GPU Compute Fleet TPM: Capacity & Reliability Leader

Perplexity; San Francisco (CA); On-site; USD 170,000 - 260,000; Full time

Job summary

Perplexity is seeking a technical program manager to own how we plan, bring online and run the GPU compute behind every product. You will forecast compute needs, bring new capacity into production, and own fleet performance, driving fixes when reliability, utilization or cost fall short.

You will work across engineering, research, Finance and providers to align capacity with product launches and model releases.

Qualifications

• 6+ years of technical program management, capacity planning or infrastructure operations in cloud environments.
• Direct experience with GPU or accelerator fleets across training and inference workloads.
• Experience working with cloud or neocloud providers on reserved capacity, capacity commitments, onboarding or vendor performance management.
• Experience with infrastructure cost modeling, unit economics or cost attribution in partnership with Finance.
• Technical depth in cluster management, scheduling and orchestration, and the data skills to build forecasts and dashboards yourself.
• The judgment to surface hard trade-offs and communicate them clearly from engineers to executives.
• Thrives in a small, agile team; has initiative and desire for ownership without much precedent to lean on.

Responsibilities

• Partner with Finance on the compute demand forecast, 3–12 months out based on customer growth, model releases and post-training plans.
• Turn the forecast into decisions with Finance and engineering leadership on commitments and budget.
• Bring new capacity online and lead transitions from contract to production including provider onboarding.
• Build a single, trusted view of the fleet: capacity, cost, allocation, uptime and utilization by provider and workload.
• Drive daily utilization, reliability and cost-efficiency; reclaim idle capacity and surface trade-offs.
• Own inference and compute requests end to end; triage and resolve GPU, deployment and quota requests.
• Manage compute provider relationships with scorecards and regular reviews of reliability and delivery.
• Partner with engineering, research and finance on capacity decisions for training, batch inference and real-time production inference.
• Build the operating system for compute: processes and tooling to plan, allocate and track compute with clear owners.

Skills

Technical PM

Capacity planning

GPU fleets

Cloud/provider onboarding

Cost modeling

Job description

Perplexity is seeking a technical program manager to own how we plan, bring online and run the GPU compute behind every product. You will forecast compute needs, bring new capacity into production, and own fleet performance, driving fixes when reliability, utilization or cost fall short.

You will work across engineering, research, Finance and providers to align capacity with product launches and model releases.