About Anthropic
Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.
About the role
As a Product Manager focused on Compute Platform, you’ll partner with Infrastructure, Compute Operations, Engineering, Finance & Strategy, and Research to build the scheduling, orchestration, and capacity management systems that power Anthropic’s compute infrastructure—the foundation on which every model training run, evaluation, and inference workload depends:
- Partner with Infrastructure to build the systems that determine how jobs are scheduled, prioritized, and allocated across Anthropic’s growing fleet of GPU and accelerator clusters—ensuring the right workloads run on the right hardware at the right time.
- Your work directly impacts cluster utilization, cost efficiency, and researcher velocity: defining the semantic layer for job scheduling, establishing resource guarantees, and making the trade-offs that keep our infrastructure running at peak capacity.
- You’ll drive the evolution of our compute platform to support increasingly diverse workloads—from large-scale training runs and fine-tuning jobs to real-time inference and batch evaluation—each with distinct scheduling requirements, priority levels, and resource profiles.
- You will define and own the strategy and roadmap across job scheduling primitives, capacity allocation policies, preemption and fairness frameworks, quota management, and the observability tooling that gives engineering and leadership confidence in how compute resources are being used.
Responsibilities:
- Deeply understand the needs of internal customers across Research, Infrastructure, Product, and Finance—from researchers who need guaranteed resources for multi-week training runs to platform teams managing inference workloads with strict latency SLAs.
- Define and iterate on the semantic layer for job scheduling: the abstractions, priority tiers, resource classes, and preemption policies that govern how work flows through our compute clusters.
- Partnering with engineering leads to design scheduling capabilities that maximize cluster utilization while honoring resource guarantees—ensuring jobs have the right prerequisites (data, checkpoints, hardware affinity) validated before launch to avoid wasted compute.
- Drive product strategy and roadmap for compute capacity management, including quota systems, fairness policies, bin-packing optimizations, and gang-scheduling for distributed workloads.
- Own the trade-off framework between utilization efficiency, job latency, cost, and reliability—making transparent prioritization decisi