Understand and track the companies powering AI.

The processors that train and run AI models, and the factories that make them. Demand for AI compute starts here; everything else in the stack exists to feed, connect, power and house these chips.

The cloud platforms and internet giants that buy most of the AI hardware, build the largest data centers, and rent capacity back out. Their capital spending is the single biggest demand signal for the rest of the list.

Every accelerator is bottlenecked by how fast it can be fed data. High-bandwidth memory sits next to the GPU; DRAM, flash and hard drives hold the training data and model outputs.

AI training runs across thousands of chips at once. The switches, optical transceivers and cabling that link them into one machine decide how much of that compute is actually usable.

AI data centers draw far more electricity per rack than the previous generation and run too hot for air alone. Generators, grid equipment and liquid-cooling suppliers are the physical limit on how fast capacity can be added.

The companies selling AI-driven software to businesses: data platforms that feed models, applications that embed them, and the tools that monitor them in production. Exposure here is usually a share of revenue, not the whole business.

The buildings, the server makers that assemble chips into racks, and the specialist clouds that rent GPUs by the hour. Closest to the physical build-out; most sensitive to its pace.