Starting with the basics: AI requires computing power. Compute needs data centers. And data centers are not just “buildings full of servers” — they are industrial complexes made of hundreds of physical components that must all work perfectly, all the time.

‍

Data Centers: The Starting Point of the Bottleneck Chain

‍

Companies like EME (EMCOR) build and wire these facilities — from switchgear rooms and cooling tunnels to backup generators and high-voltage power feeds. Without EPC contractors like EME, new AI campuses simply don’t get built.
‍

AI demand → more data centers → more EMCOR-type construction capacity

‍

‍

Racks & Servers: Where AI Actually Lives

‍

Inside each data center are tens of thousands of server racks from companies like SMCI (Super Micro) and Dell. These racks integrate GPUs, networking gear, and custom power systems. The more AI model training accelerates, the faster this layer expands.

‍

Racks get denser → more heat → more power → more stress on cooling
‍

Chips & Materials: The Metals Layer

‍

Every server rack is filled with chips made of copper, gallium, germanium, high-grade silicon, and transformer steel.
‍

This is where the strategic materials trade comes in:
‍

• Copper: FCX, TMQ, TGB
• Rare earth magnets: MP
• Antimony (key component in flame-retardant materials used in the construction of electronic equipment, server housings, and wiring within the data center environment itself): UAMY, PPTA
• Lithium (for backup storage): LAC, ALB
• Transformer steel constraints: NUE (grid metals), STLD
  ‍

Cooling
‍

AI servers run extremely hot. Cooling systems from Vertiv (VRT), Trane (TT), Johnson Controls (JCI), and Stulz keep temperatures from reaching silicon-meltdown levels.
‍

The CME outage proved the point:
‍
When cooling fails, trading stops. When cooling fails at an AI campus, the AI stops.

‍

Water: The Hidden AI Input
‍

Cooling takes enormous amounts of fresh water.
‍
AI queries indirectly consume 1–3 bottles of water per query once you include:
‍

• data-center cooling
• grid-level cooling for the power plants supplying the datacenter
  ‍

This is why water utilities and water-equipment companies are already raising long-term spending:
‍

• Water utilities: AWK, WTRG, AWR
• Water equipment & metering: XYL, ITRI, BMI
• Engineering & infrastructure: ACM, TTEK, J
  ‍

AI → more cooling → more water → more water infrastructure capex
‍
This bottleneck is growing faster than most investors realize.

‍

‍

Electricity
‍

Electricity prices are up ~35% since 2020, and Goldman now says 9 of 13 U.S. power regions are already “critically tight.”
‍

AI data centers run 24/7, drawing more power than steel mills.
‍

The utilities who can supply this power — and pass the cost into their rate base — are already preparing for an AI-driven boom:
‍

• Regulated utilities: AEP, SO, DUK, PCG
• Merchant power players: CEG, NRG
  ‍

The more AI grows, the more power grids stretch to their limits.

‍

‍

Nuclear
‍

Once you add up water, cooling, and grid strain, the world runs into a simple wall:
‍
There is no way to power AI with existing energy sources.
‍

Gas can’t keep up. Renewables can’t run 24/7.
‍
That’s why nuclear has gone from “long-term idea” to near-term buildout.
‍

Deloitte projects that U.S. data centers could consume: 176 gigawatts by 2035 — more than 5× today’s demand. At that scale, you need round-the-clock, high-density power. Nuclear is one of the few technologies that can realistically deliver that.
‍

Here’s the full nuclear trade, broken into its parts:
‍

Fuel Supply — Uranium Producers
‍

These are the companies that benefit first as reactors scale:
‍

• CCJ (Cameco) – one of the world’s largest publicly traded uranium suppliers
• UEC, UUUU – U.S. producers positioned for domestic sourcing
• DNN – long-duration resource backing
  ‍

Enrichment — The Most Critical Chokepoint
‍

44% of global enrichment historically came from Russia.
‍
The U.S. is rebuilding this capacity.
‍
Only one U.S. company is already producing advanced reactor fuel:
‍

• LEU (Centrus Energy) – currently the only U.S. company producing HALEU (high-assay low-enriched uranium) for advanced reactors, under DOE contracts, and a prime beneficiary of U.S. strategic fuel programs.
  ‍

Reactor Builders — The Companies Making the Hardware
‍

These firms build the SMRs (small modular reactors), next-gen cores, and naval reactors:
‍

• BWXT – DoD reactors + commercial SMR manufacturing
• OKLO – micro-reactors + partnerships with tech companies
  ‍

Nuclear Operators — The Utilities That Will Sell Power to AI Giants
‍

These companies have nuclear fleets already operating and selling baseload power at premium pricing:
‍

• CEG – largest U.S. nuclear fleet + merchant exposure
• NEE, D – hybrid portfolios with nuclear generation
  ‍

Construction & EPC — The Builders of the New Nuclear Fleet
‍

When the NRC (U.S. Nuclear Regulatory Commission) approvals accelerate, these are the firms that build:
‍

• FLR, J, ACM, TTEK – major engineering groups benefitting from SMR and reactor construction cycles

‍

‍