Amazon triples Nvidia chip order and adopts full physical-AI stack for robotics

Amazon tripled its order of Nvidia chips in response to surging AI demand, deepening a partnership that spans both cloud training and, increasingly, robotics. The company said it plans to adopt Nvidia's full physical-AI stack: Omniverse for simulation and digital twins, Cosmos world models, Isaac for robotics development, and Jetson edge hardware to run inference on-device across its robot fleet.

The physical-AI commitment is the strategic core. Amazon operates one of the world's largest robotics deployments in its fulfillment network, and standardizing on Nvidia's simulation-to-edge pipeline signals a bet that world models and embodied AI — not just chatbots — are the next frontier. Nvidia's Cosmos3 world models and the new Jetson Orin Nano 2 slot directly into that vision.

The move also underscores Nvidia's demand picture even as OpenAI touts a Blackwell-beating in-house inference chip: hyperscaler orders are accelerating, not slowing. Amazon separately moved to acquire Amsterdam-based DuckLabs, maker of the popular DuckDB analytics engine, folding a fast in-process database into its data stack.

The tension worth watching is vertical strategy: Amazon builds its own Trainium/Inferentia silicon yet is tripling Nvidia orders, reflecting that supply and capability, not loyalty, drive purchasing at this scale. Watch for the dollar magnitude of the order, how quickly the physical-AI stack reaches production robots, and whether the DuckLabs deal signals broader consolidation in the data-infrastructure layer.