Skip to content
Mainstage.
NVIDIA GTC 2025

GTC 2025 keynoteJensen Huang, Mar 18, 2025

Tuesday, March 18, 20252h 12mChecked 2026-09-24
Watch on YouTubeOfficial recap

Jensen Huang unveiled Blackwell Ultra, Vera Rubin, Dynamo, and new open models and robotics platforms at NVIDIA GTC 2025 in San Jose.

Where each takeaway is said2:11:48
2348
  1. 1NVIDIA Blackwell is in full production, delivering 40x the performance of Hopper for AI training and inference.
  2. 2Blackwell Ultra, the next evolution of the Blackwell AI factory platform, arrives in the second half of 2025.1:25:28
  3. 3NVIDIA Vera Rubin architecture systems are set to arrive in the second half of 2026 for major performance gains.1:44:04
  4. 4NVIDIA Dynamo, open source software, is described as the operating system of an AI factory for scaling reasoning models.1:08:15
  5. 5DGX Spark and DGX Station, personal AI supercomputers built on the Grace Blackwell platform, were announced.
  6. 6The Llama Nemotron family of open reasoning AI models was announced for building agentic AI platforms.
  7. 7Isaac GR00T N1 was announced as an open, customizable foundation model for humanoid robot reasoning.
  8. 8Cosmos World Foundation Models were released, an open customizable reasoning model for physical AI.43:04

Announced

1 available the same day
  • NVIDIA BlackwellIn full production, 40x Hopper performance for training and inferenceGA
  • Blackwell UltraNext Blackwell platform evolution, arriving second half of 2025Coming
  • Vera RubinNext architecture, systems arriving second half of 2026Coming
  • NVIDIA DynamoOpen source AI factory operating system for scaling reasoning modelsAnnounced
  • DGX Spark and DGX StationPersonal AI supercomputers on Grace Blackwell platformAnnounced
  • Llama NemotronOpen reasoning AI model family for agentic AIAnnounced
  • Isaac GR00T N1Open foundation model for generalized humanoid reasoningAnnounced
  • Cosmos World Foundation ModelsOpen customizable reasoning models for physical AIAnnounced
  • NewtonOpen source physics engine for robotics, with Google DeepMind and Disney ResearchAnnounced
  • Spectrum-X and Quantum-XSilicon photonics networking switches, fewer lasers, more power efficientAnnounced

On stage

Sources

Companies named

Related keynotes

GTC 2026Mar 16, 2026Jensen Huang
  • The keynote opened by framing the token as the basic unit of modern AI, at NVIDIA GTC 2026, SAP Center, San Jose.
  • Huang introduced DLSS 5, using 3D guided neural rendering for real time, photoreal 4K performance on local hardware.
  • Huang said computing demand has increased by 1 million times in the last two years.
6 announced2h 19mCES 2026Jan 5, 2026Jensen Huang
  • Huang said roughly 10 trillion dollars of the last decade of computing is being modernized to accelerated, AI-based computing.
  • NVIDIA Rubin, its first extreme-codesigned six-chip AI platform, is now in full production, cutting token cost to about one tenth.
  • NVIDIA showed a personal AI agent running locally on a DGX Spark desktop supercomputer, embodied in a Reachy Mini robot.
6 announced1h 32mComputex 2025May 19, 2025Jensen Huang
  • Huang told a crowd of more than 4,000 at Taipei Music Center that AI data centers are, in fact, AI factories.
  • NVIDIA detailed huge new AI installations from CoreWeave, Oracle, Microsoft and xAI, calling them gigantic factory investments.
  • NVIDIA and Foxconn Hon Hai are deepening their partnership to build an AI factory supercomputer for Taiwan, including TSMC.
9 announced1h 43m

Summaries of official keynote videos and recaps; watch the source for exact wording. Logos via logo.dev; trademarks belong to their owners.

After each keynote: the takeaways and what launched, in one short email.