A

Anduril

Builds autonomous weapons for the US and allies

Senior AI Infrastructure Engineer - Physical Infrastructure

Platform EngineeringConfirmed live in the last 24 hours
Full-timeSeattle, WA, USAIn-person

About the Job

Requirements

  • At least 10 years in a hands-on infrastructure, high-performance computing, or datacenter engineering role supporting GPU compute at scale.
  • Hands-on experience with H200/B200/B300 or comparable GPU systems, including bring-up, cabling, and firmware/driver management.
  • Experience with high-performance interconnects such as NVLink, InfiniBand, RoCE, or Spectrum-X in clusters of hundreds of GPUs.
  • Experience with high-performance parallel storage such as VAST, DDN, Weka, Lustre, or similar.
  • Kubernetes experience is required.
  • A strong automation background, including building repeatable, automated deployment pipelines rather than relying on manual processes.
  • Ability to lift or move more than 50 pounds and perform physical datacenter work, including rack/stack, cabling, and troubleshooting.
  • Eligible to obtain and maintain an active U.S. Top Secret clearance.

Responsibilities

  • Lead the vision, execution, and long-term stability of large-scale GPU training infrastructure.
  • Ensure GPU clusters are highly available, fault-tolerant, and resilient for machine-learning platform and research teams.
  • Build self-healing mechanisms to detect and isolate hardware faults proactively.
  • Tune NCCL and high-speed networking, and optimize Kubernetes, Run:AI, and Ray scheduling.
  • Replace manual triage with automated deployment tooling and deep observability to keep training infrastructure reliable and scalable.
  • Rack, stack, cable, and bring up GPU compute systems, including physical topology, power, cooling, firmware/BIOS, and burn-in validation.
  • Build and tune interconnect fabrics, including NVLink, InfiniBand, RoCE, and Spectrum-X, connecting hundreds of GPUs into low-latency training and inference clusters.
  • Integrate high-performance parallel storage to support distributed training and terabyte-scale multimodal datasets.
  • Automate cluster deployment and configuration end to end, including infrastructure-as-code for bring-up, firmware/driver management, and fabric configuration.
  • Operate and extend the Kubernetes/Run:AI environment for GPU scheduling, quota management, and multi-tenant workload isolation.
  • Own fleet health through monitoring, alerting, and rapid triage of hardware and network faults, including bad transceivers, GPU Xid errors, NCCL/collective failures, and RoCE congestion.
  • Onboard engineers and researchers to the platform and serve as their escalation point, working alongside them to debug, train, and optimize workloads when infrastructure is the bottleneck.
  • Partner with product-facing teams to understand emerging compute needs and translate them into platform capabilities.

Desired Qualifications

  • Experience with NVIDIA NVL72 rack-scale systems.
  • Experience supporting large language model token-serving or inference infrastructure alongside training clusters.
  • Network fabric tuning experience, including congestion control, adaptive routing, and quality of service for RoCE/InfiniBand at scale.
  • Familiarity with GPU/network observability tools such as DCGM and fabric telemetry, and with automated fault detection.
  • Experience supporting infrastructure as a shared platform for multiple internal customer teams with differing requirements.
  • Experience with Run:AI or similar GPU scheduling/orchestration tools.

About the Company

Anduril

Builds autonomous weapons for the US and allies

Anduril builds autonomous weapons, such as drones, cruise missiles, and robot submarines, for the US military and its allies, often developing them with its own money before any order arrives. In 2024 it beat Boeing, Lockheed Martin, and Northrop Grumman for an Air Force contract to develop Fury, a pilotless jet that flies alongside crewed fighters. Its Lattice software merges camera and radar feeds into one live map and lets a few operators direct many drones and weapons at once, including other companies' systems. Governments buy its systems under contract, and Anduril builds much of the hardware in its own factories. Oculus founder Palmer Luckey co-founded Anduril in 2017, and it is based in Costa Mesa, California.

Company size
10,001+
Company stage
Private
Total funding
$11.5B
Headquarters
Costa Mesa, CA
Founded
2017

Recently Posted Jobs

Explore more opportunities at Anduril

See more jobs