The RoleYou look after the servers and the clusters that run on them: bringing machines up from their BMC, provisioning them through automated tooling, and handing over GPU-ready Kubernetes clusters for tenants to use.

• What you will ownThe HA management cluster (etcd and core services) on the management hardware.
• Automated bare-metal provisioning for the GPU fleet: BMC/Redfish, PXE and virtual media, inspection, and lifecycle.
• The NVIDIA GPU stack on Kubernetes through the GPU Operator (driver, container toolkit, device plugin, NFD/GFD, DCGM exporter), including whole-GPU allocation and per-tenant isolation.
• The host side of the GPU fabric: NVLink and NVSwitch health through NVIDIA Fabric Manager, plus the RDMA NICs (ConnectX-8 class), jumbo frames, and GPUDirect setup on the node. The Network Engineer owns the switch fabric.
• Secrets and certificate management (Vault or OpenBao, cert-manager). Identity, SIEM, and overall security posture will sit with a dedicated security hire.
• GitOps-driven provisioning, so infrastructure changes are repeatable and auditable.
• Secure tenant decommissioning: cryptographic NVMe wipe and GPU memory zeroing between tenants.

What We Are Looking For

• Required5+ years in infrastructure or SRE, including 3+ years running Kubernetes in production with real node and cluster lifecycle ownership. On-prem or bare-metal experience counts for more here than managed-cloud Kubernetes.
• Strong Linux fundamentals: systemd, kernel modules, NUMA, cgroups, NVMe and LVM storage, and host networking (bonding, VLANs, nftables). You are comfortable debugging at the hardware and driver level.
• Hands-on server provisioning through IPMI, Redfish, or BMC tooling, with config management such as Ansible.
• The NVIDIA GPU stack on Kubernetes: the GPU Operator and its components, GPU health and telemetry through DCGM, and GPU scheduling and per-tenant allocation.
• Infrastructure as code and GitOps in production: Helm, Kustomize, and Argo CD or Flux.
• Comfortable scripting and automating in Python and/or Go.
• Self-directed: you find the problem, propose a fix, carry it out, and write it down.