NVIDIA rolled out Alpamayo 2 Super plus World Action Models to build synthetic-data foundations for robot manipulation — a week where the NVIDIA stack visibly extended toward the physical layer of AI infrastructure.
NVIDIA's developer blog stacked two posts the same day: Alpamayo 2 Super providing trajectories, reasoning traces, and auto-labels; Beyond VLAs proposing a World Action Models framework for robot manipulation. Datacenterdynamics also covered a data-center robotics startup launched by a former Microsoft researcher.
NVIDIA is widening its center of gravity from LLM/VLM messaging to synthetic data and World Action Models for robotics. Autonomous operation inside DCs sits directly on top of power, cooling, and maintenance automation, so this reads as groundwork for extending the NVIDIA stack into the physical layer of AI infrastructure. ※Product GA and adoption are unconfirmed.
Commercial shipment of Alpamayo 2 Super, adoption by hyperscalers and DC operators, and partnerships with robot startups (Physical Intelligence, Skild, Figure).