Network Engineer – AI Cloud

About the role

The company runs AI clusters for training and inference. As a Network Engineer, you own the networks behind the GPU clusters: the high-speed compute fabric, the data centre Ethernet fabric, out-of-band management, and connectivity to carriers and the internet. This is a hands-on role: you design and build networks at new sites, keep production fabrics stable and fast, and own network issues through to root cause and permanent fix.

Key responsibilities

• Design network topology for new GPU clusters and sites: compute fabric, Ethernet fabric, management network and edge connectivity, sized for scale and failure tolerance.
• Deploy, configure and validate networks at new data centre sites, from day-0 out-of-band and management switches to production fabrics.
• Bring up InfiniBand/RoCE compute fabrics and leaf-spine Ethernet fabrics for new GPU clusters.
• Review bills of materials, rack elevations and cabling plans so each site matches the design, and run fabric acceptance tests before clusters go live.
• Own the day-to-day health, configuration consistency and performance of InfiniBand/RoCE and Ethernet fabrics.
• Operate and tune subnet managers (UFM/OpenSM), QoS and congestion control on the compute fabric.
• Manage data centre edge and WAN connectivity: BGP with carriers and internet exchanges, data centre interconnect.
• Troubleshoot link, optics, cabling and fabric performance issues, and drive root cause analysis on network incidents.
• Work with vendors on firmware upgrades, defects and RMAs.

Required profile

• 4+ years in data centre, service provider or large-scale cloud networking, with hands-on production operations.
• Hands-on experience deploying and operating InfiniBand and/or RoCE fabrics, including subnet managers, QoS and RDMA.
• Fluent in both English and Mandarin, spoken and written.
• Strong troubleshooting from the physical layer up: optics, cabling, interface counters, packet captures, MTU and routing issues.
• Solid Linux skills.
• Clear written communication: design notes, change plans and incident reports others can follow.

Required skills

• Deep routing and switching knowledge: BGP, IS-IS or OSPF, ECMP, EVPN/VXLAN, leaf-spine (Clos) designs.
• Experience with at least one major vendor platform: NVIDIA, Cisco, Juniper or Ruijie.

Working arrangement: based in Singapore, with regular travel to data centres across Southeast Asia; on-call rotation required, business hours by default plus rotating coverage for critical incidents outside business hours.