NVIDIA unveiled a full “physical AI” stack: Cosmos 3, a massive multimodal world‑model that simulates real‑world physics to train robots far faster; Vera, a new high‑performance CPU built for AI‑agent workloads that can run tools, manage workflows and out‑pace traditional x86 chips; and the Isaac Groot reference humanoid robot, an open‑design platform with Jetson AGX Thor compute, dual five‑finger tactile hands, extensive sensing and torque specs meant to give researchers a standardized hardware‑software foundation for robots, autonomous vehicles and vision AI. Together, these pieces position NVIDIA as the operating layer for physical AI. The summary also notes that another company, Foundation Future Industries, is already testing its combat‑oriented Phantom Mark 1 humanoid in Ukraine, highlighting the rapid move toward real‑world, potentially military, deployment of humanoid robots.
1. NVIDIA announced Cosmos 3, a world model designed for physical AI.
2. Cosmos 3 is built on a mixture of Transformers architecture.
3. Cosmos 3 combines vision, reasoning, world generation, and action prediction in one system.
4. Cosmos 3 can understand what it sees, generate or simulate physical scenes, and help predict what should happen next.
5. Cosmos 3 was trained on one of the largest multimodal physical AI data sets ever, comprising billions of samples across text, images, video, sound, and action trajectories.
6. Axios reported the Cosmos 3 training data amounted to 20 trillion tokens of multimodal data, including real and synthetic video, images, ambient audio, text, and action sequences from humans and robots.
7. NVIDIA claims Cosmos 3 can reduce physical AI training and evaluation cycles from months to days.
8. NVIDIA launched Vera, described as the first CPU built specifically for AI agents.
9. Vera is a high‑performance, energy‑efficient CPU designed for Agentic AI workloads, reinforcement learning, and data processing.
10. NVIDIA states Vera can finish diverse agent workloads up to 1.8 times faster than traditional x86 processors.
11. Companies planning to adopt Vera include Anthropic, OpenAI, XAI, Bite Dance, Coreweave, and Oracle Cloud Infrastructure.
12. NVIDIA named Dell, HPE, Lenovo, Super Micro, Asus, Foxcon, Gigabyte, QCT, and Wistron as partners building standalone Vera CPU systems at scale.
13. Reuters reported Jensen Huang describing Vera as a possible $200 billion market, with OpenAI, Anthropic, and SpaceX among major early adopters.
14. NVIDIA announced the Isaac Groot reference humanoid robot as an open humanoid robot reference design for academic research.
15. The Isaac Groot robot is built around a unit h two humanoid chassis, stands nearly 6 feet tall, weighs about 150 lb, and has 31 degrees of freedom.
16. It is paired with dual Sharpa wave tactile five‑finger hands, adding 22 degrees of freedom, for a total of 75 degrees of freedom across body and hands.
17. The robot includes a head‑mounted stereo camera with a 140° horizontal and 102° vertical field of view.
18. It also has wrist cameras for close‑range manipulation and an IMU for motion tracking.
19. Arm torque is rated up to 120 Nm, leg torque up to 360 Nm; rated arm payload is 7 kg with a peak payload of 15 kg.
20. Onboard compute consists of an NVIDIA Jetson AGX Thor T5000 Blackwell GPU delivering 270 FP4 teraflops of AI performance, a 14‑core ARM CPU, 128 GB of unified memory, and configurable power draw from 40 to 130 watts.
21. Leading institutions such as AI2, ETH Zurich, Stanford Robotics Center, and UC San Diego’s Advanced Robotics and Controls Lab will use the Isaac Groot reference design.
22. NVIDIA plans to collaborate with US, European, and South Korean humanoid robot makers in addition to Unitry.
23. Unitry is based in China, and some US lawmakers have expressed concerns about using Unitry systems in federally funded research.
24. NVIDIA states that software updates for the Isaac Groot platform are routed through NVIDIA chips, with protections like secure boot and confidential computing built in.
25. NVIDIA launched the Cosmos Coalition, bringing together companies including Agile Robots, Black Forest Labs, Generalist, LTX, Runway, and Skilled AI.
26. Foundation Future Industries’ Phantom Mark1 humanoid robots have been tested in Ukraine for dangerous logistics tasks.
27. Reports indicate two Phantom robots were sent to Ukraine earlier this year for pilot testing focused on supply pickup near hazardous areas.
28. Foundation Future Industries secured a $24 million Pentagon contract.
29. Foundation’s leadership has discussed future combat roles for humanoids, including eventual handling of weapons used by humans.
30. Foundation acknowledges a significant gap between a humanoid that can perform a slow logistics demo and one that can operate reliably in a real firefight.
31. Battery life, durability, water/dust/shock resistance, terrain handling, manipulation reliability, and cost are cited as major barriers for military humanoids.
32. Business Insider reported that Foundation believes humanoids could carry out more complex military missions within 5 to 10 years.