AI and high-performance computing (HPC) demand more power per rack. As a result, the need for high-density data centers is growing. Yet many existing facilities were designed for lower loads. What if your environment is still low-density and a customer suddenly needs high-density? In this article, we explain what “high-density” really means, how cooling and power need to evolve, and how you can scale up an existing site without turning everything upside down. We close with a real-world example from our data center near Basel.
What do we mean by a high-density data center?
A high-density data center is one where power per rack sits well above the usual range. Most data halls run at about 3-12 kW per rack. AI and high-performance computing (HPC) often push this to tens of kilowatts, with training peaks approaching or exceeding 100 kW. This shift affects everything: power distribution, cooling, and daily operations.
In short:
- Density shift: 3 to 12 kW per rack in conventional data halls; as high as 100 kW per rack for AI/HPC.
- More heat, more often: higher continuous thermal load plus peak events.
- Implication: air-only cooling is frequently not enough.
Why AI and HPC change the cooling equation
As models and datasets grow, accelerators (GPUs/TPUs) draw more power and generate more heat for longer periods. To keep performance stable and hardware reliable, most organisations need to move beyond air-only to liquid-assisted cooling.
- Higher thermal loads: AI/HPC racks run hotter than traditional servers.
- Rising energy use: accelerator-rich nodes lift overall facility demand.
- Sustained peaks: training and simulations mean prolonged high-intensity windows.
A practical path to high-density
A straightforward step for existing data halls is to install rear-door heat exchangers (RDHx). A heat exchanger at the back of the rack captures server exhaust and cools it with water before the air returns to the room. Because water absorbs about 3,000 times more heat than air, you can achieve significant efficiency gains without redesigning the entire hall. RDHx can also be combined with direct-to-chip liquid cooling for the hottest components as densities increase.
Designing for high-density: power, operations, compliance
- Power delivery: plan for sustained high loads and peaks; revisit distribution, redundancy and protection.
- Operations: liquid-assisted cooling shifts routines from “keep water out of the data hall” to controlled water circuits at the rack or component level; scale and rollout are phased and technology-dependent, so processes and training are essential.
- Compliance & data residency: upgrades should maintain national and EU requirements while providing the low-latency AI workloads need.