Nvidia GTC 2025 got underway with a compelling keynote from CEO Jensen Huang. This has started a new generation of artificial intelligence. Addressing a whole house, Huang announced that AI has come to “an inflection point.” In the last decade, AI transitioned from basic perception to generative AI and now agentic AI and it will be interesting to see what Nvidia GTC has.
New AI Chips: Blackwell Ultra and Rubin AI
One of the biggest news at Nvidia GTC 2025 was the announcement of Nvidia’s upcoming chip architectures. Huang revealed the Blackwell Ultra, which would come out in late 2025, followed by the Rubin AI chip with high performance in 2026. An even more advanced version, Rubin Ultra, will be available in 2027. All the new AI chips are directed towards enabling more complex calculations, thereby making AI models smarter and faster.
Nvidia expects its data center infrastructure business to skyrocket, projecting it will grow to $1 trillion by 2028. It is a mirror of the through-the-roof demand for GPUs by top cloud service providers. AI companies across the world are racing to adopt Nvidia’s cutting-edge technologies in an attempt to dominate the aggressive market.
The Rise of Physical AI and Robotics
Beyond generative AI, Huang referred to the coming breakthrough: “physical AI”-controlled robots. Physical AI understands real-world mechanisms such as friction, inertia, and object permanence which is contrary to traditional AI which operates on virtual data.
To speed up robotics growth, Nvidia unveiled Isaac GR00T N1, an open-source foundation model to help build humanoid robots. The model is compatible with Cosmos AI, which generates simulated training data for robots. Specialists believe that this technology will enable more efficient and economical training in robotics.
Benjamin Lee, an engineering professor at the University of Pennsylvania, highlighted the importance of synthetic data for training AI. “It’s expensive and time-consuming to train actual robots in the real world,” he said. “With Nvidia’s method of synthetic data, more researchers can experiment with AI, not just big tech firms.”
Omniverse and Cosmos: The Future of AI Training
Huang also introduced innovations in training AI. Nvidia’s Omniverse, a physics-based simulation platform, will now co-exist with the Cosmos AI model. This combination is able to render high-quality, low-cost photorealistic video, replacing traditional methods such as using human demonstrations or actual footage.
A massive industry shift is already in motion. US carmaker General Motors is to integrate Nvidia technology into its autonomous car fleet. The agreement will see the auto firm working with Nvidia to develop advanced AI systems using Omniverse and Cosmos.
Halos System for Automotive AI
Among the other major learnings of Nvidia GTC 2025 was the launch of Halos, a self-driving vehicle safety system that is powered by AI. Nvidia takes pride in being the first in the world to have its entire line of AI safety code thoroughly inspected. This system aims to advance the automation of cars while ensuring maximum safety standards.
Newton: The Open-Source Physics Engine
To push AI robotics to the next level, Nvidia launched Newton, an innovative physics engine that was developed in collaboration with Google DeepMind and Disney Research. Newton will make it possible to have realistic physics simulations, which will render AI models more realistic in real-world scenarios.
As Huang concluded his address, a small humanoid robot named Blue emerged on the stage. It beeped, followed his commands, and stood beside him—a clear indication that AI-powered robotics is no longer a fantasy.
The Future is Here
Nvidia GTC 2025 showcased some of the most pioneering AI breakthroughs ever. Nvidia is expected to shape the future of artificial intelligence. The industry is entering into an era where AI can reason, act, and learn like never before through reinforcement learning, Omniverse, and AI-driven automation.
