SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Salary: USD 320,000 - 405,000 / annual
Anthropic is building reliable, interpretable, and steerable AI systems. The Infrastructure Capacity Planner role sits within the Capacity Engineering team, which owns the full lifecycle of infrastructure growth—from planning through supply management to delivery and utilization optimization.
In this role, you will own the medium-range demand forecast for all resource classes Anthropic consumes: accelerators (by chip/interconnect class), CPU (by shape), storage (by tier and access pattern), network egress (by path), and managed services (by SKU). Your forecasts will be driven by model roadmap, RL/inference growth, evaluation volume, and retention policy—not trend lines alone.
Key responsibilities include:
- Building and maintaining the multi-resource demand forecast across accelerators, compute, storage, and network, informed by product and research roadmaps
- Running a weekly plan-vs-reality loop: comparing planned allocations against observed fleet occupancy, surfacing unrecorded trades and stale allocations, and driving variance to zero
- Qualifying each incoming capacity tranche against the forecast before contract signature, ensuring right shape, region, quarter, and supporting-resource envelope
- Partnering with Finance and cost-efficiency teams to translate forecasts into core cost drivers and identify where savings opportunities will emerge
You should have hands-on experience with capacity, demand, or supply planning for large-scale technical infrastructure (cloud, HPC, hyperscale, or internal platforms). You build forecasting models yourself using SQL, Python/pandas, or proper forecasting/optimization stacks rather than specifying them for others. You understand data-center resource classes deeply—why storage and egress don't forecast like GPUs, why headline chip counts rarely bind, and how to articulate the difference between forecasts, plans, and allocations. You prefer simple, inspectable models over opaque ones and instrument your own forecast error.
Preferred experience includes cloud/neocloud reserved-capacity onboarding, demand planning for accelerator fleets, data-center delivery (power, space, network, site acceptance), accelerator health and fleet-health SLOs, lifecycle/state-machine services for infrastructure assets, and onboarding new hardware generations into existing schedulers.
