AI and HPC data center networking

Most recently I led network engineering for the first phase of a multi-gigawatt AI and HPC data center campus, building the network from an empty site. The case study describes what that involved.

What I help with

Fabric design for GPU and HPC clusters. Front-end and storage fabrics on EVPN-VXLAN, and how they connect to the rest of the campus.

Out-of-band and management networks. Management networks for thousands of servers, PDUs and network devices, built so they stay reachable when production is down.

Automation from day one. Fabrics, IP addressing and DNS defined in Terraform before the first rack arrives, so scale doesn’t mean hand-typed configs.

Long-haul and campus fiber. Working with optical vendors and carriers on diverse dark-fiber paths back to metro data centers.

Working with server and platform teams. Translating what Kubernetes and HPC teams need into network requirements, and the other way around.