Jensen Huang Unveils NVIDIA's AI Revolution: From Quantum Computing to Agentic AI and Industrial AI Factories

Jensen Huang opens with a poetic vision of intelligence as a factory producing tokens—the fundamental units of AI. He emphasizes how tokens have unlocked new frontiers, transforming images into scientific data that help explore alien atmospheres, detect hidden dangers on Earth, and even predict diseases before they manifest. This metaphor sets the stage for understanding AI as a generative process that fuels innovation across many fields.

"Tokens have opened a new frontier... Tokens transform images into scientific data... They see disease before it takes hold." 1:04
"This is how intelligence is made. A new kind of factory. Generator of tokens." 0

Key points:

• Tokens are the building blocks of AI intelligence.
• AI’s impact spans from space exploration to healthcare.
• Intelligence is framed as a manufacturing process, producing valuable outputs.

Jensen recounts NVIDIA’s pioneering work in accelerated computing, starting with CUDA libraries that reformulate algorithms for parallel processing on GPUs. He highlights a broad ecosystem of over 400 specialized libraries, each accelerating different domains—from computational lithography in semiconductor fabs to genomics and weather modeling. The introduction of CUDA-Q marks NVIDIA’s entry into quantum computing acceleration, bridging classical and quantum systems.

"You have to reformulate your algorithm... and so we created libraries to help." 5:29
"CUDA-Q is for quantum computing... quantum classical computing based on GPUs." 9:13
"Our entire quantum algorithm stack is now accelerated on Grace Blackwell 200 and the speed up is utterly incredible." 11:45

Key points:

• Accelerated computing requires new software paradigms, not just hardware.
• NVIDIA’s CUDA-X suite includes domain-specific libraries enabling breakthroughs in science and industry.
• CUDA-Q extends GPU acceleration to quantum-classical hybrid computing.
• Grace Blackwell 200 is a key platform for quantum algorithm acceleration.

Jensen introduces the concept of Agentic AI—AI systems that don’t just generate text or images but can reason, plan, and act autonomously, including physical locomotion and tool use. This marks a shift from generative AI to embodied AI, where robots become intelligent agents capable of interacting with the real world. He connects this evolution to NVIDIA’s roots in GeForce graphics, which laid the foundation for accelerated AI computing.

"Agentic AI... the embodiment of that Agentic AI... the ability to walk or reach out and grab something, use tools." 16:09
"The fundamental technology to enable agents, which are basically information robots and embodied AI, physical robots." 16:33
"Everything I'm going to show you today... is computer simulation, not animation." 17:24

Key points:

• Agentic AI represents a new wave of AI focused on reasoning and physical interaction.
• Robotics and embodied AI are natural extensions of generative AI capabilities.
• NVIDIA’s graphics and simulation technologies underpin this evolution.

Jensen unveils Grace Blackwell NVL72, a monumental leap in AI hardware—a “thinking machine” designed as one giant virtual GPU composed of 72 interconnected Blackwell dies. This system replaces traditional CPU-GPU setups with a tightly integrated architecture connected by NVLink, a high-bandwidth, low-latency interconnect fabric. The Blackwell system delivers 30-40x performance improvements over previous generations, enabling complex reasoning models that generate vastly more tokens during inference.

"It was designed to be one giant GPU... one giant virtual GPU." 22:22
"NVLink spine... 130 terabytes per second bandwidth... more than the world's entire internet traffic." 25:47
"Blackwell can generate a ton more revenues as a result of the architecture." 29:15

Key points:

• Grace Blackwell NVL72 integrates 72 GPU dies into a single virtual GPU.
• NVLink interconnect provides unprecedented bandwidth, surpassing global internet traffic.
• Designed to support reasoning AI models that require massive token generation and throughput.
• Represents a generational leap in AI supercomputing performance.

