Why we start with live power
The scarcest part of an AI factory isn’t the GPUs. It’s the megawatts behind them. So we begin where the power is already on.
Notes from the build.
All posts
6 posts
Designing a hall for racks that draw over 100 kW
Rack-scale GPU systems changed the unit of design. A data hall is now planned around liquid, weight and power per rack, not floor area.
What “dedicated” means in a dedicated GPU cluster
No shared tenancy, no noisy neighbors, no surprises in the scheduler. A cluster that’s yours, run by people who know it.
Bringing up a cluster: the first 72 hours
Between the last rack landing and the first job running, there’s a checklist. Here’s what’s on it, and why we don’t skip steps.
Planning capacity around a training run
Training runs have a shape: a ramp, a long plateau and a tail. Capacity that fits that shape costs less than capacity that doesn’t.
What an existing building gives you
Old industrial buildings were made for heavy loads, tall equipment and serious power. That makes them better starting points for AI than you might expect.