Software Engineer, GPU Infrastructure - HPC

About the Team OpenAI’s Infrastructure Operations team is responsible for the availability, reliability, and operational excellence of one of the world’s largest AI infrastructure networks. The team owns day-to-day operations of production AI networks across Industrial Compute's data centers, working with colocation providers, deployment teams, and hardware vendors to deliver highly available GPU infrastructure for AI training and inference workloads. About the Role We are seeking an Infrastructure Operations Engineer to operate and improve the large-scale Ethernet fabrics that support GPU clusters, storage systems, and management infrastructure. This role combines hands-on production operations with automation, observability, and incident response across a global AI network. The ideal candidate has experience operating high-availability data center, cloud, AI, or HPC networks and can move comfortably from physical-layer troubleshooting to routing and fabric behavior, change execution, and root-cause analysis. You will partner closely with network architecture, systems engineering, GPU engineering, storage engineering, security, deployment, site operations, service providers, colocation partners, and hardware vendors to raise reliability and reduce operational toil. Key Responsibilities Own the operational health, availability, and reliability of production AI network infrastructure across Industrial Compute's data centers. Monitor, troubleshoot, and resolve network incidents while meeting service-level objectives (SLOs), reducing Mean Time to Detect (MTTD), and minimizing Mean Time to Recovery (MTTR). Operate and maintain large-scale Ethernet fabrics supporting GPU compute, storage, and management networks. Execute production network changes, maintenance windows, and capacity expansions with minimal customer impact. Manage the hardware lifecycle, including switch and optics replacements, RMA coordination, software upgrades, and preventive maintenance. Support new A

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. THE OPPORTUNITY We are looking for Senior Software Engineers to join our team. This is a specialized, high-impact role sitting at the intersection of high-performance computing (HPC) and Large Language Model (LLM) engineering. You will not just be building the automated "speedometer and diagnostic" suite for our next-generation AI infrastructure; you will be defining the roadmap, driving key technical decisions, and taking full ownership of the future of this work. RESPONSIBILITIES Benchmarking : Evaluate, run and automate standard LLM quality benchmarks (GSM8K, MMLU) alongside custom performance suites for specific workloads (e.g., long-context window, KV cache reuse, disaggregated serving). DevEx Improvement : Develop and maintain internal GPU-enabled development environments (similar to GitHub Codespaces). You will ensure the team has seamless, high-performance "dev machines" optimized for model experimentation. Tool Development : Build and contribute to open-source tools such as InferenceMAX and genai-bench to automate model evaluation, benchmarking and analysis. System Profiling : Use profilers like PyTorch Profiler, NVIDIA Nsight Systems and py-spy to collect performance profiles, identify bottlenecks, and debug the compute/networking stack. Monitoring & Observability : Develop real-time dashboards and alerts to monitor system health, model startup times, and runtime performance. Continuous Integration : Auto

ABOUT BASETEN Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F , led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to to ship AI products. THE ROLE Baseten is building its own GPU infrastructure for large-scale inference. As we move into large scale, high-density NVIDIA systems, the hardest failures are intermittent, cross-layer, and difficult to prove: RoCE congestion, InfiniBand stalls, ECN/DCQCN mis-tuning, bad optics, RNIC issues, host kernel stalls, GPU driver problems, and workload symptoms that look like network problems, but are not. We are hiring a Lead Software Engineer to build a first-class observability and root-cause analysis system for GPU fabrics. This is a hard distributed systems problem, not a dashboarding problem. The system will collect high-volume signals from switches, hosts, active probes, and inference services; reduce and correlate them in real time; understand topology and service ownership; and produce actionable diagnosis while an incident is still unfolding. This role sits at the boundary between networking and inference software. RDMA data paths, GPUDirect transfers, prefill/decode disaggregation, KV cache movement, request routing, and workload backpressure can all create fabric symptoms or hide real fabric failures. The goal is to tell an operator, quickly and with evidence, whether an incident is caused by the fabric, host, NIC, GPU, RDMA path, scheduler, or serving layer — and what to do next. EXAMPLE INITIATIVES Real-time telemetry engine — Build the ingestion, reduction, storage, and query path for high-cardinality fab