Jensen explains NVIDIA’s unified architecture that spans cloud AI, enterprise AI, personal AI, and edge AI. He introduces the RTX Pro Server, an enterprise-grade system that supports a wide range of workloads—from AI to Omniverse simulations to traditional enterprise applications—running on a single platform. This system bridges AI-native architectures with traditional IT environments, supporting Linux, VMware, Kubernetes, and more.

"This server is the only server in the world that runs everything the world has ever written and everything NVIDIA has ever developed." 37:31
"The architecture of the systems, the architecture of the software are completely different." 39:13
"One architecture - From Cloud AI, Enterprise AI, Personal AI, to Edge AI." 34:14

Key points:

• NVIDIA offers a single architecture adaptable to diverse AI and enterprise workloads.
• RTX Pro Server integrates AI-native and traditional IT stacks seamlessly.
• Supports broad software ecosystems including Windows, Linux, Kubernetes, and VMware.

Jensen frames AI data centers as “AI factories” designed solely to produce intelligent tokens at scale, fundamentally different from traditional data centers. These factories are revenue-generating facilities, akin to car factories, and are becoming critical infrastructure for countries and companies. He highlights Europe’s rapid expansion of AI infrastructure, with plans to increase AI computing capacity tenfold in two years, and the establishment of AI technology centers across seven European countries.

"These AI data centers are not data centers in the classical sense... They have one job and one job only: to produce intelligent tokens." 39:56
"AI factories are now part of a country's infrastructure." 41:27
"In total in just two years we will increase the amount of AI computing capacity in Europe by a factor of ten." 44:16

Key points:

• AI factories are specialized data centers focused on token generation for AI workloads.
• They represent a new industrial revolution, transforming economies and infrastructure.
• Europe is aggressively building indigenous AI infrastructure and ecosystems.
• NVIDIA is partnering with governments and companies to accelerate this growth.

Jensen highlights NVIDIA’s deep collaborations with European companies and governments, including Siemens, Schneider Electric, and startups like Mistral. He stresses the importance of sovereign AI models that respect data ownership and cultural context, supported by NVIDIA’s Nemotron platform which enhances open-source models with reasoning and extended context capabilities. Partnerships with companies like Perplexity enable regionalized AI applications tailored to local languages and needs.

"Your data belongs to you. It is the history of your people, the knowledge of your people, the culture of your people." 52:54
"We're announcing that we're going to build an AI cloud together here to deliver their models as well as deliver AI applications for the ecosystem." 49:02
"Nemotron models have top of the leaderboard performance." 52:05

Key points:

• Emphasis on data sovereignty and regional AI model customization.
• Nemotron enhances open models with advanced reasoning and context.
• Strong partnerships with European industry leaders and startups.
• Building AI clouds and ecosystems tailored for European markets.

Jensen showcases Agentic AI as a leap beyond one-shot generative models, capable of multi-step reasoning, planning, and tool use. He presents a demo where AI agents collaboratively plan and execute a complex task—starting a food truck in Paris—by breaking it down into market research, menu design, financial planning, and marketing. NVIDIA’s AI agent platform provides tools to build, train, guardrail, and deploy such agents securely across clouds and on-premises environments.

"Agentic AI is a giant step function from one-shot AI." 56:20
"One prompt... must have generated 10,000 times more tokens." 59:16
"We created a platform for you... a framework, and a set of tools... to build those agents." 1:00:11

Key points:

• Agentic AI enables autonomous, multi-agent problem solving.
• NVIDIA provides a full software stack for agent development and lifecycle management.
• Agents can be deployed flexibly across public clouds, private clouds, and edge devices.
• Demonstrates practical applications of AI agents in business workflows.

Jensen introduces DGX Spark, a compact AI supercomputer designed for desktops, and DGX Cloud Lepton, a cloud orchestration platform that connects developers to a global network of GPUs across multiple clouds and regions. Lepton enables seamless deployment and scaling of AI models and agents, integrating with platforms like HuggingFace for one-click deployment. This infrastructure supports developers wherever they are, accelerating AI innovation globally.

