GPU Cluster Architect

Nebius
USA2025-08-21

About the job

We are seeking a GPU Cluster Architect to drive the design of our next-generation AI infrastructure. In this high-impact, hands-on role, you will make end-to-end architectural decisions across compute, networking, and storage — ensuring our platforms can meet the massive scale, performance, and reliability requirements of modern AI workloads. This is a high-impact, hands-on architecture role where you’ll define how tens of thousands of GPUs are interconnected, cooled down, powered, and optimized across multiple data center sites.

Responsibilities

Cluster Design: Architect scalable GPU cluster topologies including compute nodes, interconnect (InfiniBand, Ethernet), storage, and control planes.

Performance Modeling: Analyze AI/ML workloads (e.g. LLM training, inference) to inform design tradeoffs across latency, bandwidth, and GPU density.

Network Architecture: Align with network architect relevant design and validate low-latency, high-throughput interconnects (e.g., InfiniBand HDR/NDR, RoCEv2) at POD and DC scale.

Storage Integration: Work with storage teams to optimize performance for training datasets, checkpointing, and others.

Reliability & Monitoring: Understand and analyze signal from monitoring systems to the detect flows in design

Collaboration: Partner with site reliability, networking, storage, and DC engineering teams to operationalize and scale your architecture.

Qualifications

Minimum

5+ years of experience designing clusters.

Deep understanding of modern GPU architecture (NVIDIA, AMD, etc.).

Experience with HPC interconnects (InfiniBand & RoCE).

Solid background in systems architecture, networking, and hardware reliability.

Experience in scripting for automation and telemetry pipelines (Python, Go, etc.)

Preferred

No preferred qualifications listed.