Staff Networking Platform Engineer, Infrastructure

Location: Remote (US). Must be based in Dallas, San Francisco or New York City. Remote for now, with hybrid planned further down the road.

Pay:

• L6 (Staff): $270,400–$365,800
• L7 (Sr. Staff): $341,400–$461,800

Compensation is not an issue for the right person. The role pays nearly all cash: the bonus is folded into base, and there is an equity component on top.

A late-stage AI cloud company backed by NVIDIA, with a valuation in the billions, runs AI compute out of data centers around the world. It is now building a managed Kubernetes platform that turns large fleets of bare-metal GPU compute into dedicated, isolated clusters for customers running AI training and inference. This role owns networking across the whole platform, from Kubernetes cluster networking down through the data center fabric and GPU interconnect, and turns customer connectivity, isolation and performance needs into managed platform features.

What you'll do

• Define how networking features are requested, provisioned, changed, observed and removed
• Automate tenant and node network lifecycles with Kubernetes-native tooling, including safe recovery from partial failures
• Manage BGP, routing policy, VLANs and overlays, plus VIPs, load balancers and public IPs, through fabric APIs
• Automate GPU networking (NVL72, InfiniBand, RoCE, GPUDirect) and validate isolation and performance
• Own IPAM and inventory integration so intended config, applied state and inventory stay in sync
• Set segmentation and traffic policy with security engineering, and standards enforced through GitOps and drift reconciliation
• Own network telemetry, troubleshooting and incident response

What you bring

• Deep networking: TCP/IP, BGP, routing policy, ECMP, segmentation, interconnects
• Production ownership of bare-metal leaf-spine data center networks with VLANs, VRFs and EVPN/VXLAN
• Kubernetes and Linux networking: CNI behavior, Cilium or similar, network policy
• Kubernetes controllers, operators or custom resources in Go, with GitOps and controlled rollouts
• Multitenant infrastructure or managed Kubernetes products
• Securing networks that run customer workloads

Nice to have: Netris, UFM or NetBox; NVIDIA AI networking (InfiniBand, Spectrum-X, RDMA, RoCE, congestion control); networking for large GPU clusters, including multi-site