According to BlockBeats, on May 21, NVIDIA CEO Jensen Huang stated that the open-source release of DeepSeek R1 has propelled the widespread application of inference AI technology, accelerating the demand for computation.
Models like OpenAI's o3, DeepSeek R1, and xAI's Grok 3 demonstrate a shift in AI from perception and generative models to long-term reasoning models that require enhanced computational capabilities. These models tackle complex problems, with computational demands 100 times that of traditional AI.
NVIDIA CEO Jensen Huang pointed out that the scale of post-training and model customization is enormous, with computational demands potentially exceeding pre-training levels by several orders of magnitude. Driven by extended testing times and new inference models, the demand for inference is accelerating.