AI, DEEPSEK, from AI, from AI, revealed the costs and revenues of the V3 and R1 models on Saturday, and revealed the daily theoretical cost rate by up to 545 %.
However, Hangzhou warned that actual revenues are still much lower due to pricing structures and limited income.
This is the first time that Deepseek has shared details about its profit margins from the tasks of inference, and it is the stage that trained AI models create outputs, such as Chatbot responses. This disclosure comes amid the fluctuations of the global artificial intelligence market, with the rapid rise in Deepseek increased already of artificial intelligence shares outside China.
In January, artificial intelligence shares decreased sharply after they acquired Chatbots R1 and V3 jars around the world.
Dibsic previously claimed that she spent less than $ 6 million on training chips, much less than billions invested by American competitors such as Openai. The company also used the NVIDIA H800 chips, which are less advanced than those available to American companies, providing investors’ doubts about high capital expenditures by American artificial intelligence companies.
In the GitHub post, Deepseek estimated that if the H800 SIM rental cost is $ 2 per hour, the total daily inference cost of V3 and R1 is 87,072 dollars. Daily theoretical revenue from these models is $ 562.027, which translates into annual revenue capabilities of more than $ 200 million.
However, the company made it clear that the actual profits are lower due to the low prices for V3, free access to the web and application, and the use of the reduced developer during peak hours.
