NVIDIA launched Cosmos 3, Nemotron 3 Ultra, and previewed the RTX Spark superchip at Computex.
world-models
14 tracked signals on world-models.
Runway's GWM Worlds 2 turns video generation into real-time interactive simulation via WorldPrompt
“turns high-fidelity video and audio generation into real-time interactive simulation”
Perceptron AI wants to unify VLMs, VLAs, and world models into 'embodied foundational models' for real-time physical-world interaction.
“we want to move away from the distinction between VLM, VLA, world models, whatever you want to call it, to what we call embodied fundamental models”
NVIDIA Cosmos 3 Edge brings 4B world model robot control to on-device hardware
World Labs is acquiring a robotics company to apply its large world models to physical spaces.
“we are building the next frontier of AI which is what we call spatial intelligence”
NVIDIA Cosmos-H-Dreams enables real-time generative simulation for surgical robotics training.
NVIDIA introduces World-Action Models (WAMs), robot policies fine-tuned from pretrained world/video models to act.
“WAM World-Action Model: a policy that starts from a pretrained world-model or video”
Video models get their intelligence from LLMs, so the next Sora won't be a better video model but a video agent.
“In the near term, the next Sora won't be a better video model, but a video agent.”
NVIDIA launches Cosmos 3, a frontier foundation model for physical AI combining reasoning, world, and action models.
“Physical AI systems must understand the real world before they can act within it.”
Physical AI's real bottleneck is scarce, natural robot training video, not the models.
“AI is no longer the hard part, the data is what's important.”
MoonLake AI is building modeling infrastructure for practical physical, embodied AI grounded in causality.
“The North Star for AI, and indeed for cognitive science, will always be to understand and develop how to create embodied intelligence.”
NVIDIA's Cosmos-based World Action Models aim to generalize robot manipulation beyond training demonstrations
World models are the most promising path to closing AI's sample efficiency gap with humans.
“the two major problems that we have left to solve is intelligence per watt and intelligence per sample.”
NVIDIA GTC Berlin 2026 promotional content highlights AI, robotics, and digital twins convergence.
“We will be able to make progress towards solving problems that have so far been completely off limits.”