"DGX Spark enables you to have essentially the Grace Blackwell system on your desktop." 35:36
"DGX Cloud Lepton provides on-demand access to a global network of GPUs across clouds, regions, and partners." 1:07:21
"Lepton is a cloud of clouds... one model architecture, one deployment, and you can run it everywhere." 1:03:28

Key points:

• DGX Spark brings supercomputer power to individual developers.
• DGX Cloud Lepton orchestrates AI workloads across multiple cloud providers and private infrastructure.
• Integration with HuggingFace simplifies model training and deployment.
• Enables scalable, flexible AI development and inference worldwide.

Jensen celebrates NVIDIA’s longstanding presence in Europe and its partnerships with industrial giants like Siemens, BMW, Mercedes-Benz, and Toyota. He highlights the use of Omniverse for creating photorealistic digital twins of factories, warehouses, and infrastructure, enabling design, simulation, and optimization before physical deployment. He announces the world’s first industrial AI cloud in Europe, designed for real-time simulation and AI-driven robotics training, marking a new era in manufacturing and industrial automation.

"We're partnering with Siemens on everything from design to simulation to digital twins of factories." 1:12:11
"Omniverse was built so that digital twins look real and behave realistically for robots to learn." 1:17:41
"We're announcing today that we are going to build the world's first industrial AI cloud here in Europe." 1:18:30

Key points:

• Industrial AI revolution leverages digital twins and AI to transform manufacturing.
• Omniverse enables physics-accurate simulations critical for robotics and factory design.
• The industrial AI cloud will support real-time design, simulation, and robot training.
• Strong European industrial partnerships accelerate AI adoption in manufacturing.

Jensen details NVIDIA’s Drive platform for autonomous vehicles, emphasizing safety through the Halos system, which integrates diverse sensors, redundant compute, and emergency protocols. He highlights the use of synthetic data generated in Omniverse for training AV models to handle edge cases. The Isaac platform supports robotics development with hardware, software, and simulation tools, enabling humanoid robots to learn in virtual environments obeying physics laws before operating in the real world.

"NVIDIA Drive built on the Halos safety system lets developers build safe autonomous vehicles." 1:22:08
"Developers use NVIDIA Omniverse and Cosmos to generate realistic synthetic training data." 1:22:28
"Isaac Open Robotics Development Platform... robots learn in a virtual world that obeys the laws of physics." 1:28:58

Key points:

• Drive platform integrates AI, safety, and redundancy for autonomous vehicles.
• Synthetic data from Omniverse enhances AV training diversity and robustness.
• Isaac platform accelerates robotics development with simulation and AI.
• Humanoid robots learn locomotion and manipulation in physics-accurate virtual worlds.

Jensen closes by reflecting on the dawn of a new industrial revolution powered by AI. He underscores the explosive growth in AI inference workloads and the necessity of specialized “thinking machines” like Blackwell to handle reasoning tasks. He expresses excitement about Europe’s commitment to AI infrastructure and innovation, thanking partners and encouraging the audience to embrace the transformative potential of AI. The keynote ends on a warm, personal note with the robot Grek, symbolizing the future of embodied AI.

"The next wave of AI has started... It requires inference workloads to explode." 1:35:09
"These Blackwells will go into new types of datacenters, essentially AI factories." 1:35:41
"I'm so happy to see that Europe is going all in on AI." 1:35:57

Key points:

• AI inference demand is growing exponentially, driving need for advanced hardware.
• Blackwell systems are designed specifically for reasoning and token generation.
• Europe is rapidly expanding AI infrastructure and ecosystem.
• The future includes widespread embodied AI and intelligent agents transforming industries and daily life.

---

This keynote by Jensen Huang at GTC Paris 2025 offers a sweeping narrative of NVIDIA’s technological innovations and strategic vision, from foundational GPU acceleration to quantum computing, agentic AI, industrial digital twins, and autonomous systems. It highlights the convergence of hardware, software, and ecosystem partnerships that are enabling a new era of AI-driven industrial and societal transformation, with Europe playing a central role in this unfolding story.