Storage Infrastructure at CoreWeave
Massive GPU clusters require petabytes of checkpointing bandwidth without network congestion. CoreWeave storage layers coordinate NVMe-oF target fabrics using low-latency userspace daemons written in Rust and C. Fast block indexing relies on lock-free LSM trees in host memory. Disk writes bypass the Linux page cache via direct I/O ring submissions. Corruptions cannot happen. We implement end-to-end checksumming across every 4KB sector, catching bit rot before weights touch GPU memory.
As Engineering Manager, you steer the teams building high-throughput scratch storage and persistent volume fabric supporting tens of thousands of GPUs. Disclosed base compensation ranges from $182,000 to $242,000 USD plus equity and benefits.
Core Production Responsibilities
- Guide architectural direction for NVMe-over-Fabrics (NVMe-oF) RDMA target daemons and parallel file storage systems.
- Supervise performance tuning of asynchronous block drivers utilizing io_uring and SPDK primitives.
- Coordinate incident triage and post-mortems for storage cluster network partitions and drive failure recovery loops.
- Foster an engineering culture focused on rigorous automated testing, failure injection, and deterministic benchmarking.
Target Experience
- Demonstrated leadership experience managing systems or infrastructure engineering teams building distributed storage or Linux block devices.
- Practical familiarity with Rust or C/C++ in asynchronous, low-level network or file-system programming environments.
- Working knowledge of Linux storage topologies, NVMe namespaces, RDMA/RoCEv2 networks, and POSIX semantics.
- Compensation: $182,000 to $242,000 USD base salary with equity incentives across US office hubs or qualified remote locations.
