At the world’s largest technology exhibition, NVIDIA showcased several products and developments at CES 2025.
In addition to introducing its newest graphics cards, NVIDIA also showcased various new developments and technologies showcased at CES 2025.
Addressing the media, Jensen Huang, NVIDIA CEO, presented various other NVIDIA products, such as NVIDIA Cosmos, NVIDIA Omniverse, Isaac Groot, and Project Digits.
What are the latest technologies from NVIDIA at CES 2025? Read the following discussion to the end.
List of NVIDIA’s Latest Technologies Showcased at CES 2025
1. Blackwell Architecture
As the GPU architecture that powers the NVIDIA GeForce 5090, it incorporates numerous AI technologies.
Blackwell architecture as the basis for NVIDIA’s latest graphics cards – NVIDIA CEO at CES 2025
One of the most significant improvements is the ability to handle up to 4,000 AI TOPS of AI operations. This is a significant jump from the RTX 4090, which was capable of 1,300 AI TOPS.
AI TOPS, or AI Tera Operations Per Second, is a measure of how many trillions of operations a processor can perform in one second. This metric is crucial for assessing the performance of AI hardware, particularly Neural Processing Units (NPUs) and GPUs designed for AI tasks.
With better power efficiency and higher performance, Blackwell is designed to support intensive data processing and more complex AI models.
2. Agentic AI – The Future of AI Interaction with Workers
NVIDIA envisions that every office will eventually require AI technology. Every worker will be assisted by AI to assist them with their work.
To manage this, Agentic AI will help workers address various issues they encounter, including conducting automated monitoring.
Agentic AI Schematic – NVIDIA CEO at CES 2025
With Agentic AI, NVIDIA is taking AI to a more sophisticated level of interaction. This technology enables AI systems to communicate naturally with users, providing a more personalized and responsive experience.
Agentic AI Diagram – NVIDIA CEO at CES 2025
Agentic AI is expected to transform the way companies interact with customers, including deciding on actions based on KPI (Key Performance Indicator) simulations.
Agentic AI’s ability to learn and adapt to user preferences makes it a critical asset for companies focused on AI-based customer service.
3. NVIDIA Omniverse, Create Digital Twin Simulations
During research and development, developers are already accustomed to conducting in-depth simulations through applications like SolidWorks, which has CFM simulations.
Just imagine these simulations applied on a larger scale, with more variables to produce increasingly accurate and precise results.
Through digital twin simulation, real-world physical products, systems, or processes can be accurately analyzed. This technology also allows for monitoring and maintenance without the need for direct inspection.
With its simulation capabilities and domain randomization, Omniverse offers a highly scalable training environment for real-world applications.
Developers can use NVIDIA Omniverse to build geospatially accurate, physics-based scenarios.
4. NVIDIA Cosmos – Physical AI
What if the entire real world were simulated in AI? NVIDIA Cosmos seeks to address this challenge.
NVIDIA Cosmos is a comprehensive platform designed for the development of physical AI, such as robots and autonomous vehicles.
This form of AI modeling is called World Foundation Models (WFMs), which simulate everything in the world, including the environment and its interactions.
Cosmos will utilize the AI and CUDA pipeline, enabling developers to process large amounts of data using the NVIDIA Blackwell platform.
NVIDIA Cosmos includes:
- Automatically derived regressive World Foundation Models
- Diffusion-based World Foundation Models
- Advanced tokenizer and Nvidia Cuda-accelerated data pipeline
Cosmos models ingest text images or video requests and output the world state as a video.
Cosmos prioritizes the unique requirements of AV and robotic use cases, such as real-world environmental lighting and object permanence awareness.
With its ability to understand physical dynamics and spatial relationships, Cosmos enables realistic simulations and the development of more sophisticated robotics applications. This technology opens up new opportunities for more natural human-machine interactions.
5. Thor Processor – Processor for Self-Driving Cars
To support the growing autonomous vehicle industry, NVIDIA introduced the Thor processor, the latest generation of its vehicle processor.
