Many existing data centers were designed around 5–15 kW per rack and air cooling. That baseline worked well until AI workloads changed the load profile completely. Today, a single rack can reach 60–100 kW, and air cooling becomes difficult to manage beyond roughly 30 kW.
At that point, switching to liquid cooling is not just a technology choice. It becomes a facility decision.
One thing that is often underestimated is that retrofitting is not a simple upgrade. It introduces structural implications that need to be addressed early:
- CDU weight can exceed raised floor capacity
- Pipe routing requires slab penetrations that must be coordinated
- Plant rooms may not have enough space for additional equipment
These are not late-stage details. They directly affect whether a retrofit is viable.
New builds solve many of these issues by design, but they come with higher capital cost and longer timelines. That’s why most operators end up somewhere in between, often introducing liquid cooling in dedicated zones rather than across the entire facility.
In practice, the biggest risks don’t come from the cooling technology itself, but from decisions made too late.
If you are working on similar upgrades, I found a detailed breakdown of retrofit vs new build from gbc engineers, including structural risks and planning mistakes to avoid right here.
Retrofit or Rebuild: What Actually Drives the Decision?
