Infrastructure Software Engineer

About Etched

Etched is building AI chips that are hard-coded for individual model architectures. Our first product (Sohu) only supports transformers, but has an order of magnitude more throughput and lower latency than a B200. With Etched ASICs, you can build products that would be impossible with GPUs, like real-time video generation models and extremely deep & parallel chain-of-thought reasoning agents.

Job Summary

Building cutting-edge model-specific ASICs requires crafting custom infrastructure and toolchains to support ultra-fast, reliable, and scalable development across the stack - from simulation to silicon. We build this infrastructure as software - and we engineer it with the same best practices we apply to our products. We use the same rigor, design discipline, and quality standards and testing as we do to our ASIC, software, and platform.

You will lead the development and adoption of next-generation infrastructure tooling, enabling Etched ASIC, Software, and Platform engineers to iterate faster, build more reliably, and push the boundaries of AI performance. This includes building and scaling our hybrid high-performance compute (HPC) cluster, optimized for massively parallel CI, EDA workflows, Emulation, and hardware-aware job execution.

You’ll also architect and implement a state-of-the-art observability stack with LLM integration and a strong emphasis on streaming health and performance telemetry, log aggregation, distributed tracing, insight generation, synthetic testing, and smart alerting - across CI pipelines, simulation clusters, and service endpoints.

This role demands a strong software engineering mindset, quality instincts, and deep understanding of systems. It’s not just about writing scripts - it’s about writing code that builds and manages infrastructure with precision, repeatability, and intent.

Key responsibilities

• Architect and Scale Distributed Compute Systems: Design and build the orchestration layers that drive our hybrid high-performance clusters—enabling simulation, synthesis, and continuous integration of AI ASICs at unprecedented scale.

• Build Infrastructure-as-Code Systems: Develop and maintain a fully programmable infrastructure control plane to ensure reproducibility, auditability, and rapid iteration across the entire stack.

• Optimize End-to-End Developer Experience: Create tools and abstractions that empower engineers to harness massive parallelism without worrying about the underlying complexity..

• Workload Elasticity, Reliability, and Efficiency: Prototype and execute workload orchestration and migration strategies between on-premise and cloud environments, balancing performance, storage availability and replication, uptime, and cost across heterogeneous hardware and compute backends.

• Implement real-time telemetry, tracing systems that surface insights from millions of metrics, enabling proactive debugging and system optimization.

• Push the Limits of Observability: Build a full observability stack that includes dashboards, alerting, automated responses, and a synthetic testing framework to proactively test infrastructure performance and reliability for various application and data flows, ensuring we remain proactive against issues impacting development and productivity workflows.

Representative projects

• Design and deploy a fully automated, scalable hybrid HPC cluster, combining bare-metal servers and switches with cloud instances, provisioned through MaaS and orchestrated via SLURM and Kubernetes, optimized for mixed EDA workloads and parallel CI pipelines.

• Develop a real-time observability system for ASIC toolchain jobs and distributed builds, integrating Prometheus, Grafana, and VictoriaMetrics with streaming telemetry, tracing, and alerting to detect performance regressions before they hit silicon.

• Architect and implement a programmable infrastructure-as-code control plane, using Terraform, Ansible, and Puppet, to version, audit, and redeploy every layer of Etched's development stack with deterministic reproducibility.

• Create a zero-downtime interactive development environment that provisions and connects Jupyter and VS Code sessions to GPUs and high-memory nodes via a secure zero-trust network, abstracting away cluster state and machine failures.

• Prototype and evaluate dynamic workload migration strategies between on-premise and cloud environments to optimize for latency, reliability, and cost across simulation and synthesis pipelines.

• Design a synthetic testing and fault injection framework to valid