Senior Software Engineer - Cluster Networking

You will be the technical owner of how our clusters communicate internally and externally. This includes the CNI data plane, the overlay mesh, and the nodes connecting clusters across regions and providers.

• Own and evolve the Kubernetes networking architecture for GPU clusters running at multi-thousand-node scale
• Design, operate and scale the overlay network - CNI, mesh and VPN topologies (Tailscale, WireGuard), and the gateways that connect control and data planes
• Design, operate and scale the L7 gateways/load balancers/tunnels (Envoy, Cloudflare)
• Find and eliminate scale ceilings: packet loss under load, control-plane saturation, IP address management exhaustion, and the failure modes that only appear above a few thousand nodes
• Build the scale-test environments and validation suites that let us catch networking regressions before they reach production, rather than during a training run
• Diagnose hard, ambiguous problems across the stack - where a symptom in Slurm or a training job traces back to a mark collision, a stale route, or a saturated tunnel
• Partner with cloud and neocloud providers on network topology, requirements and capabilities as we bring up new clusters
• Provide senior technical judgement to a distributed team, and depth in the Custer Networking domain.

Software Engineering, Computer Graphics, Artificial Intelligence (AI), GPU (Graphics Processing Unit), Computer Games, Cloud Computing, Network Architecture/Engineering, Topology, VPN (Virtual Private Network), Load Balancing, * BS/MS in Computer Science, Electrical Engineering or a related field, or equivalent experience

• 6+ years of professional experience in systems, network or infrastructure software engineering
• Deep command of Kubernetes networking architecture and CNI standards, with production experience operating Calico strongly preferred
• Proficiency designing and maintaining modern mesh and VPN networking topologies - Tailscale, WireGuard or equivalent
• Strong Linux networking fundamentals: routing, netfilter and iptables/nftables, packet marking, network namespaces, and how these interact with container runtimes
• Demonstrated ability to debug distributed network problems at scale - packet capture, tracing, and correlating behaviour across many hosts to find a single root cause
• Proficiency in Go, Python, C or a comparable systems language
• Clear written and verbal communication, and the ability to work effectively with engineers across multiple time zones

Ways to stand out from the crowd:

• Direct experience architecting and operating massive-scale Kubernetes topologies across thousands of concurrent nodes
• Experience with high-performance fabrics in AI or HPC environments - InfiniBand, RoCE, or RDMA over converged networks
• Upstream contributions to Calico, Cilium, Tailscale, or Kubernetes networking SIGs
• Experience operating networking across multiple public clouds and on-premises environments simultaneously
• Familiarity with Slurm or other HPC schedulers running on Kubernetes

Your base salary will be determined based on your location, experience, and the pay of employees in similar positions. The base salary range is 184,000 USD - 287,500 USD for Level 4, and 224,000 USD - 356,500 USD for Level 5.

NVIDIA is a pioneer in accelerated computing, known for inventing the GPU and driving breakthroughs in gaming, computer graphics, high-performance computing, and artificial intelligence. Our technology powers everything from generative AI to autonomous systems, and we continue to shape the future of computing through innovation and collaboration. Within this mission, our team, Managed AI Research Superclusters (MARS), builds and scales the infrastructure, platforms, and tools that enable researchers and engineers to develop the next generation of AI/ML systems. By joining us, you’ll help build solutions that power some of the most sophisticated computing workloads globally.

We are looking for a senior networking engineer to lead the network architecture of our GPU superclusters. Our platform runs frontier model training across tens of thousands of GPUs on multiple clouds, and we are scaling it toward clusters of ten thousand nodes and beyond. At that size, networking stops being a configuration exercise and becomes the hardest engineering problem on the platform - and it is currently one of the areas where we most need depth.