World Economic Forum Records Conversation With NVIDIA CEO Jensen Huang in Davos

jp.weforum.org reported that a one-on-one session between NVIDIA CEO Jensen Huang and Larry Fink, interim co-chair of the World Economic Forum and president and CEO of BlackRock, was recorded at the 2026 Annual Meeting in Davos.

The World Economic Forum recorded a one-on-one session between NVIDIA's CEO and the forum's interim co-chair at its Annual Meeting in Davos. The available information does not confirm specific technical claims made in the conversation, and it is not possible to determine a direct connection to related technology trends.

What can be confirmed is that the World Economic Forum recorded a one-on-one session with Huang and Fink at its 2026 Annual Meeting in Davos. The session's argument and the content of the remarks cannot be assessed based on the available information. See the original source for details. The relevant question raised by related coverage from this newspaper is not what was said in the conversation, but how far the technologies underpinning robot AI have advanced. One area is computing resources. NVIDIA Explains Full-Stack Design to Address AI Memory Constraints reported that the company discussed pressure on GPU memory from large models and long contexts, and also mentioned HBM3e and HBM4. However, the article summary does not provide design specifications or performance figures. It is therefore not possible to determine from this information how the computing infrastructure described relates to actual robot processing. At another layer, the issues include infrastructure for collecting training data and running models, as well as the ability to manipulate objects in real-world settings. [Hugging Face Releases Robot Training Platform LeRobot v0.7.0, Supporting Conversational Control and Large-Scale Data Training](/en/news/hugging-face-releases-lerobot-v0-7-0-robotics-training-platform-adding-4e8d3f) covered about 300TB of open Physical AI data on the Hub and training capabilities for large models. Meanwhile, Panasonic HD and Others Jointly Develop T-Rex Robot Manipulation AI, With a 65% Average Success Rate Across 12 Tasks reported an average success rate of 65% across 12 real-world tasks involving touch. These are separate efforts, and no link to results based on NVIDIA's computing infrastructure has been confirmed. Evaluating data infrastructure, computing resources and robot manipulation as a single process would require details on how they are connected and performance in real-world operations. Three points warrant further attention. First, the full transcript or video could show which issues were discussed in detail. Second, how computing resources and memory specifications apply to robot training and inference. Third, whether the scale of training data and manipulation performance in real-world settings are reported under the same evaluation conditions.