Abstract: Along with increasing use of artificial intelligence (AI), power supply for AI data centers has become critical for ensuring their continuous operation to provide multitude of services. AI data centers rely heavily on utility grids for their high power demand. However, it is known that utility grids are vulnerable to weather and other conditions which lead to outages. Hence, we design a microgrid that makes use of solar PV and battery technologies to increase reliability of power supply for AI data centers. It also reduces their carbon footprint and power demand from the utility grids. A real-world 5.85-MW AI data center in Wyoming state is used as a case study. The microgrid is designed to provide 60% of the center power demand from solar PV and 40% from utility. Additionally, it has a battery storage system capable of providing power for up to 18.2 hours. The battery is charged during high solar production hours and supplies critical loads in the data center at night. A prototype of the microgrid is built for evaluating the design concepts and the system performance. The testing results show that the prototype works properly. It supplies critical and non-critical loads in an automated manner using a microcontroller. For example, at battery state of charge (SOC) of 90% all the loads are operational. At the battery SOC of 40%, the microcontroller automatically turns off non-critical loads while keeping critical loads. The microgrid increases reliability of power supply to data centers because it has battery back-up power. The solar PV system adds to the microgrid reliability because it can supply power to data centers under utility outage conditions. The microgrid design is applicable to all data centers where space and resource are available for solar PV. The microgrid enhances AI data center reliable operation while reducing power demand stress on utility grids.

Keywords: Artificial intelligence (AI), battery, converter, data center, inverter, microgrid, microcontroller, solar PV

Pages: 106-118