Data Centers Rethink Power Architecture for AI Workloads

The panel discussion, centered on the debate between AC and DC power, revealed that the industry is moving beyond viewing batteries as simple emergency backups. Instead, experts are positioning energy storage as an active, intelligent layer capable of smoothing power profiles and managing high-frequency load fluctuations inherent to AI processing. James Li of Ampace noted that these new workloads demand a departure from legacy data center environments, requiring battery systems that provide both rapid response and long-term reliability.

Aaron Schott, also of Ampace, emphasized that advanced lithium systems at the UPS layer can absorb the erratic power spikes typical of AI hardware. This shift is particularly urgent in Europe, where grid constraints and strict Energy Efficiency Directive requirements are colliding with the need for scalable, high-density computing. As companies like Eaton and Ramboll collaborate with hyperscalers, the consensus is clear: future infrastructure must prioritize interoperability and load management to remain viable under the intense demands of the current AI-driven hardware cycle.