Lionel Bonaventure | AFP | Getty Images
Chinese artificial intelligence startup company DeepSeek On Saturday, some cost and revenue data related to the HIT V3 and R1 models allegedly had a theoretical cost-based ratio of up to 545% per day, although it warned that actual revenue would be significantly lower.
This marks the first time the Hangzhou-based company has revealed any information about its lower-margin “computer-intensive reasoning” task, a post-training phase involving training AI models for prediction or execution of tasks, such as through chatbots.
Revelation may be further Slump in January The popularity of the world emerged after Web and App Chatbots powered by their R1 and V3 models.
The sell-off is partly due to DeepSeek’s claim that it spent less than $6 million on the chips used to train the model, much less than the chips spent by competitors like Openai.
Chips DeepSeek The claim that it uses NVIDIA’s H800, which has far fewer features than OpenAI and other U.S. AI companies can use, has led investors to even raise more questions about U.S. AI companies’ commitments to spend billions of dollars on cutting-edge chips.
DeepSeek said in a GitHub post posted on Saturday that assuming the cost of renting a one-h800 chip is $2 per hour, its V3 and R1 models have a daily inference cost of $87,072. In comparison, these models generate theoretical daily income of $562,027, resulting in a cost rate of 545%. In the year, this revenue totaled more than $200 million.
However, the company added that its “real revenue is significantly lower” because using its V3 model is less expensive than the R1 model, and only some services are free Pay less during off-peak hours.
