Nvidia Unveils ‘Graphics 3.0’ to Boost AI Productivity in Factories

Nvidia has introduced its concept of “Graphics 3.0,” asserting that AI-generated graphics will significantly enhance productivity in factories and warehouses. This initiative aims to leverage generative AI (genAI) tools to create graphics that can train robots and automate various processes in the physical world. The announcement was made by Ming-Yu Liu, vice president of research at Nvidia, during a keynote presentation at SIGGRAPH 2025 held in Vancouver, BC.

The company believes that these AI-generated graphics can play a crucial role in several applications, from training robots to performing tasks in real-world environments to assisting AI systems in automating the design of equipment and structures. Liu emphasized the transformative potential of this technology, stating, “We believe we are now in Graphics 3.0…being superpowered by AI.”

This new approach comes as Nvidia’s graphics processing units (GPUs) are already widely utilized by various text-based genAI models and virtual assistants. Beyond these applications, the company envisions a future where AI can manage physical systems such as robots, traffic signals, home appliances, and autonomous vehicles. Jensen Huang, the CEO of Nvidia, spoke about the implications of this technology, noting in a video address that robots will redefine work across industries, including agriculture and manufacturing.

Despite the promise of Graphics 3.0, Nvidia acknowledges challenges in its implementation. Unlike virtual AI, physical AI relies on data that may not be readily available. This necessitates the creation of synthetic data through the simulation of virtual environments. Huang remarked, “Robots don’t learn from code. They learn from experience. But real-world training is slow and expensive.” To address this, Nvidia has developed AI models and simulation tools designed to generate the necessary pixels for training robots and other physical AI applications.

During the keynote, Aaron Lefohn, vice president of research at Nvidia’s real-time graphics lab, highlighted the need for innovative tools that allow artists to conceptualize and create more efficiently. Nvidia’s Cosmos AI models are intended to enable robots to receive commands, sense their surroundings, reason, plan, and execute tasks effectively in the physical realm. According to Sonia Fidler, vice president of research at Nvidia’s spatial intelligence lab, this approach is essential because “Physical AI can’t scale through real-world trial and error. It’s unsafe, time-consuming, and expensive.”

For instance, the training of autonomous vehicles is conducted in virtual environments, as the prospect of repeatedly crashing cars to gather data is impractical. In conjunction with this initiative, Nvidia introduced Omniverse NuRec, a platform that converts real-world sensor data into fully interactive simulations for training robots. This platform provides various tools and AI models for constructing, simulating, rendering, and enhancing 3D digital environments.

Nvidia’s capability to reconstruct virtual worlds is facilitated by 2D data collected from cameras and sensors, with every pixel labeled based on an understanding of the sensor data. Fidler cautioned that achieving perfect visual understanding is challenging due to inherent ambiguities.

The company also unveiled advanced AI material-generation tools aimed at creating more realistic graphics, incorporating detailed visual elements such as reflectivity and surface textures. Lefohn explained that 3D experts and engineers can interact with AI assistants using straightforward language to articulate their requirements, streamlining the creative process.

As Nvidia continues to push the boundaries of AI-generated graphics, the implications for productivity in various sectors remain significant. The development of Graphics 3.0 could usher in a new era of efficiency, fundamentally changing how industries approach automation and robotics.