Capacity, Tier and density are decided here
Two systems carry the entire facility: the power train from the grid connection down to the rack PDU, and the thermal loop from the chip back out to the sky.
From the grid to the rack
Everything else in a data center is negotiable. The power train is not — it defines the capacity, the Tier, the CAPEX and, for an AI campus, whether the project is possible at all.
One redundant unit per capacity block
Survives one component failure
Grid failure is a design case
The utility will fail. What matters is that the IT load never notices. Run the sequence and watch where the load sits at every second.
- 0 msGrid failure detected
- 0 msUPS carries the full IT load
- 2 sGenerator start signal
- 10 sGenerators at rated speed
- 13 sParalleling and synchronisation
- 15 sATS transfers load to generators
- 20 sCooling plant restarted in sequence
Verified on site during L3 and L4 commissioning at full load bank.
Density decides everything
A 10 kW rack and a 120 kW rack are different buildings. Pick the density you need and the cooling architecture that can actually carry it appears.
Direct-to-chip liquid
- 01GPU cold plate
- 02Rack manifold
- 03CDU
- 04Heat rejection
Tell us what you are building
Send the site status, the grid offer and the target load. You get a concept, a power and cooling strategy and an indicative CAPEX back.