Product Manager - AI Inference & Model Serving

Mirantis is the Kubernetes-native AI infrastructure company, enabling organizations to build and operate scalable, secure, and sovereign infrastructure for modern AI. They are seeking a deeply technical Product Manager to own AI inference and model serving for their control plane for GPU infrastructure and distributed AI workloads.

Responsibilities:

• Own product strategy, roadmap, and lifecycle for inference and model serving, including serverless inference, dedicated endpoints, autoscaling, routing, KV cache management, and the related observability
• Lead deep technical discovery with NeoClouds, sovereign clouds, and enterprise platform teams, and translate findings into prioritized requirements and architecture direction
• Partner with engineering on system design trade-offs across runtime integration, GPU scheduling, network, storage, and serving topology, including disaggregated serving and multi-model serving
• Define positioning grounded in measurable outcomes: latency distributions, throughput per GPU, utilization, tail reliability, and cost per tokens
• Drive go-to-market execution: pricing and packaging, reference architectures, sizing guides, PoC playbooks, and direct engagement with customers, analysts, and ecosystem partners

Requirements:

• 7+ years in product management, technical product management, or a senior technical role owning AI/ML and inference product(s)
• Strong understanding of production AI inference, including model serving, serverless execution, dedicated endpoints, autoscaling, routing, workload placement, observability, and reliability
• Proven capability to reason about performance trade-offs across GPU, network, storage, orchestration, and runtime layers, and to translate low-level technical capability into business value such as TTFT, throughput per GPU, and TCO
• Working knowledge of modern inference runtimes (vLLM, SGLang, TensorRT-LLM, Dynamo, Triton) and the optimization patterns that matter in production: continuous batching, KV cache management, cold starts, prefill versus decode, disaggregated serving, and multi-model serving
• Credibility with engineering leaders and infrastructure operators, including comfort in production architecture reviews and technical commercial conversations with platform engineering buyers