Due to the rapidly growing use of Artificial Intelligence (AI), global electricity consumption of data centres is predicted to increase significantly over the next five years, with 2030 forecasts approaching 3% of all total global electricity consumption.

Data centres vary in their function and use, and can differ in their incorporation of an Uninterruptible Power Supply (UPS) to their topology. In the context of Tier I data centres defined by the Uptime Institute, a data centre topology features a centralized UPS where there is only one power-electronics pathway from its internal energy sources to data centre racks. Another potential solution is to have distributed UPS systems within data centres, where internal energy sources are divided into small-scale energy sources, also known as rack-level energy systems, in parallel with IT equipment, e.g., as per the data centre model with distributed UPS recently published by the US Department of Energy. For the sake of clarity, the energy source with the UPS systems can be a battery storage or a rotary generating unit.

There is an ongoing debate on suitable grid-code requirements for large loads, including data centres, to ensure system reliability and stability as we integrate more large loads into power systems. One of the key requirements for data centres relate to their performance and behavior during severe voltage conditions, also known as high/low voltage ride-through requirements. This is of particular importance given the recent data centre tripping incidents that happened in the ERCOT system. Considering a data centre with a centralized UPS, data centre controls typically switch operation to the UPS in order to protect its equipment when there is a severe voltage disturbance in the grid. In this case, the data centre does not comply with fault ride-through requirements defined by system operators, which requires them to stay connected during severe voltage disturbances. This is particularly of concern for data centres with a centralized UPS.

This paper first highlights the potential issues with fault ride-through operation of data centres with centralized UPS systems. This paper will then investigate the potential impacts of distributed UPS topology on the data centre performance in fault ride-through operation during severe voltage disturbances. More importantly, we will investigate how and to what extent the transition time duration between different data centre modes of operation (grid-connected mode and UPS-connected mode) may influence the fault ride-through performance of the data centres with centralized and distributed UPS systems. Finally, fault ride-through performance analysis will be performed considering both strong and weak grid conditions, as well as different system fault types.

Understanding the technical limitations and benefits of both centralized and distributed UPS systems in data centres would add certainty to data centre project developers in the design and development stage.