From Grid to Chip to Ambient: Engineering the AI Factory Energy Stack

Dr. Eugene van Rooyen, Product Line Manager, Eaton

AI data centers have crossed a threshold: we are no longer designing facilities, but city‑scale energy systems that convert power into tokens, and tokens into artificial intelligence. Rising rack densities, shrinking deployment timelines, and grid constraints demand a shift from enterprise design approaches to end‑to‑end energy stack optimization, from grid interconnection and power conversion to compute delivery and thermal rejection.

On the power side, three interdependent domains must scale together: IT distribution in the whitespace (480 V AC and emerging 800 V DC architectures), liquid cooling infrastructure where CDUs and high‑amperage busways become mission‑critical loads, and centralized cooling plants where medium voltage enables efficient delivery of industrial-scale power. Physics governs these choices, with current driving footprint and fault domains defining availability.

On the thermal side, the primary efficiency lever is the temperature delivered to silicon. Advanced cold plate designs with low thermal resistance enable hot‑water cooling aligned with chip roadmaps, reducing reliance on chillers and reframing performance around compute output per unit energy. Thermal resistance becomes the key architectural driver of system efficiency.

At scale, resilience requires coordinated power quality, ride‑through for cooling loads, and digital control layers. The advantage lies in engineering the entire energy stack—unlocking step‑change gains in efficiency and availability.

Dr. Van Rooyen specialises in translating customer needs and business objectives into scalable, profitable product portfolios, working at the intersection of engineering, operations, and commercial teams. His experience spans full product lifecycle ownership, portfolio optimisation, and growth of new products and customer segments.