Nvidia CEO Huang: DeepSeek's R1 reasoning AI consumes 100x more compute than non-reasoning AI

CNBC TelevisionAbout 2 min readMar 20, 2025Watch original
THE SUMMARYAI-generated

Key Concepts:

  • R1 DeepSeek: An open-sourced reasoning model developed by Nvidia.
  • Reasoning AI: Artificial intelligence that can reason, break down problems, and verify solutions.
  • Compute Consumption: The amount of computational resources required to run an AI model.
  • Quiet Period: A restricted communication period for publicly traded companies before earnings announcements.

R1 DeepSeek and its Capabilities

The core topic revolves around Nvidia's R1 DeepSeek, described as the "first open-sourced reasoning model." The key characteristic of R1 is its ability to "reason," meaning it doesn't just provide answers but breaks down problems step-by-step. This involves:

  • Self-Questioning: The AI asks itself questions during the problem-solving process.
  • Option Generation: It considers multiple potential solutions.
  • Verification: It verifies the correctness of its answers, exemplified by solving a quadratic equation and plugging the solution back in to confirm its validity.

Compute Consumption of Reasoning AI

A central point of discussion is the compute consumption of reasoning AI like R1. The speaker emphasizes that reasoning AI consumes "100 times more compute than a non-reasoning AI."

The "Deep Seek Issue" and Communication Strategy

The conversation addresses a past event where Nvidia's stock experienced a "600 billion dollar hit" following a "Deep Seek issue." The interviewer suggests that Nvidia could have mitigated the impact by explaining the positive aspects of R1. The response indicates that Nvidia was in a "quiet period" at the time, restricting their communication. However, the speaker acknowledges that the announcement of R1 still occurred shortly before the end of the quiet period. The speaker suggests direct communication in the future to address such issues promptly.

Significance of R1

The speaker highlights that R1 is a "fantastic thing" because it is the first open-sourced reasoning model. This openness is presented as a significant advancement in the field of AI.

AI summaries can miss context or contain errors. Check important details against the original video.

Go a little deeper.

Have a question about this video? Load its transcript to open the video chat.