The opportunity
Anthropic's Infrastructure organization is foundational to our mission of developing AI systems that are reliable, interpretable, and steerable. The systems we build determine how quickly we can train new models, how reliably we can run safety experiments, and how effectively we…
What you'll do
Own the technical strategy and roadmap for node lifecycle management: ingestion, bring-up, health checking, and automated repair
Drive cross-team initiatives to build and scale AI clusters across multiple clouds and accelerator families
Design and operate the systems that detect, isolate, and remediate unhealthy: hardware automatically, driving up fleet MTBI and minimizing stranded capacity
Define infrastructure architecture, ensuring the hardest problems get solved: - whether by you directly or by working through others
Work closely with cloud providers and internal research/inference/product: teams to shape long-term compute, data, and infrastructure strategy
Establish and evolve operational excellence practices (incident response, postmortem culture, on-call)
What they're looking for
- + years of software engineering experience, including time as a technical lead setting direction for a team
- Experience managing large scale compute infrastructure at hyperscale (10K+: nodes), including capacity management and efficiency
- Depth in one or more of: Kubernetes internals (scheduler, autoscaler, kubelet, Karpenter), cluster orchestration systems (Mesos, Borg-like), or node provisioning pipelines
- Low-level systems experience: kernel, virtualization, device drivers, firmware, or hardware health/diagnostics daemons
- Familiarity with high-performance networking (EFA, RDMA, InfiniBand) for distributed ML workloads.
- Demonstrated ownership of production reliability for high-throughput, latency-sensitive systems
- Contributions to relevant open-source projects (Kubernetes, Linux kernel, container runtimes, etc.)
- Skill in quickly understanding systems design tradeoffs and keeping track of rapidly evolving software systems
