Nvidia presented a range of new technologies at its GTC 2025 event, focusing on advancements in GPUs, AI infrastructure, robotics, and quantum computing. The company introduced the GeForce RTX 5090, a graphics card built on the Blackwell architecture, featuring improvements in energy efficiency, size reduction, and AI-assisted rendering capabilities. Nvidia highlighted the increasing role of AI in real-time, path-traced rendering and GPU performance optimization.
In the data center sector, Nvidia announced the Blackwell Ultra GB300 family of GPUs, designed to enhance AI inference efficiency with 1.5 times the memory capacity of previous models. The company also introduced MVLink, a high-speed interconnect technology that enables faster GPU communication, and Nvidia Dynamo, an AI data center operating system aimed at improving management and efficiency. The DGX Station, a computing platform for AI workloads, was also unveiled to support enterprise AI development.
Nvidia’s automotive division revealed a partnership with General Motors to develop AI-powered self-driving vehicles. The company stated that all software components involved in the project have undergone rigorous safety assessments. Additionally, Nvidia introduced Halos, an AI-powered safety system for autonomous vehicles, integrating hardware, software, and AI-based decision-making.
In robotics, Nvidia introduced the Isaac GR00T N1, an open-source humanoid reasoning model developed in collaboration with Google DeepMind and Disney Research. The Newton physics engine, also open-source, was announced to enhance robotics training by simulating real-world physics for AI-driven robots. Nvidia also expanded its Omniverse platform for physical AI applications with the launch of Cosmos, a generative model aimed at improving AI-driven world simulation and interaction.
The company announced new AI models under the Llama Nemotron family, designed for reasoning-based AI agents. These models are optimized for enterprises looking to deploy AI agents that can work autonomously or collaboratively. Nvidia stated that members of the Nvidia Developer Program could access Llama Nemotron for development, testing, and research.
Nvidia revealed its latest advancements in quantum computing, including the launch of the Nvidia Accelerated Quantum Research Center in Boston. The company is collaborating with Harvard and MIT on quantum computing initiatives. During the conference, Nvidia hosted Quantum Day, where CEO Jensen Huang discussed the evolving role of quantum computing and acknowledged previous underestimations of its development timeline.
The company introduced a roadmap for future GPUs, including the Vera Rubin GPU, set for release in 2026, followed by Rubin Ultra NVL576 in 2027, which is projected to deliver 15 exaflops of computing power. Nvidia also announced the Feynman GPU, scheduled for 2028, designed to advance AI workloads with enhanced memory and performance capabilities.
Following the conference, online discussions reflected various perspectives on Nvidia’s announcements. Many users expressed interest in the concept of physical AI and its potential applications.
AI expert Armughan Ahmad shared:
AI is shifting from simple chat assistants to autonomous agents that execute work on our behalf.
While AI strategist Vivi Linsi commented:
With the arrival of agentic and physical AI, AI has developed from talking to Doing – are you ready?
Nvidia’s announcements at GTC 2025 highlight the company’s continued investment in AI, data centers, robotics, and quantum computing, positioning itself at the forefront of next-generation computing infrastructure.
