219. Spiky AI Loads Are Forcing a Rethink of Data Center Power

From The POWER Podcast · 00:28:30

Show by POWER

The way artificial intelligence (AI) data centers draw power may be a harder problem for utilities than how much they draw. Training clusters can swing load by large amounts in milliseconds, and many facilities drop offline when the grid is stressed, which can make a grid problem worse. In this episode of The POWER Podcast, POWER Executive Editor Aaron Larson talks with Brad Witter, senior vice president of Industrial Digital at Generac, about what fundamentally changes with AI loads and how the power architecture around them is evolving. Topics include: • Why AI training loads are spikier than those of traditional data centers, and why that matters for both grid operators and on-site generation. • What utilities are asking of hyperscalers, including bringing their own generation and limiting load swings. • How battery energy storage and grid-forming inverters can act as a buffer between data centers and the grid. • Why validating these systems depends on utility-scale simulation, since no one can build a half-gigawatt proof of concept. • How Generac is approaching integrated power solutions, and where AI itself can help solve the problems it creates. Witter also shares what he thinks will separate the data centers that get their power strategy right from those that don’t. For more on the conversation, read the companion article at: https://www.powermag.com/spiky-ai-loads-are-forcing-a-rethink-of-data-center-power.

More episodes of The POWER Podcast

More episodes

PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.