Nvidia CEO Jensen Huang has defended the company against market concerns over AI advancements, stating that investors misinterpreted the capabilities of Chinese firm DeepSeek. In a recent pre-taped interview, Huang emphasized the importance of post-training methods in AI development.
In January, DeepSeek released a competitive open-source reasoning model named R1, built with weaker chips and less funding than Western-made AI models. This led to a $600 billion loss in market capitalization and a significant decline in Nvidia’s stock value. However, Huang claims that the dramatic market response was due to investors’ misinterpretation of DeepSeek’s advancements.
According to Huang, post-training methods are crucial for AI development, allowing models to draw conclusions or make predictions after training. He stated that the need for computing power will grow as these methods diversify and become more complex.
Huang acknowledged that pre-training is still important but emphasized that post-training is where true intelligence lies. The CEO called DeepSeek’s innovations “incredibly exciting” and praised the energy surrounding the open-sourcing of R1.
Nvidia has been defending against concerns over model scaling for months, with Huang previously stating that it had simply shifted from training to inference. He reiterated this stance on Thursday, emphasizing the intensity of post-training methods and their potential for further model improvements.
Huang’s comments may serve as a preview for Nvidia’s first earnings call of 2025, scheduled for February 26. The CEO’s statements have been echoed by rival AMD, whose CEO Lisa Su described DeepSeek as driving innovation that’s “good for AI adoption.”
Source: https://www.businessinsider.com/nvidia-ceo-jensen-huang-addresses-deepseek-stock-sell-off-2025-2