AMD | Making Open-Model Inference Work at Scale

Open-weight AI models are creating new opportunities for cloud providers, AI startups, and enterprises to build revenue-generating inference services. But success depends on more than deploying a model. It requires delivering competitive performance, managing infrastructure costs, and attracting developers to production-ready platforms. This session examines the technical and commercial decisions behind serving open models at scale, from balancing latency and throughput to managing GPU utilization and cost per token. Drawing on deployments using AMD Instinct MI355X GPUs, Ramine Roane, CVP, Product Application Engineering at AMD, explores how hardware and software choices shape inference economics, and how platforms such as OpenRouter connect providers with developer demand. Attendees will gain practical insights into the economics of AI inference, strategies for maximizing GPU utilization, and approaches for turning open models into sustainable revenue streams.