L A T E S T I N T E L L I G E N C E

As AI workloads generate unprecedented power demands, hyperscale data centres face greater pressure to accommodate denser, hotter, and more energyintensive computing environments. This paper emphasizes best practices for powering modern AI racks, focusing on intelligent rack PDUs, including updated sizing guidelines for outlets and branch circuits, power quality monitoring, and infrastructure scalability.

POWERING AI RACKS: THE NEW NORMAL

Modern AI applications, especially those involving large language model( LLM) training or real-time generative inference, are driving an explosive demand for parallel compute capacity. To meet this demand, hyperscalers are making significant investments in highwattage GPU( Graphics Processing Unit) clusters, rapidly increasing rack power density, and reconfiguring data centre floor space. In the US alone, data centre energy consumption is projected to rise from 61 GW in 2025 to 130 GW in 2030.

In response, data centre operators must rethink their power delivery strategies at the cabinet level. While a typical rack previously supported 15 – 20 kW, racks supporting AI infrastructure now exceed 60 kW and can reach up to 100 kW.

• Density Metric Per Rack

• Extreme 100 + kW

• High 60 – 100 kW

• Medium 30 – 60 kW

• Low 20 – 30 kW

This change requires careful consideration of every element of the cabinet’ s power setup – from designing power input and heat output to monitoring and redundancy. Intelligent rack power distribution units( PDUs) supporting three-phase power, which were once passive components in a rack, now play a crucial role in maintaining uptime, efficiency, and safety. The table below highlights the maximum capacity load data for common three-phase power types typically used in racks supporting AI workloads. �

WHITE PAPER

RACK POWER SOLUTIONS FOR MODERN AI APPLICATIONS:

LEVERAGING INTELLIGENT PDUs AND MONITORING

DOWNLOAD WHITEPAPER

PRESENTED BY

www. intelligentdatacentres. com 21