NVIDIA Accelerates Robotics & AI Development, Unveils New Products
NVIDIA is significantly intensifying its focus on robotics and what it terms “physical AI,” a strategic shift prominently featured at the recent SIGGRAPH conference in Vancouver. The company views this domain, where AI systems interact with and understand the real world, as the next major frontier in artificial intelligence, with the pace of development described as “incredible” by Rev Lebaredian, vice president of Omniverse and simulation technologies at NVIDIA. This commitment is underscored by a suite of new hardware, software, and AI models designed to accelerate the creation and deployment of intelligent autonomous systems.
Physical AI, as defined by NVIDIA, extends beyond traditional AI models that operate solely in digital environments, such as large language models. Instead, physical AI systems are end-to-end models capable of perceiving, reasoning, and performing complex actions in the three-dimensional physical world. This paradigm shift, moving from “Software 1.0” (human-coded serial code) to “Software 2.0” (software writing software via GPU-accelerated machine learning), aims to enable robots and autonomous systems to sense, respond to, and learn from their physical surroundings. Industries from manufacturing and logistics to healthcare and smart cities are expected to be transformed by this advancement.
Central to NVIDIA’s strategy is a “three-computer solution” for physical AI robotics, covering the entire lifecycle from foundational model training to real-time on-robot operation. For the intensive training of large robot foundation models, NVIDIA leverages its DGX AI supercomputers. For simulation and testing, the NVIDIA Omniverse platform, with its physically accurate digital twins, plays a crucial role, allowing developers to test and optimize robot fleets virtually before real-world deployment. Finally, for on-robot inference and real-time operation, NVIDIA offers the Jetson AGX Thor, a compact, energy-efficient computer designed to run multimodal AI reasoning models directly on the robot.
Further bolstering its AI infrastructure, NVIDIA announced new RTX Pro Servers at SIGGRAPH, featuring the Blackwell architecture. These RTX PRO 6000 Blackwell Server Edition GPUs are designed to accelerate enterprise workloads, including AI, machine learning, data analytics, and 3D graphics, delivering up to 45 times better performance and 18 times higher energy efficiency compared to CPU-only systems. Global system partners like Cisco, Dell Technologies, HPE, Lenovo, and Supermicro will offer these new 2U mainstream servers, making Blackwell power more accessible for enterprise and industrial AI.
The company also expanded its Nemotron AI resource family and introduced new world simulation SDKs and libraries, crucial for developing sophisticated AI agents. The Nemotron family, including Nemotron Nano 2 and Llama Nemotron Super 1.5, provides advanced reasoning capabilities for building smarter AI agents capable of handling complex, multi-step tasks. These models offer high accuracy and efficiency, with features like a hybrid architecture and quantization (NVFP4) to lower reasoning costs. Leading companies such as CrowdStrike, Uber, Magna, NetApp, and Zoom are already utilizing or planning to use these models.
New NVIDIA Omniverse libraries and Cosmos world foundation models (WFMs) were also unveiled, accelerating the development and deployment of robotics solutions. These include Omniverse NuRec libraries with RTX ray-traced 3D Gaussian splatting for large-scale world reconstruction, enabling developers to capture and reconstruct the real world in simulation. Additionally, Cosmos models like Cosmos Reason, a 7-billion-parameter vision language model, enable robots and AI agents to “reason” by understanding physics and common sense, crucial for planning and interacting with their environment. These advancements facilitate the creation of physically accurate digital twins and the generation of synthetic data, which is vital for training physical AI models and building AI agents that comprehend the physical world. Industry leaders such as Amazon Devices & Services, Boston Dynamics, Figure AI, and Hexagon are already embracing NVIDIA’s simulation and synthetic data generation capabilities.
NVIDIA’s strategic push into robotics and physical AI, supported by a comprehensive ecosystem of hardware, software, and advanced models, signifies a profound commitment to transforming industries and enabling a future where intelligent machines seamlessly integrate with our physical world.