ChipsSeptember 4, 2026via AWS Machine Learning Blog
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
Why it matters
A vendor tutorial on running continuous ML pipelines (synthetic data, training, evals) at scale on managed GPU clusters—relevant to practitioners building physical AI systems who need to understand cluster economics and goodput optimization.
Key signals
- NVIDIA Cosmos 3 used for physical AI workloads
- SageMaker HyperPod persistent cluster architecture
- Amazon EKS deployment
- Synthetic data generation + post-training + closed-loop evaluation pipeline
- GPU goodput optimization as primary metric
- Model factory (continuous pipeline, not single-job training) framework
- NVIDIA Cosmos 3 used for physical AI model factory pipeline
- Workload includes: synthetic data generation, post-training, closed-loop evaluation
- Infrastructure: Amazon SageMaker HyperPod on EKS (persistent, resilient cluster)
- Success metric: GPU goodput (not just raw throughput)
- Published Sept 4, 2026 (vendor blog/how-to format)
The hook
Physical AI model factories demand persistent, resilient compute. Here's how to architect one on SageMaker HyperPod.
Building a Physical AI system takes a continuous pipeline, not a single training job. This post shows how to run that model factory (synthetic data generation, post-training, and closed-loop evaluation with NVIDIA Cosmos 3) on a persistent, resilient Amazon SageMaker HyperPod cluster on Amazon EKS, …