About the Team Compute Foundations builds the software that manages OpenAI’s GPU compute infrastructure across sites, data centers, and infrastructure providers, supporting model training and inference. Our systems turn large, heterogeneous fleets of machines into dependable compute for research and products. We build Kubernetes-based control planes, controllers, services, and APIs that coordinate the lifecycle of machines and clusters. We connect global infrastructure management with the realities of bare-metal systems, giving clients consistent interfaces across differences in hardware, topology, and provider behavior. About the Role You will build distributed systems that provision, configure, and manage compute throughout its lifecycle. Your work will connect global services and Kubernetes controllers with the systems that bring machines online, update them safely, and recover them when something goes wrong. This role combines software architecture with an understanding of how machines and data centers work. You might design a lifecycle API, improve controller performance under high concurrency and provider rate limits, or trace a provisioning failure from an API through reconciliation to network boot or host configuration. You will help these systems remain reliable as the fleet expands across sites and generations of GPU hardware. We value depth in relevant systems and the ability to connect layers. You do not need to arrive as an expert in every component of the stack. In this role, you will: Design, build, and operate Kubernetes-based controllers and distributed services that coordinate infrastructure across sites, isolate failures, and scale as GPU capacity grows. Define APIs and resource models that let clients request and track lifecycle operations through consistent interfaces across hardware platforms and providers. Build provisioning and configuration services that coordinate network boot, hardware management interfaces, and the deployment of firmware,

This role will support the fleet infrastructure team at OpenAI. The fleet team focuses on running the world’s largest, most reliable, and frictionless GPU fleet to support OpenAI’s general purpose model training and deployment. Work on this team ranges from Maximizing GPUs doing useful work by building user-friendly scheduling and quota systems Running a reliable and low maintenance platform by building push-button automation for kubernetes cluster provisioning and upgrades Supporting research workflows with service frameworks and deployment systems Ensuring fast model startup times though high performance snapshot delivery across blob storage down to hardware caching Much more! About the Role As an engineer within Fleet infrastructure, you will design, write, deploy, and operate infrastructure systems for model deployment and training on one of the world’s largest GPU fleet. The scale is immense, the timelines are tight, and the organization is moving fast; this is an opportunity to shape a critical system in support of OpenAI's mission to advance AI capabilities responsibly. This role is based in San Francisco, CA. We use a hybrid work model of 3 days in the office per week and offer relocation assistance to new employees. In this role, you will: Design, implement and operate components of our compute fleet including job scheduling, cluster management, snapshot delivery, and CI/CD systems. Interface with researchers and product teams to understand workload requirements Collaborate with hardware, infrastructure, and business teams to provide a high utilization and high reliability service You might thrive in this role if you: Have experience with hyperscale compute systems Possess strong programming skills Have experience working in public clouds (especially Azure) Have experience working in Kubernetes Execution focused mentality paired with a rigorous focus on user requirements As a bonus, have an understanding of AI/ML workloads About OpenAI OpenAI is an AI resea

About the Team The Workload team is responsible for designing and running OpenAI’s LLM training and inference infrastructure that powers frontier models at massive scale. Our systems unify how researchers train and serve models, abstracting away the complexity of performance, parallelism, and execution across vast GPU/accelerator fleets. By providing this foundation, the Workload team ensures that researchers can focus on advancing model capabilities while we handle the scale, efficiency, and reliability required to bring those models to life. About the Role We are looking for an engineer to design and implement the dataset infrastructure that powers OpenAI’s next-generation training stack. You will be responsible for building standardized dataset interfaces, scaling pipelines across thousands of GPUs, and proactively testing performance bottlenecks. In this role, you will collaborate closely with the multimodal researchers, and other infra groups to ensure datasets are unified, efficient, and easy to consume. In this role, you will: Design and maintain standardized dataset APIs, including for multimodal (MM) data that cannot fit in memory. Build proactive testing and scale validation pipelines for dataset loading at GPU scale. Collaborate with teammates to integrate datasets seamlessly into training and inference pipelines, ensuring smooth adoption and a great user experience. Document and maintain dataset interfaces so they are discoverable, consistent, and easy for other teams to adopt. Establish safeguards and validation systems to ensure datasets remain reproducible and unchanged once standardized. Debug and resolve performance bottlenecks in distributed dataset loading (e.g., straggler systems slowing global training). Provide visualization and inspection tools to surface errors, bugs, or bottlenecks in datasets. You might thrive in this role if you: Have strong engineering fundamentals with experience in distributed systems, data pipelines, or infrastructure.