The opportunity
Crusoe is on a mission to accelerate the abundance of energy and intelligence . As the only vertically integrated AI infrastructure company built from the ground up, we own and operate each layer of the stack — from electrons to tokens — to power the world's most ambitious AI workloads.
What you'll do
Hardware Offload for Storage Protocols: Design and implement software that offloads storage-protocol work (NVMe-oF target/initiator processing, encryption, compression, checksums, erasure coding) from the host CPU onto DPUs/SmartNICs, measurably reducing host CPU overhead per I/O.
Control-Plane Integration: Build and maintain the integration between Crusoe's storage orchestration layer and DPU-resident services, including provisioning, health/telemetry, and failover of offload paths.
Cross-Team Fabric Work: Partner with the storage-networking function on the RDMA/RoCE and NVMe-oF fabric work that DPU offload depends on, and with Compute/Hypervisor on SR-IOV/vDPA-based virtualization of DPU resources for multi-tenant isolation.
Vendor Framework Integration: Evaluate and integrate vendor DPU SDKs and frameworks (e.g., NVIDIA DOCA, BlueField SNAP) to accelerate storage virtualization (virtio-blk/virtio-fs offload) without building bespoke firmware from zero.
Observability & Testing: Build monitoring and testing infrastructure specific to DPU-offloaded paths — this is new surface area for the team, and strong tooling here determines how fast we can safely iterate.
Performance Benchmarking: Drive IOPS/throughput/latency benchmarking and quantify host-CPU-cycles-saved per offloaded operation, to make the case for where offload investment pays off next.
What they're looking for
- Technical Point of Contact: Serve as the storage team's technical lead on DPU-related architecture decisions, including with hardware vendors.
- + years of professional software engineering experience, with deep, hands-on: systems-level programming (C, C++, or Rust) and production experience in Go, Java, or similar for control-plane/orchestration work.
- Direct, hands-on experience with SmartNICs/DPUs in production: NVIDIA BlueField (2/3), Intel IPU, ConnectX 6/7, or comparable hardware-offload platforms.
- Working knowledge of NVMe-oF, RDMA/RoCE/InfiniBand, and the storage protocols: DPU offload commonly accelerates.
