CES 2025: Nvidia Leads with AI Innovations

"Intelligence is generated in a new way, a factory of tokens that revolutionize how we understand the world. From words to knowledge, from ideas to videos, AI tokens transform endless possibilities into reality."

In this article:

In a monumental CES 2025 presentation, Nvidia CEO Jensen Huang illuminated the stage with a transformative speech about the role of tokens in building a new era of artificial intelligence. Tokens, Huang explained, have become fundamental building blocks in creating advanced AI, enabling computers to perceive, interpret, and create with unprecedented precision. This post explores how Nvidia is leading this revolution and its impact across industries.

GeForce: The Foundation of Modern AI

Huang recalls Nvidia’s history, starting in 1993 with the MV1, which turned PCs into gaming consoles. The introduction of the programmable GPU in 1999 revolutionized modern graphics and paved the way for significant AI advances. With the RTX Blackwell development, Nvidia continues to set performance and energy efficiency standards in graphics computing, integrating AI to optimize pixel processing and create stunning images.

The RTX Blackwell series announcement 50 highlights GPUs’ ability to perform complex AI tasks like neurotexture compression and neuromaterial shading, producing high-quality images with lower computational cost. This innovation is redefining the industry, making it possible to create real-time graphics with precision previously thought impossible.

Huang highlighted the importance of scalability laws in the evolution of AI, explaining how they are fundamental for the development of more sophisticated and effective models. The first scalability law establishes a direct link between the model size, the amount of training data, and the available computational capacity. This means that the larger the model and the more extensive the training data, the greater the need for computational capacity to ensure the AI model's effectiveness. With the exponential growth of multimodal data, which includes text, images, and sounds, the ability to efficiently process and interpret this information becomes increasingly essential for advancing AI technology.

The second law, known as post-training, highlights the crucial role of techniques such as reinforcement learning with human feedback. This method allows AI models to enhance their abilities through a continuous process of practice and adjustment, similar to human trial-and-error learning. Human feedback acts as a mentor, guiding the model toward specific improvements in its skills, whether in solving mathematical problems or logical reasoning capabilities.

The third law, test time, refers to the model's intelligence in applying reasoning and effectively allocating its computational resources during real-time interaction. This capability is critical for AI to provide optimized and accurate responses, especially in applications requiring quick and complex decisions. This law emphasizes the importance of scalable reasoning, where the model can break down complex problems into smaller steps and evaluate different solutions before reaching a conclusion.

One of Huang's most revolutionary announcements was the launch of Nvidia Cosmos, a global foundation model designed to understand and simulate the physical world with unprecedented precision. This model is trained on 20 millions of hours of high-definition video, capturing complex physical dynamics such as gravity, inertia, friction, and spatial relationships. Cosmos is a platform that integrates autogenerative and diffusion models capable of creating virtual video states of the world, which is essential for applications in robotics and industry.

Cosmos enables the convergence of physical simulation and artificial intelligence in a way that allows the generation of incredibly realistic synthetic data. This feature is crucial for training robots with data that simulates real-world conditions, reducing reliance on physically captured data, which is expensive and time-consuming to collect. Additionally, Cosmos can simulate a wide range of scenarios and environmental conditions, such as variations in weather and lighting, providing a robust platform for testing and validation in AI policy development.

This advancement promises to revolutionize how robots and automated systems are integrated into industrial and everyday environments, allowing them to operate more efficiently and safely. With Cosmos, developers can generate virtual environments that accurately reflect real-world dynamics, providing a solid foundation for creating robots that not only operate in their workplaces but are also capable of predicting and reacting to variables and changes in real time. This represents a significant step toward a new era of robotics and industrial automation, where simulation and AI combine to create smarter, more adaptable machines.

UniTree G1: Humanoid Robots in Real Life

During the event, we had the opportunity to explore the UniTree G1, an affordable humanoid robot that is gaining popularity among developers. With a starting price of $28,000 in the US, the UniTree G1 comes equipped with a built-in microphone and is being integrated with chat GPT to enable more natural interactions with users.

The G1 can be remotely controlled and features integrated depth cameras and LIDAR for environmental perception. These features allow the G1 to efficiently navigate various environments, avoiding obstacles and performing pre-programmed movements such as dancing or handshaking. For developers, the G1 is a versatile platform to explore new applications in robotics.

Furthermore, the UniTree G1's flexibility allows significant customization. With the ability to add different accessories and components, developers can adapt the robot to meet specific needs. The unit has an AI-controlled walking mode, as well as collision prevention modes and even maneuvers such as the "hand position." These features make the UniTree G1 not only an innovative tool but also a showcase of current robotic capabilities.

Autonomous Vehicles and the Future of Mobility

Nvidia is at the forefront of the autonomous vehicle revolution, collaborating with automotive giants to develop technologies that will make autonomous driving a reality. With the Thor processor, Nvidia offers 20 times the processing capacity of the previous generation, enabling the integration of high-resolution sensors and the execution of complex calculations necessary for autonomous navigation.

Additionally, Nvidia's data platform, which combines real and synthetic data, is setting new standards for safety and efficiency in autonomous vehicles, accelerating the development and deployment of these technologies in the market.

General Robotics: The Next Frontier

Huang foresees a future where general robotics becomes an everyday part of human life, acting as multifunctional assistants in various aspects of daily life and industry. Nvidia, with its Isaac Groot project, is at the forefront of this movement, developing a technological infrastructure that enables the creation of sophisticated humanoid robots. These robots are being designed to perform a wide range of complex tasks with high precision and safety, from industrial activities to personal assistance in homes.

The Isaac Groot project offers a set of essential tools for developers, including foundation models for robots, data pipelines, simulation frameworks, and the powerful Thor robotics computer. One of the most innovative elements is the capability to simulate human movements through imitation learning, where a small number of human demonstrations can be expanded into a vast synthetic dataset using teleoperation and random domain simulations. This allows robots to quickly learn and adapt their skills in virtual environments before being deployed in the real world.

The impact of this advancement goes beyond simply enhancing industrial operations. It promises to transform human-machine interaction, fostering more intuitive and efficient collaboration. In a scenario where humanoid robots are integrated into daily operations, new opportunities arise, not only in terms of operational efficiency but also in how humans interact with technology. This continuous and collaborative integration paves the way for a new era of innovation where robotic technology acts as an extension of human capabilities, improving quality of life and revolutionizing industries.