Thor Processor in the Hands of NVIDIA CEO – NVIDIA CEO at CES 2025
With 20 times the processing power of the previous generation, Thor can handle a wide range of sensor data in real time, enabling safer and more efficient autonomous driving technology.
The Thor processor is capable of processing input from multiple sensors, including high-resolution cameras, lidar, and radar. This input is converted into tokens for input into the transformer.
Sensor input for the brain of an autonomous vehicle – NVIDIA CEO at CES 2025
With its numerous sensors, Thor has the potential for significant safety. Its safety dedication could even surpass that of a human driver, who has limited focus and blind spots when driving.
Its safety standards have even earned ISO 26262 and ASIL-B/D certification, the highest functional safety standards for cars.
Sensor data and geospatial data are combined with an omni-map to build a 3D environment. Variation in driving scenarios can then be generated from drive logs or an AI traffic generator.
The neural reconstruction engine uses the autonomous vehicle’s sensor logs to create a high-fidelity 4D simulation environment, replaying previous drives in 3D, generating scenario variations to strengthen the training data.
Finally, Edify 3DS automatically searches through existing asset libraries or creates new assets to create simulation-ready scenes.
This processor is expected to be the brain of smart cars from Toyota, Tesla, BYD, Mercedes, WHMO, ZUK, and JLR.
Even Volvo, Lucid, Rivian, and Aurora will develop self-driving trucks based on the Thor processor.
This robotics processor can also be used to create full-fledged robots. These could be AMRs, humanoid robots, brains, or manipulators.
6. Isaac Groot – Multipurpose Robot
What if the Thor chipset were used to create multipurpose robots?
In an effort to advance the development of humanoid robotics, NVIDIA introduced the Isaac Groot platform.
NVIDIA’s Isaac Groot humanoid robot can be trained to perform a variety of human tasks – NVIDIA CEO at CES 2025
This platform provides tools for creating multipurpose robots by leveraging synthetic data and simulated workflows.
With Isaac Groot, developers can improve robot development efficiency through simulation-based training.
NVIDIA diagram: Isaac Groot – NVIDIA CEO at CES 2025
To train this humanoid robot to perform a task, a skilled worker will first use an Apple Vision Pro and enter a special simulation room.
This worker will act as the robot to demonstrate its work. Its movements will be recorded as data. This allows operators to capture data without a physical robot.
This movement data can then be stored and replicated using GrootMimic to create a series of desired movements.
Ultimately, the humanoid robot can perform repetitive and high-risk tasks.
7. Project Digits – An AI Supercomputer the Size of a Mini PC
Project DIGITS is making headlines by presenting an AI supercomputer that’s as compact as a Mini PC but with the power of a supercomputer.
Project Digits is the size of a mini PC but has the AI capabilities of a supercomputer from NVIDIA – NVIDIA CEO at CES 2025
This AI supercomputer runs NVIDIA’s entire AI stack. All NVIDIA software runs on it. DGX Cloud runs on top of it. It resides somewhere, wirelessly, or connected to your computer. It can even be a workstation, accessible like a cloud supercomputer.
Based on a state-of-the-art 110GB chip, DIGITS is designed to deliver powerful AI computing capabilities and make them accessible to a wider range of users in fields ranging from research to creative content production.
Project Digits has a compact form factor, just the size of a mini PC – NVIDIA CEO at CES 2025
You can use it as a Linux workstation if you like. If you use a PC, a Mac – anything – it can act as a cloud platform.
Even more impressively, you can connect it with ConnectX, and it has a built-in NIC and GPU Direct. This includes connecting two Digits as a stack.
This AI supercomputer is expected to be available to the market in May.
Final Words
These are some of the NVIDIA technologies at CES 2025 that will bring advancements to various aspects, from gaming, office work, industry, research and development, to self-driving cars.
With these innovations, NVIDIA is not only strengthening its position as a pioneer in graphics and AI technology but also paving the way for advanced applications in the future.
Each product introduced demonstrates the immense potential to empower various industries, create richer user experiences, and address challenges in the digital age.