The stock jumped: the CEO of Nvidia revealed a series of new products tonight

The chip maker Nvidia It is among the interesting companies appearing at the annual CES exhibition. Every year CEO Jensen Huang takes the stage and launches the company’s big announcements. As part of this year’s exhibition, Nvidia held the first press conference that officially opened the event. A few hours before the press conference, Nvidia’s stock jumped about 4%, in anticipation Why the company will present on stage.

The launch event opened with a video that introduced many companies and their developments. For example, they presented: a robot that prepares alcoholic beverages, a robot that helps children get vaccinated and talks to them, etc. All technologies presented were developed or related in one way or another to Nvidia. Immediately afterwards, Jensen Huang took the stage wearing a black sequined jacket. Huang presented the history of Nvidia since 1993, when Nvidia started out with a processor that later became a GPU.

The event opened with the company’s new graphics card based on the new architecture, GeForce RTX Blackwell. Huang explained that it is a video card with 92 billion transistors. What amazed the crowd were the prices: the RTX 5070 model with 1,000 AI TOPS capabilities will cost $549. The most expensive model of all, RTX 5090 with capabilities of 3,400 AI TOPS will cost 1,999 dollars. It should be noted that the products will be available to the public already this month.

Huang even referred in a short sentence to the company’s collaboration with the Toyota automobile company, in the development of the next generation of the company’s vehicles – but did not expand on this further. Also, during the event, Huang introduced a desktop personal computer as well as a platform to accelerate the development of robots and autonomous cars.

 

NVIDIA CEO Jensen Huang shows off the GeForce RTX 5090 Blackwell GPU with 92 billion transistors and 4 petaFLOPS of AI compute pic.twitter.com/IzuQ2btXo9

– Tsarathustra (@tsarnick) January 7, 2025

Project DIGITS : a personal supercomputer for artificial intelligence

One of Nvidia’s key announcements from the press conference is NVIDIA Project DIGITS, a desktop computer for artificial intelligence that will help developers, researchers, scientists and students access the processing power of the NVIDIA Grace Blackwell platform – which combines GPU and CPU components. The computers include the company’s new superchip, GB10 Grace Blackwell, which offers peta-flops (10 15) with artificial intelligence performance and is designed for the development, optimization and running of large AI models of up to 200 billion parameters. It was also announced that thanks to the company’s communication technology, ConnectX, embedded in computers, it will be possible to connect two supercomputers and it will be possible to run models of up to 405 billion parameters.

For example, users will be able to develop and run artificial intelligence models on their personal desk, and connect to accelerated computing infrastructures in the cloud or in a data center. Users will be able to access an extensive library of artificial intelligence software from Nvidia in order to experiment and develop prototypes and more.

Each such unit includes 128 gigabytes of shared memory and up to 4 terabytes of NVMe storage. The company even boasts that the computer is able to provide such strong performance using only a standard power outlet. According to the company, the new computer will be available for purchase from Nvidia and leading partners starting in May 2025, at a price of $3,000 per computer.

NVIDIA Cosmos: A Modeling Platform for Robots and Autonomous Cars

Another major announcement that Nvidia launched is a new software platform that offers groundbreaking World Foundation Models and other resources – where the goal is to promote and improve the development, training and testing of physical artificial intelligence systems such as autonomous vehicles and robots. World models are artificial intelligence models that understand the dynamics of the real world, including physics and spatial properties. These models know how to predict dynamics such as movement, force and spatial relationships.

These models do this based on texts, images, video and sensor data. Developers using such models can build virtual simulations in order to optimize and accelerate the adoption processes of autonomous machines – so that they can “understand and feel” the real world. For example, to train an autonomous vehicle on the streets of the city of Herzliya, without it actually doing so “in the real world”.

The company also launched the NVIDIA Cosmos platform and it even offers these models in question. The platform includes safety tools for artificial intelligence models and an accelerated work environment for video processing – and the platform comes in an open model so that developers can adapt the models to their needs according to the requirements of their target application.

Huang also announced on stage that Nvidia’s next step is to integrate the digital twin platform of the Omniverse (Nvidia’s Metaverse) with physical AI applications. Using this technology, it will be possible to develop autonomous machines and robots, and this will actually be the next era of industrial AI. The company noted that artificial intelligence models that obey the laws of physics are expensive and complex, as they require huge amounts of real-world data and a lot of computing power. The models that Nvidia offers allow developers to easily generate large amounts of physics-based synthetic data to train and adapt their existing models. Huang added that “we created NVIDIA Cosmos to democratize the field of Physical AI, and make the field of robotics accessible to every developer.”

Agentic AI : Smart agents

The CEO of Nvidia also referred to the announcements concerning the next generation of artificial intelligence and the personalized artificial intelligence “agents”. In order for these agents to work well, the systems need a variety of creative artificial intelligence models, so the company announced a new family of large language models Based on Llama, these models will help developers create and deploy AI agents in various applications, such as customer service, Fraud detection and more.

The company is launching three sizes to allow developers to implement agents: Nano (a model with high cost efficiency but appropriate cost, low response time suitable for smartphones and laptops); Super (high traffic rate on a single GPU); and Ultra (the highest level of accuracy, data center scale and require the most powerful performance). The new family of models was developed, among others, by employees of the AI ​​Software group in Nvidia Israel.

Among the examples given on stage was the sports market, a market that generates 500 billion dollars worldwide, with growth of hundreds of billions. In the coming years, coaches of teams and leagues will rely on game video analysis to improve player performance. With the help of the artificial intelligence agents that have visual perception, athletes will be able to gain access to deeper insights to improve their sports performance.

***Full disclosure: The reporter was a guest of CTA, the governing body of CES