CEO of NVIDIA Corp, Jensen Huang, delivered an electrifying keynote address at the company’s highly anticipated GTC 2025. The CEO donning his iconic leather jacket, characteristic humour and technical mastery set the stage for the next frontier in artificial intelligence. The two-hour-long keynote address to a packed house brimming with developers, business leaders and various stakeholders highlighted the rapid pace of the AI revolution and how NVIDIA has been integral to this growth story. From bits about the earlier products of the chipmaker, pushing hardware limits, how it is enabling big tech to scale AI, and transforming the AI-driven enterprises, the keynote gave a glimpse of what lies ahead. Huang was speaking to an audience at the event that has been dubbed the Super Bowl of AI. The event saw an overwhelming participation of 25,000 in-person attendees at the venue in San Jose, California. In case you missed the keynote, here is a TLDR version of the address that showcased the trajectory of AI, flanked by some futuristic presentations and some surprises. Key takeaways from the address by Jensen Huang Physical AI: The next phase Huang's address touched upon the dawn of physical AI, which, according to him, represents the next major wave in AI innovation. He explained that "physical AI" is AI that comprehends physical concepts like friction, inertia, cause and effect, and object permanence. "The ability to understand the physical world, the three-dimensional world, is what's going to enable a new era of AI, which we call Physical AI, and it's going to enable robotics," he said. The CEO sees it as a transformative phase in AI that follows perception AI, generative AI, and agentic AI. Physical AI will be the foundation for robotics, marking a massive shift in AI applications. While earlier models mainly processed data and generated content, physical AI will allow machines to navigate and operate in real-world environments. Huang believes that physical AI, by advancing robotics, will bring in new industries and partners, further expanding the AI ecosystem. However, it is not an easy road ahead. The CEO also pointed out the massive computational challenges that prevail in developing physical AI. These challenges include scaling AI models to process real-world physics; training AI without human-in-the-loop limitations to accelerate learning; and efficiently using data to teach AI how to interact with the physical world. Huang has categorised physical AI as the next logical progression in AI development, indicating his optimism about the fusion of AI and robotics. GeForce 5090 and Blackwell Architecture Last year’s GTC conference centred around Nvidia’s Blackwell AI chips; this year too, it was among the main attractions in Huang’s address. The CEO presented an enhanced version of the chip named Blackwell Ultra. The Ultra is the latest in the company’s GPU evolution. The company introduced the GeForce 5090 that has been powered by the Blackwell GPU architecture. When compared to the earlier 4090, the new flagship GPU is 30 per cent more efficient, and Huang emphasised that it delivers remarkably improved AI-powered graphics and performance. On Gen AI’s evolution "AI is no longer just about perception; it is about creation." Huang traced the genesis of AI, which essentially moved from perception (computer vision and speech recognition) to generative AI (text-to-image, image-to-text, protein synthesis, etc.). He remarked that this transition from retrieval-based computing to generative computing signalled a fundamental shift in how we approached problem-solving. Vera Rubin Huang also unveiled the next generation of AI chips named Rubin that has been named after the astronomer who discovered dark matter. The GPU, which is expected to be shipped in 2026, has two components – a CPU called Vera and a new GPU design called Rubin. Vera is the company’s first custom CPU design, and it is expected to be twice as fast as the CPU on the Grace Blackwell chips from 2024. Vera and Rubin, when combined, can manage 50 petaflops during inference, which is more than the 20 petaflops on the current Blackwell chips. A petaflop, or PF, is a unit of measurement for computer processing speed. Rubin also comes with 288 gigabytes of fast memory. The rise of Agentic AI Much of 2025 has been about how agentic AI will be joining the workforce, kickstarting a plethora of use cases across industries. Huang too seemed bullish about the ongoing trend. "Agentic AI is the next leap, AI that perceives, reasons, and acts on its own," he said. He said that the world was at a $1 trillion computing inflection point. The CEO pointed out that AI computing demand has been accelerating rapidly, driven by the rise of reasoning AI and agentic AI. While discussing the future of agentic AI, the CEO also announced the open Llama Nemotron family of models with reasoning capabilities. These models have been designed to offer developers and enterprises a business-ready foundation for developing AI agents that can work autonomously or as connected teams to accomplish complex tasks. Huang said that the models which are built on Llama models are capable of delivering on-demand AI reasoning capabilities. The company has enhanced the new reasoning model family during post-training to improve multistep math, reasoning, coding, and complex decision-making. CUDA-X GPU and autonomous vehicles The CEO said that CUDA-X GPU-accelerated libraries and microservices are now powering industries across the spectrum, with CUDA reaching a tipping point in accelerated computing. He also revealed that the company will be open-sourcing its cuOpt decision optimisation platform. On the other hand, he highlighted how AI’s growing influence is extending to robotics, self-driving cars, and factories, with General Motors adopting NVIDIA AI, simulation and accelerated computing for next-gen vehicles and automation. The CEO also introduced NVIDIA Halos, a safety system that integrates NVIDIA’s automotive hardware, software and AI research in the safety of autonomous vehicles. Vision for AI and robotics Towards the end of his address, Huang showcased a host of announcements related to robotics. The CEO envisions robotics as a $10 trillion industry; he also claimed a shortage of 50 million workers. In order to accelerate robotics, NVIDIA has introduced Isaac GR00T N1, the first open and customisable foundation model for humanoid reasoning. The company also introduced new NVIDIA Cosmos world foundation models, offering greater control over AI-driven world generation. Using the Omniverse platform and Cosmos, NVIDIA is enabling infinite, controlled data creation. He also introduced the Newton open-source physics engine, developed with Google DeepMind and Disney Research. His address concluded with a live demo featuring Blue, a small AI-powered robot that resembles Disney’s Wall-E. The affable robot won applause and admiration.
