LIVE: Nvidia CEO Jensen Huang speaks at Computex

Key Concepts

• Agentic AI: A computing paradigm where AI models (Large Language Models) act as "brains" within a "harness" (orchestration layer) to reason, plan, use tools, and perform productive work.
• AI Factories: Large-scale, highly complex infrastructure designed to generate "tokens" (the unit of AI output/revenue) with maximum efficiency.
• Vera Rubin: NVIDIA’s next-generation, multi-rack, pod-scale supercomputing system designed specifically for Agentic AI.
• Vera CPU: A new CPU architecture built from the ground up for agents, prioritizing single-threaded performance, low latency, and high bandwidth over traditional core-count virtualization.
• Extreme Co-design: The methodology of designing chips, racks, networking, and software stacks simultaneously to ensure maximum throughput and reliability.
• Cosmos 3: A frontier open-model for Physical AI, enabling robots to understand, reason, and generate actions in the physical world.
• RTX Spark: A new line of PCs (laptops, desktops, workstations) reinvented for the age of agents, featuring integrated NVIDIA AI hardware.

---

1. The Arrival of Agentic AI

Jensen Huang declared that "useful AI" has arrived. The shift is moving from simple generative AI to Agentic AI, where software agents orchestrate models, memory, and tools to perform complex tasks.

• Economic Impact: Using GitHub as a case study, Huang noted that 30–40 million software developers are seeing their output triple due to AI. This productivity boost transforms AI from a cost center into a GDP generator.
• The Agentic Pattern: An agent consists of a Large Language Model (LLM) for thinking, a "harness" for orchestration (acting as an operating system), and tools (like CUDA X libraries) for execution.

2. Vera Rubin: The Infrastructure for Agents

Vera Rubin is NVIDIA’s most ambitious system, designed to handle the disaggregated and distributed nature of agentic workloads.

• Technical Specifications: It features 6 trillion transistors per board, 18,000 components, and utilizes liquid cooling at 45°C.
• Manufacturing Efficiency: Through extreme co-design, assembly time for a rack has been reduced from two hours to five minutes.
• Confidential Computing: The system ensures data is encrypted at rest, in motion, and in use, which is critical for enterprise AI.

3. Vera CPU: Built for Agents, Not Humans

Traditional CPUs were designed for human users (virtualization/multi-tenancy). Vera CPUs are designed for "impatient" agents that require nanosecond-level responsiveness.

• Key Performance Metrics:
  • Instructions Per Clock (IPC): 10 instructions fetched/decoded per clock.
  • Bandwidth: 3.6 TB/s fabric bandwidth; 1.2 TB/s LPDDR5X memory bandwidth.
  • Efficiency: Designed to maximize "tokens per watt," as compute is now synonymous with revenue.
• Performance: Vera delivers 1.8x the agentic sandbox performance of traditional x86 CPUs and 3x–6x speedups in real-world workloads like SQL and stream processing.

4. Physical AI and Robotics

Huang emphasized that while language models learn from internet text, physical AI requires a new kind of data: first-person perspective data.

• Cosmos 3: An open-frontier model for physical AI that uses a mixture of transformers to understand, simulate, and predict physical world actions.
• Isaac Groot: A reference humanoid robot platform. It provides a fully integrated stack (simulation, data generation, and runtime) to accelerate robotics research, featuring 31 degrees of freedom and 6 feet of height.

5. Reinventing the PC: RTX Spark

In partnership with Microsoft, NVIDIA is reinventing the PC for the first time in 40 years.

• RTX Spark: A new chip (N1X) combining Blackwell GPU cores and Grace CPU cores. It is designed to run local agents, allowing users to have a "personal AI" that works 24/7.
• Use Case: Huang demonstrated an agent designing a house in real-time, moving files between Rhino, Blender, and generative AI models, showcasing how agents simplify complex workflows.

6. Notable Quotes

• "Compute is revenue. Performance per watt is your revenue." — Jensen Huang, on the economics of AI factories.
• "All of the CPUs until now were created for people... This CPU is built for agents." — Jensen Huang, regarding the Vera CPU architecture.
• "The world is no longer limited by the number of people. Therefore, those agents are going to use more tools than ever." — Huang, arguing against the idea that AI will eliminate software jobs.

Synthesis and Conclusion

The keynote marks a fundamental transition in computing: the move from human-centric applications to agent-centric systems. NVIDIA has evolved from a GPU company into a full-stack AI infrastructure company. By providing the hardware (Vera Rubin, Vera CPU), the software (Agentic Toolkit, OpenShell), and the models (Neotron 3, Cosmos 3), NVIDIA is positioning itself as the backbone of the next decade of industrial and personal computing. The core takeaway is that the "AI Factory" is the new economic engine, and the ability to generate profitable tokens at scale is the primary competitive advantage for the modern enterprise.