For network, storage, redundancy, and because nobody runs a hall at 100 percent.
What if
- Racks that fit
- Heat to reject
- tons
- White space
- Runs compute
- of the power
- Energy a year
- GWh
Every rack profile
| Rack | Racks that fit | GPUs | White space |
|---|
Under the hood
Capacity used to be a floor-area question. It is now power first, heat second and floor last, and most planning conversations still run in the old order.
Density broke the rule of thumb
A hall planned at 10 kW a rack and one planned at 130 kW are different buildings. Any estimate carried over from the old number is off by ten times.
Power became the unit of account
Nobody buys square feet any more, they buy megawatts, and PUE plus held-back capacity eat the difference before a single GPU turns on.
The binding constraint moved
Below roughly 40 kW a rack, floor area binds and air works. Above it, the liquid loop binds. Which side you are on changes the entire build.
- Of the power at the fence that never reaches a server, at PUE 1.2
- Published platform figures, as of July 2026
Four inputs, in the order they actually get decided. The example follows whatever you have set above.
Which end do you know?
The power is the siting conversation: a fixed service, and you need to know what it holds. The fleet is procurement: the racks are decided, and you need the megawatts before you promise a date.
Pick the rack, not the chip
The highest-leverage input. Moving from 45 kW of air to a 130 kW GB300 changes the cooling, the floor loading, the distribution and the schedule. Pick the closest profile or set your own; 50 kW and up is treated as liquid cooled.
The tax between the fence and the floor
PUE is total facility power over IT power. At 1.2, every 100 MW at the fence delivers about 83 MW to the servers. New builds reach 1.1 to 1.2; much of the installed base sits nearer 1.5.
Capacity you build and refuse to sell
Nobody fills a hall to the last kilowatt. Network, storage and redundancy take a slice first. 15 percent is an honest default; 0 is the theoretical ceiling.
Published figures for the systems halls are being planned around. Air-cooled numbers are practical ceilings, not hard limits.
| Rack | kW per rack | GPUs | Cooling | Notes |
|---|
A sizing tool that hides its assumptions is a sales tool. Here is the whole model, so you know exactly which constant to argue with.
- IT capacity
- facility MW / PUE
- Usable IT
- IT capacity × (1 − margin)
- Racks
- floor(usable IT kW / kW per rack)
- Heat to reject
- racks × kW per rack, in kW
- Cooling tons
- heat kW / 3.517
- Liquid loop
- 90% of heat, for liquid racks
- White space
- racks × 30 sq ft air, × 40 sq ft liquid
- Total shell
- white space × 2.0
- Energy a year
- deployed kW × PUE × 8,760 hours
Heat equals power. Essentially all the electricity into a rack leaves it as heat.
The ton is 3.517 kW. Liquid racks send 90 percent of their heat to the loop and 10 percent to room air, roughly what cold plates leave behind.
White space is 30 sq ft per air rack and 40 per liquid rack, aisles and coolant units included. The shell doubles it for electrical and mechanical rooms.
Energy is billed on what you run: deployed load × PUE × 8,760 hours. Training clusters run close to continuously.
A US home uses about 10,500 kWh a year. The comparison is there to be legible, not precise.
This is a planning heuristic for the first conversation, not an engineering submittal. Nothing here replaces a mechanical engineer.
Part of Infrabench
Free tools for the people who build data centers, on public data with every constant on the page.
Built by Adrian Mucha
A product designer for the software most designers avoid: data center tools, AI tooling and operator consoles where a wrong click costs real money.
- See the portfolioadrianmucha.us
- Open to new workBased in Los Angeles