What Did NVIDIA CEO Jensen Huang Announce in the GTC 2026 Full Keynote?

TL;DR

NVIDIA’s AI factory innovations turn computing infrastructure into factories that generate tokens, the building blocks of AI. Jensen Huang presents AI factories alongside NVIDIA’s CUDA X and systems platforms, backed by a CUDA installed base spanning hundreds of millions of GPUs and computing systems. Read on to see how CUDA’s ecosystem, architectural compatibility, and continuing software updates reduce computing costs over time.

Transcript

This is how intelligence is made. A new kind of factory, generator of tokens, the building blocks of AI. tokens have opened a new frontier, turning data into knowledge and drawing on all we have learned. Tokens are harnessing a new wave of clean energy and unlocking the secrets of the stars. In virtual worlds, they help robots learn and in the phys... Read More

Key Insights

• Nvidia is enhancing AI inference with the Vera Rubin architecture, which offers a 10x increase in throughput for high-value tiers.
• Groq's technology integration with Nvidia aims to boost token generation speed by 35x, enhancing AI factory efficiency.
• Open Claw is a pivotal open-source framework likened to Linux and HTML, enabling the development of personal AI agents.
• Nvidia's AI factories focus on maximizing token throughput per watt, driving down computing costs and increasing efficiency.
• The AI inference inflection point has arrived, with Nvidia's architecture supporting diverse AI models across multiple domains.
• Nvidia's CUDA X libraries are central to its strategy, providing domain-specific acceleration across various industries.
• Nvidia is committed to open models, contributing to one of the largest AI ecosystems with nearly 3 million models.
• Strategic partnerships with companies like Meta and Nebius are expanding Nvidia's reach in AI cloud services.

Questions & Answers

Q: What are NVIDIA’s AI factories?

AI factories are a new NVIDIA platform designed to generate tokens, which the keynote calls the building blocks of AI. They sit alongside NVIDIA’s CUDA X and systems platforms.

Q: What did NVIDIA CEO Jensen Huang cover in the GTC 2026 full keynote?

The keynote covers NVIDIA’s three platforms: CUDA X, systems, and AI factories. It also emphasizes ecosystems and the layers of artificial intelligence, from infrastructure and chips to platforms, models, and applications.

Q: Why does NVIDIA describe tokens as the building blocks of AI?

The keynote says tokens turn data into knowledge and draw on what has been learned. It presents tokens as useful across virtual worlds, robotics, physical work, science, transportation, and healthcare.

Q: What are NVIDIA’s three platforms?

NVIDIA identifies CUDA X, systems, and AI factories as its three platforms. The keynote says it will discuss all three, with particular attention to their ecosystems.

Q: Why is CUDA’s installed base important to NVIDIA?

NVIDIA says its installed base attracts developers, who create algorithms and breakthroughs that lead to new markets and ecosystems. Those ecosystems then enlarge the installed base, accelerating the platform’s flywheel.

Q: How large is NVIDIA’s CUDA installed base?

After 20 years of work on CUDA, NVIDIA says hundreds of millions of GPUs and computing systems around the world run it. CUDA is present in every cloud, every computer company, and nearly every industry served by NVIDIA.

Q: How does NVIDIA reduce computing costs over time?

Accelerated computing initially speeds up applications, while continuing software updates deliver additional cost reductions throughout the hardware’s life. Because NVIDIA GPUs are architecturally compatible, a new optimization can benefit millions of installed systems.

Q: What does CUDA support across the AI ecosystem?

CUDA supports every phase of the AI life cycle, every data-processing platform, and scientific solvers of many kinds. Its ecosystem includes thousands of open-source tools, compilers, frameworks, and libraries, plus a couple of hundred thousand public projects.

Summary & Key Takeaways

• Definition: AI factories are NVIDIA’s new platform for generating tokens, the building blocks of AI.

• Definition: NVIDIA presents CUDA X, systems, and AI factories as its three platforms.

• Number: GTC featured 450 sponsoring companies, a thousand technical sessions, and 2,000 speakers.

• Number: CUDA reached its 20th anniversary after NVIDIA worked on the architecture for 20 years.

• Definition: CUDA uses SIMT, single instruction, multi-threaded, to let scalar code spawn into multi-threaded applications.

• Tool: CUDA added tiles to help programmers work with tensor cores and mathematical structures foundational to artificial intelligence.

• Number: CUDA includes thousands of open-source tools, compilers, frameworks, and libraries, with a couple of hundred thousand public projects.

• Number: Hundreds of millions of GPUs and computing systems around the world run CUDA.

• Compare: CUDA’s installed base attracts developers, breakthroughs, markets, and ecosystems, which then create a larger installed base.

• Definition: Architectural compatibility lets NVIDIA support installed GPUs and extend their useful life through continuing software updates.

• Tool: A new NVIDIA software optimization can benefit millions of installed systems.

• Definition: Accelerated computing and continuing software updates reduce computing costs over time.