At GTC 2025, NVIDIA CEO Jensen Huang introduced groundbreaking advancements in AI, robotics, and computing. He highlighted AI’s role in graphics, autonomous systems, and data centres, unveiling the GeForce 5090, Blackwell architecture, and AI factories.
GTC NVIDIA
On March 18, 2025, Jensen Huang, Nvidia’s CEO, led a virtual GTC 2025 AI Conference in San Jose, California.
He presented a keynote that outlined some of the most significant developments in AI, robotics, and accelerated computing while announcing that NVIDIA is at the forefront of innovation in technology.
This article will highlight his keynote in detail, reporting on all of the announcements and innovations that are a breakthrough in the future of AI.
Jensen Huang opened the conference with the light-hearted comment that he was speaking freely and without a script or teleprompter to emphasise the genuineness of Nvidia’s announcements.
He explained that these tokens enable AI to convert images into scientific information, which has critical uses in areas such as disease prognosis and space exploration.
Additionally, he highlighted how AI is improving automation and enhancing human experiences by bridging different industries and disciplines.
Huang then introduced the GeForce RTX 5090, describing it as a leap in power efficiency and size.
‘AI is transforming computer graphics. Both real-time ray tracing and predictive rendering are accelerating image quality while restoring rendering speeds,’ he said.
He shared insights about generative AI and how it represents an exciting shift from the computing we’re familiar with today—where we mainly retrieve stored data—to a more dynamic model that empowers AI to generate context-driven content on the fly.
While addressing the idea of how AI has evolved, Huang presented a concept he referred to as Agentic AI, which aims to reason, plan, and utilise tools to complete tasks in a more efficient manner.
He pointed out that this kind of AI can understand and interact with the real world, which opens up tons of possibilities for robots and automation.
But he also acknowledged that there are still some hurdles, like getting data, training AI without people having to constantly, and making the most of computing resources. These things are still stopping AI from getting even bigger and better.
He discussed that AI is now facing an exponential increase in computational demands, requiring 100x more computing power than before.
He explained that reinforcement learning is crucial in this development, allowing AI models to improve over time through trial and error using synthetic data.
Huang also highlighted that major cloud providers (CSPs) are expanding their AI computing capacity at an unprecedented rate to keep up with the rising demand.
He introduced the idea of AI factories, explaining that these are not just traditional data centres but specialised facilities designed to generate intelligence rather than just store information.
He noted that NVIDIA’s CUDA libraries, including CUDA X, are playing a critical role in accelerating AI workloads across various industries like physics, biology, and telecommunications.
He proudly stated that 6 million developers across 200 countries are using CUDA, proving its widespread impact.
Huang emphasised that AI is now deeply integrated into various industries, including manufacturing, robotics, and self-driving vehicles.
He announced a partnership between NVIDIA, Cisco, and T-Mobile to enhance AI-powered communications networks.
He also highlighted NVIDIA’s collaboration with GM, which is focused on advancing AI in autonomous driving, in-vehicle systems, and enterprise applications.
Safety, he said, remains a top priority, with NVIDIA implementing strict assessments and transparency measures in AI-powered automotive technologies.
Addressing the future of AI infrastructure, Huang spoke about the challenge of balancing token generation speed while maintaining high-quality AI outputs.
He then unveiled Blackwell, NVIDIA’s next-generation GPU architecture, which he claimed delivers 40 times the performance of the previous Hopper architecture.
He confidently stated that Blackwell would redefine efficiency in AI computing, making older models obsolete.
Looking toward the future, Huang shared that NVIDIA is building digital twins of AI factories in the Omniverse, allowing companies to optimise AI workflows before physical deployment.
He also introduced Blackwell Ultra, an enhanced version of the Blackwell GPU, which will push the boundaries of AI performance even further.
He then teased upcoming projects like Vera Rubin and Feynman, named after renowned scientists, signalling NVIDIA’s ongoing commitment to scientific breakthroughs powered by AI.
Huang predicted that in the near future, all software engineers will work with AI assistants, completely transforming how software is developed.
He announced Spectrum X, a high-performance AI computing platform designed to help enterprises transition into AI-driven operations.
He also introduced co-packaged silicon photonics, a breakthrough technology aimed at drastically reducing energy consumption in data centres while increasing efficiency.
Huang closed his keynote by discussing AI’s role in robotics, stating that AI-powered robots will soon help address labour shortages by performing complex real-world tasks.
He unveiled Isaac Groot N1, a new AI-powered humanoid robot system designed with advanced perception capabilities and rapid response mechanisms.
He also introduced Newton, a physics engine co-developed with DeepMind and Disney Research that enables highly realistic robotic simulations.
Huang concluded on a positive note, stating that the open-source release of Groot N1 will transform the future of AI-driven robotics.
Also read: Mira Murati’s AI business is hoping to raise a whopping $2 billion in seed funding
This post was last modified on April 12, 2025 5:13 am
Pick your task, get the best AI model for it — images, video, slides, research,…
Learn what Agentic AI is, how it works, and how it differs from Generative AI.…
Discover the 13 best free online vocal remover AI tools for 2026, designed to isolate…
Explore the top 13 yield farming platforms for 2026, featuring secure, trusted, and high-APY crypto…
Explore the best AI learning platforms for 2026, including Coursera, edX, Udacity, and more. Learn…
Explore the 13 best Polygon wallets in 2026, comparing security, DeFi access, hardware and mobile…
