NVIDIA’s Jensen Huang on Reasoning Models, Robotics, and Refuting the “AI Bubble” Narrative

In this talk

1. 0:28

   AI adoption and the changing shape of work

   Huang says grounding, reasoning, and links to search have helped AI give more reliable answers. He argues that AI changes the tasks people do without necessarily eliminating the purpose of their jobs: radiologists, for example, can review more scans and help more patients. He also points to new jobs in chip, computer, and AI factory construction, as well as future robot maintenance.

2. 15:44

   Why open source and a broad AI stack matter

   Huang describes AI as a five-layer stack: energy, chips, infrastructure, models, and applications. He argues that open models let startups, universities, and established industries adapt AI to specialized needs, while closed models can still be valid businesses. He cautions against treating one company or one model as the sole path to national AI leadership.

3. 29:25

   Falling costs, specialization, and new applications

   Huang says hardware, algorithms, and model design together are driving large declines in AI costs, and that open research such as DeepSeek’s benefits labs and startups. He expects models and companies to specialize in areas such as coding rather than trying to excel at everything. Looking ahead, he highlights digital biology, including protein understanding and generation, and reasoning systems for cars and robots.

4. 46:00

   Robotics, energy, and the AI bubble question

   Huang expects reasoning and multimodal systems to help robots handle situations beyond their training, and says many forms of moving machinery could become robotic. He predicts more industry-specific AI solutions because physical systems need extremely high reliability. On the bubble question, he points to demand across accelerated computing, healthcare, finance, autonomous vehicles, and research, while arguing that infrastructure capacity and energy supply remain constraints.