Next-Generation Solid-State Transformers for Powering Artificial Intelligence Data Centers: A Survey of Emerging Topologies

The rapid growth of AI datacenters and the transition toward megawatt scale power consumption of single server racks demand a fundamental redesign of datacenter power distribution architectures. Conventional low voltage ac (LVac) systems face increasing limitations in efficiency, scalability, and material usage. Solid state transformers (SSTs) offer a pathway to compact medium voltage ac to low voltage dc (MVac LVdc) conversion, enabling MVac distribution directly to the server racks. However, today’s industrial fully modular SST demonstra-tors show limited power density of roughly 0.2 MW/m3, which is insufficient for supporting the targeted high compute densities. This paper surveys emerging single-stage and quasi single-stage SST topologies with the potential for breaking through the power density barrier and achieving 1 MW/m3at 99% efficiency. Alternatives based on low-frequency transformers for near term adoption of LVdc distribution are also discussed. The paper highlights the need for a holistic comparative benchmarking of the different distribution architectures together with protection methods and SST concepts, especially including future MVdc distribution and MVdc-LVdc SSTs close to the server racks.