WattsB4Bots
The AI buildout is a physical thing. Watts, silicon, steel — concrete, copper, and transistors under every chatbot. This site maps the stack, tier by tier.
Power · Silicon · Steel
The shape of the stack
Chatbots run on something. This site is a field guide to that something — how it’s built, who makes each piece, and where each tier bottlenecks.
AI's insatiable electricity demand is remaking the grid.
A single GPT-class query uses roughly ten times the energy of a Google search. Hyperscalers are signing multi-decade PPAs with nuclear operators, reopening shuttered reactors, and queuing gas turbines into the 2030s.
The chip stack is where the bottleneck lives.
GPUs, custom accelerators, HBM memory, and the fab-equipment duopoly are each single points of failure. Capacity is the binding constraint on AI — and whoever controls the silicon controls the pace of the race.
Data centers are the factories of the AI era.
$300B+ in planned data-center construction, new cooling physics, and a complete re-architecture of networking between racks and between sites. REITs, cooling specialists, and system integrators build it all.
The map — 37 companies
Every company on this list sits on the critical path of an AI training run or inference request. Power, silicon, or steel — one role each. No narratives without a ticker.
Tap a pillar to expand
Research only. Not investment advice. Do your own diligence.
Start here
A field guide to the AI infrastructure stack, transistor to gigawatt, in nine plates.
Tier-by-tier teardown of a single piece of the stack, starting with the GPU.
The 37 public companies closest to the AI buildout, as a reference sheet.