AI clusters are exploding in size and pushing the limits of today’s data center power and network designs. In this episode, host Rob Coyle sits down with JP Buzzell, VP and Data Center Chief Architect at Eaton, to break down why power architecture has become the next big bottleneck and how the industry can respond.
JP explains how data center design has hit a turning point. Density jumped from a few kilowatts a rack to more than 100 kW almost overnight. The move from many small, varied workloads to massive, synchronized GPU clusters has changed everything from physical layouts to network physics. Packet loss now dictates rack layout. Space inside the rack has run out. Cooling had to evolve not just for efficiency, but simply to fit the gear. And power distribution needs a complete rethink.
The conversation covers the shift toward direct current power, why AC inside the rack cannot scale to the clusters coming next, and the role OCP is playing in driving standards, regulation, and cross-industry cooperation. JP also shares the human motivation behind this work, from medical breakthroughs to a cleaner, safer grid.
This is a clear and practical explanation of a topic that will shape the future of data centers for years to come.
00:00 – Why data center design is changing
02:15 – The jump in rack density
04:20 – Liquid cooling’s real driver
06:50 – Power congestion inside the rack
08:30 – JP Buzzell’s background
10:45 – Building hyperscale GPU clusters
14:40 – The power bottleneck hits
18:00 – Training vs inference reliability
20:40 – We’re still in the “light bulb era” of AI
22:10 – What slows DC adoption
25:00 – Inside the OCP power distribution effort
29:30 – Grid stability and load drops
34:00 – Modular power and offsite build
38:45 – The next wave of power tech
41:00 – Global standards and interoperability
51:55 – Final message: feed the good wolf
