Segments

The Infrastructure of Tomorrow

Five tiers, mapping how a dollar of hyperscaler capex flows through the AI / automation buildout. The taxonomy is the navigation: each subsegment is a curated 5–7 company peer set with its own scoring methodology, calibrated against a worked-example golden set.

17 live0 calibrated0 planned· 17 subsegments across 5 tiers
I

Compute

3 subsegments

The silicon that does the actual training and inference work. Where roughly 30–40% of every dollar of hyperscaler capex lands. The thesis starts here because everything else is built to feed it.

II

Datacenter

4 subsegments

Where the silicon physically lives. Mechanical and electrical content per megawatt, the construction labor to build, and the operators monetizing GPU capacity. The bottleneck shifts from chips to power and cooling once a rack pushes past 100 kW.

III

Power Stack

5 subsegments

The binding constraint of the buildout once buildings exist. Generation, grid, and the speculative pre-revenue developers solving the next-decade power gap. Nuclear gets two adjacent buckets because operating IPPs and uranium miners are fundamentally different securities.

IV

Embodied AI / Robotics

2 subsegments

Where compute becomes physical action. The public investable layer is the supply chain — automation platforms, sensors, perception. Humanoid platforms are mostly private; track that exposure through TSLA's profile rather than a separate bucket.

V

Critical Inputs

3 subsegments

The raw materials and specialty chemicals that everything upstream depends on. National-security and permitting overlays dominate fundamentals here in a way they don't elsewhere.