This week, from March 17 to March 21, the largest Nvidia GTS 2025 chip manufacturer conference was held in San Jose, USA, dedicated to innovations in the field of artificial intelligence. It was attended by 25,000 representatives from more than a hundred companies. Among them are Waymo, Microsoft and Ford. During the conference, Nvidia makes it clear that the AI revolution will happen soon and will affect all areas of people's lives.
Finam described the main novelties of GTC 2025.
AI robots
For many years, AI has existed in the form of chatbot software, automation tools, and content generators. But GTC 2025 has shown that AI is also penetrating the physical world.
For example, Nvidia introduced a vision-action VAM AI model. This model does not just analyze information, but perceives it and acts in the real world.
Robots will become another manifestation of AI in the real world. Businesses are already preparing to automate processes in manufacturing, logistics, and retail.
During the conference, Nvidia unveiled the Nvidia Isaac GR00T N1, the world's first open and fully customizable basic AI model for creating humanoid robots. The company will pre-train the GR00T N1 on synthetic and real data and release a range of models for robotics developers.
Company CEO Jensen Huang also demonstrated the 1X NEO Gamma humanoid robot, which is capable of doing cleaning. It was trained using the GR00T N1 model.
Blackwell Ultra and Vera Rubin AI chips
Nvidia has announced new chips for creating and deploying artificial intelligence models. Blackwell Ultra is the continuation of the range of Blackwell chips. According to developers, the novelty will be able to generate more tokens per second, which means more content.
Nvidia says cloud service providers will be able to use Blackwell Ultra for premium AI services for applications requiring rapid response, allowing them to generate 50 times more revenue from the new chips compared to the Hopper generation released in 2023.
Blackwell Ultra will come in two versions: the GB300 with two Nvidia Arm processors and the B300 with just the GPU.
It is expected to be available as early as the second half of the year.
In addition, Nvidia plans to begin shipping systems based on next-generation GPUs in the second half of 2026.
The system consists of two main components: the Vera processor and the new Rubin GPU. It is named after American astronomer Vera Rubin.
The combined superchip will replace the current Grace Blackwell at the top of Nvidia's AI processor hierarchy. According to the company, Vera is Nvidia's first proprietary processor based on an architecture they have named Olympus. Nvidia officials claim that Vera will be twice as fast as the Grace Blackwell processor used in last year's chips. The superchip can also support up to 288 gigabytes of fast memory, which is one of the main features AI developers are looking at.
In the second half of 2027, according to Huang, there will be a Rubin Ultra superchip that will combine multiple GPUs.
It is also known that Nvidia's next superchip will be called Feynman and will appear in 2028.
AI supercomputers
Nvidia has unveiled the Nvidia DGX personal supercomputers based on the Blackwell AI platform. DGX Station and DGX Spark. They will enable AI developers, researchers and data scientists to develop prototypes, customize and output large models on desktops. Users can run these models locally or deploy them on the NVIDIA DGX Cloud or any other accelerated cloud or data center infrastructure. DGX Spark and DGX Station harness the power of the Grace Blackwell chip, which was previously only available in data centers.
DGX Spark is the world's smallest artificial intelligence supercomputer. It is powered by the NVIDIA GB10 Grace Blackwell superchip, which features a powerful NVIDIA Blackwell GPU with fifth-generation tensor cores and FP4 support, providing up to 1,000 trillion operations per second for computing with the latest AI models, including the NVIDIA Cosmos Reason base world model and the NVIDIA GR00T N1 robot base model.
DGX Station is the first desktop system to utilize the NVIDIA GB300 Grace Blackwell Ultra Desktop superchip. It delivers PC-level data center performance. DGX Station is the first desktop PC to utilize NVIDIA's GB300 Grace Blackwell Ultra Desktop superchip. It features 784 GB of coherent memory to accelerate large-scale training and data analysis tasks.
Nvidia Dynamo and Llama Nemotron
In addition to processors, Nvidia unveiled Nvidia Dynamo, an open source software library available for AI developers. According to Huang, it is essentially an artificial intelligence factory operating system designed to efficiently accelerate and scale AI logic output models.
Nvidia also announced a new family of Llama Nemotron reasoning models that will allow developers and companies to create their own AI agents that are more advanced than current ones. The new reasoning models are also open source.
