AWS's open-source robot toolchain is a cloud funnel with Nvidia inside

The code is free. Where the training runs, how models reach the fleet and whose simulation stack sits underneath are not neutral choices.

AWS's new open-source Physical AI Toolchain is less a gift to roboticists than a packaging exercise that routes robot development through AWS compute and Nvidia's Isaac stack. Launched Oct 8, it ties synthetic data, training, simulation, edge deployment and a field-to-cloud retraining loop into one workflow.^[1][2]

Read the repo, not the launch post. The Apache-2.0 code is mostly infrastructure-as-code for running Nvidia components (Cosmos, Isaac Lab, Isaac Sim, GR00T, OSMO) on SageMaker, Batch, EKS and EC2 GPU instances, with Nvidia Jetson as the edge target. When we read it, the edge-deployment module was marked "Planned", although launch coverage describes Greengrass edge deployment.^[1][3]

Nvidia is the constant. Microsoft publishes its own open-source Physical AI Toolchain pairing Azure with the same Isaac and OSMO stack. The hyperscalers compete for the compute bill; Nvidia sits under both.^[4]

AWS shut RoboMaker in 2025 and says the new toolchain is not a direct replacement. AWS also says Amazon runs more than 1 million robots and Telexistence has deployed more than 300 humanoids in convenience stores (per AWS; not independently verified). Pricing for the underlying services: not published.^[1][2]