Director, Site Reliability Engineering — AI Accelerator Infrastructure

Director, Site Reliability Engineering — AI Accelerator Infrastructure

Santa Clara, CA (3 Days Onsite)

You will build and lead Our Client's Site Reliability Engineering function from the ground up — owning the infrastructure that development, validation, and customer-facing deployments run on. This spans colocation facilities, on-premises lab clusters, cloud environments (AWS, Azure, GCP), and the platform services customers use to collaborate with our Client on hardware and software deployments.

You are a hands-on engineering leader. You will establish SRE as a discipline at our client, hire and grow the team, set the technical direction, own SLOs for critical systems, and be the senior escalation point when things go wrong — all in parallel. You will partner closely with the Director of Dev Ops Engineering, whose pipelines and automation run on the infrastructure you own, and work directly with hardware and software development teams to ensure HPC infrastructure meets their workload requirements.

• Own the SRE function end-to-end: define the team's charter, establish SRE as a discipline within our client's engineering culture, and drive buy-in across hardware, software, and executive stakeholders who have operated without a dedicated SRE team.
• Hire, develop, and retain a team of 3–5 SRE engineers; establish a culture of operational excellence, ownership, and continuous improvement from day one.
• Define the SRE technical roadmap: reliability architecture, automation priorities, capacity planning, and on-call model — and execute against it with your hands on the keyboard where needed.
• Serve as the senior technical escalation for critical incidents — guiding cross-team triage, driving RCA, and ensuring systemic fixes rather than point patches.
• Translate operational signals and infrastructure health into clear, actionable narratives for engineering leadership and executive stakeholders.
• Partner with the Director of Dev Ops Engineering to align infrastructure reliability with pipeline and automation delivery; the two functions must operate as a unified platform.
• Direct a dedicated Data Center & Lab Technician team — your hands and feet across on-premises and colocation facilities; set their work priorities, establish operational standards, and ensure physical infrastructure execution aligns with the SRE technical roadmap.

• Establish SRE process from a zero baseline: define SLIs and SLOs, build error budgets, design on-call rotations, and create the incident management framework our client currently lacks.
• Own 24×7 reliability across colocation, on-premises lab clusters, cloud, and customer-facing platform services — designing for failure domains, progressive delivery, and strict change control at every tier.
• Own the full observability stack (metrics, traces, logs) and instrument it from the ground up — Prometheus, Grafana, and/or Datadog — with SLO visibility, alert design, and E2E signal quality.
• Evolve incident and problem management into a data-driven discipline: automated triage workflows, pattern detection across recurring failures, and every P0/P1 producing a written RCA with tracked systemic fixes.
• Own Fin Ops and capacity planning as a unified discipline across all three infrastructure tiers — cloud (AWS, Azure, GCP), colocation, and on-premises: establish spend visibility and attribution across every tier, model TCO comparatively, drive workload placement decisions based on cost and performance, and anticipate infrastructure needs for new silicon programs and customer deployments.
• Own the migration from ad-hoc JBOD-based storage and point-in-time snapshots to an enterprise-grade shared storage platform spanning on-premises, colocation, and cloud tiers — covering architecture, vendor selection, data protection design (snapshots, replication, DR), and integration with HPC workloads and development environments

There is no automation baseline today. You will build it.

• Drive IaC-first discipline across the team — Terraform, Ansible, and production-quality…