On February 5th, CITIC Securities Research Report stated that the training and inference computing power requirements for DeepSeek R1 are relatively low, mainly due to the optimization and coordination of algorithms, frameworks, and hardware implemented in DeepSeek R1. The scaling law on the pre training side in the past is gradually moving towards a broader space. In the stage of deep inference, the future computing power demand of the model will still show explosive growth. Adequate computing power demand is still crucial for the performance improvement of artificial intelligence models.
Scan code to share