Capacity Planning for Bursty AI Workloads
Why GPU infrastructure must be planned around peak demand, power, cooling, network fabric, scheduling, and faster hardware cycles.
Author
CDCMP-certified Engineering Tech Lead at vCluster
Peter is a CDCMP-certified engineer who specialises in the lifecycle of high-density compute environments. His work connects physical infrastructure with DevOps automation, from bare-metal provisioning and high-performance networks to software-operated GPU estates. Before joining vCluster, Peter spent three years at DeepL, where he helped deploy one of Europe's largest private GPU clusters. His work covered NVIDIA H100 and GB200 systems, power and thermal monitoring, hardware lifecycle automation, and InfiniBand and RoCE fabric operations.
Earlier roles at Twitch, Brandwatch, and Cisco took him through global data center builds, remote operations, capacity expansion, structured cabling, site audits, and secure hardware decommissioning. He has delivered infrastructure across six continents, from greenfield sites to liquid-cooled high-density racks. Peter approaches infrastructure as a total lifecycle, including supply-chain integrity, hardware roots of trust, automated provisioning, day-to-day operations, and responsible decommissioning.