and inference demands fast, dense storage at scale. Micron sits at the center of that constraint, alongside Western Digital, which supplies the hard drives and flash storage that data centers stack by the petabyte. The data infrastructure theme captures 83 companies whose revenues rise directly when enterprises, hyperscalers, and telecoms pour capital into physical compute and connectivity.

This is not a software story. It is a concrete, capital-intensive buildout — one where the winners are determined by who supplies the pipe, the switch, and the power socket.

Networking Is the New Bottleneck

Once memory is provisioned, data has to move. Arista Networks builds the high-speed Ethernet switching fabric that connects GPU clusters inside the largest AI data centers. Its gear is the standard choice for hyperscale operators who need deterministic, low-latency interconnects at 400G and 800G speeds. As GPU cluster sizes grow, the switching layer scales with them — more nodes means exponentially more east-west traffic, and that traffic has to traverse Arista hardware.

Beyond the data center wall, the traffic has to reach end users. Carriers like TELUS, KT Corp, and PLDT are investing in fiber densification and 5G backhaul to handle the explosion in data volume that AI-generated content and edge inference will create. VEON extends the same dynamic into high-growth emerging markets across Eastern Europe and Central Asia, where mobile data consumption is compounding fast.

Power and Sensing Complete the Stack

Data centers are power-hungry by design, and that creates a direct line to NRG Energy. As hyperscalers sign long-term offtake agreements to secure reliable electricity, diversified power producers with flexible generation capacity become infrastructure partners, not just utilities. The energy constraint on AI expansion is real, and companies that can guarantee consistent power delivery at scale earn durable contracts.

At the edge of the network, Planet Labs adds a sensing layer the rest of the stack depends on — daily satellite imagery that feeds geospatial data into analytics platforms, agricultural models, and climate risk systems. It is a different kind of data infrastructure, but the logic is identical: raw physical input that software cannot generate on its own.

Why This Theme Has Structural Depth

Data infrastructure is not a trade on a single product cycle. Memory, storage, switching, and power all benefit from the same secular driver — the world is generating more data, processing more of it locally and in the cloud, and doing so with AI models that are orders of magnitude more resource-intensive than the workloads they replace. Western Digital benefits when storage density demands rise. Arista Networks benefits when cluster sizes expand. Micron benefits when memory bandwidth becomes the binding constraint on model throughput.

The 83 companies in this theme span the full stack from semiconductor to satellite. Investors looking for exposure beyond the hyperscalers themselves — the picks-and-shovels layer of the AI economy — find it here. Explore the full ETF library for fund-based access to the theme.