One thing I find interesting about AI data centers is that the construction method is no longer only about cost or speed.
Rack density seems to be changing the decision.
If a facility is designed around older assumptions, for example 5 to 10 kW per rack, then moving to AI workloads can create major issues for cooling, power distribution and sometimes even structural planning. Purpose built modular systems can be attractive because they are designed around higher density loads from the beginning.
But I do not think modular is automatically better. For large hyperscale campuses, traditional construction can still make sense because the owner has more control over design, procurement and long term flexibility.
How do you usually approach this question? Is rack density now one of the first things you check before comparing modular and traditional construction?
This article gives a good breakdown of the issue: Is Traditional Data Center Construction Running Out of Time?
At What Point Does AI Rack Density Change the Construction Method?
