Supercompute Operations Engineering

This team owns 100% GPU uptime, end to end, 24/7, across both compute and facility. It owns incident response and the engineering behind it, holds central control authority from the customer back to the facility, runs day-to-day sustainment of live datahalls, and leads deployment engineering for new datahall bring-up. The team will also build the automation for continuous response improvement. The team is expected to have system level knowledge about the whole datacenter site, end to end. The team engineers solutions to enable GPU uptime, working directly with technicians in the datahalls and field to execute the team’s engineering planning.

• Own monitoring architecture, alerting, signal quality, and automation of monitoring across GPU and host telemetry (DCGM, Prometheus/Grafana) and facility telemetry (BMS, EPMS, DCIM, SCADA, CDU / direct liquid cooling).
• Own compute hardware change incorporation - physical, firmware, software.
• Be part of 24/7 iincident response, mitigation, and control authority - end to end - from customer impact back to the facility.
• Own day-to-day site sustainment operations, including resolving issues impacting rack-level compute and facility infrastructure.
• Disposition daily issues across both Site Operations (rack compute) and Facility Operations (power and cooling).
• Lead acceptance and deployment testing campaigns to bring new data halls fully online.
• Develop production automation for new datahall bring-up, automation behind alerting and automated incident response.
• Develop and maintain standard operating procedures and planning for facility and site technicians. Includes change / upgrade incorporation planning on both rack level and facility.
• Own and document the live, changing state of datahall operation as new halls come online.
• Enable Site Operations and Facility Operations technicians to do their day to day jobs more efficiently and safely to increase GPU uptime.
• Create predictive and responsive automation to alerts and incidents to enable your response to be faster and more impactful.
• Lead cross-functional reliability projects spanning compute, network (InfiniBand/RoCE), storage, and facility signal boundaries, including grey failures that cross those boundaries (e.g., NVLink flapping, thermal throttling, power sequencing).
• Truly root cause systemic and design failure classes across hardware and integrated facility and operations systems: turn fleet data (MTBF/AFR, repeat-RMA trends) into proactive alerting, standard responses, vendor RCA inputs, and supplier corrective actions.

…