Infrastructure Explained · No. 01
How AI Gets Power
From generation to a GPU rack, in six steps.
The Intelligent Infrastructure Review · August 1, 2026 · 4 min read
Training and running AI models consumes electricity in a specific place, at a specific hour, through equipment that has to be built before the first server is racked. The path from a power plant to a processor crosses several owners and several approval processes, and the slowest one sets the schedule.
Diagram
Infrastructure Explained · No. 01
The Intelligent Infrastructure Review
How AI Gets Power
The delivery path from generation to compute. Each handoff has its own owner, cost, and lead time.
- 01
Generation
Utility, on-site, or contracted
- 02
Transmission
High-voltage delivery
- 03
Substation
Step-down and metering
- 04
Service entrance
Site interconnection
- 05
Facility distribution
Switchgear, UPS, backup
- 06
Compute
Racks, cooling, workload
Key takeaways
What to carry out of this
- 01
Power reaches compute through a chain of separately owned systems — generation, transmission, distribution, and the facility itself.
- 02
The binding constraint is usually deliverable capacity at the site, not the amount of generation available in the region.
- 03
Equipment lead times for transformers and switchgear can govern the schedule more than construction.
- 04
On-site generation and storage can shorten the wait, but they change the operating and permitting profile of the site.
The decision-maker question
For your site, which step in this chain is the long pole — capacity, equipment, or approval — and what is the plan for it?
Go deeper
From Grid Burden to Grid Asset
How data centers can strengthen the communities that host them, through dedicated generation, storage, and controls.
Read it →Follow Infrastructure Explained
One topic, one diagram, one page — sent as each installment publishes.