chokepoints.ai
SUBSCRIBE
10 layers580 nodes2,376 dependencies9 chokepoints112 bottlenecks6,500+ companiesnode size = companies identified
CHOKEPOINTS · DATACENTRE DESK · FIELD ATLAS

The anatomy of a datacentre

A datacentre is one machine with fourteen organs. Power is its binding constraint, heat its by-product, and a handful of upstream suppliers its chokepoints. Click any part of the plate to open its detail sheet: how it works, what it costs, who builds it, and where it pinches.

$740-750bnMicrosoft, Amazon, Alphabet and Meta combined 2026 capex guidance219GWGlobal data centre capacity demand by 2030, McKinsey projection1.4%North American primary-market colocation vacancy, a record low125GW vs 45GWGB grid connection queue versus GB peak electricity demand128-160+ weeksStandard power transformer lead time, Wood Mackenzie Q2 2025 survey120-150kW/rackCurrent Nvidia rack power against a 5-10kW legacy baseline$10.7m to $40m/MWBuild cost per MW across three scope bases: shell-only, AI-optimised shell, and fully loaded with compute~4xUK industrial electricity price against US and Nordic competitors, cited by OpenAI when pausing Stargate UK (to-verify)

Plate I · general arrangement · click a numbered part or use the index →

The chapters

The drawing above is the machine. The chapters below are the market it sells into, the country it is being built in, and the stack that constrains it.

CONTEXT PLATES II-VII
THE MARKETII III
THE UNITED KINGDOMIV V
THE STACKVI VII