DeepSeek today revealed some cost and revenue data related to the popular V3 and R1 models, claiming a theoretical cost-to-return ratio of up to 545% per day, despite warnings that actual revenue could be lower.
This is the first time the Chinese AI startup has disclosed any information about its profit margins from less computationally demanding “inference” tasks, the post-training stage that involves trained AI models making predictions or performing tasks, such as via chatbots.
The theoretical return on investment that DeepSeek claims is a whopping 545%.
The disclosure from the Hangzhou-based company could further shake up the AI stock market outside China, which plunged in January after web and app chatbots powered by the company's R1 and V3 models soared in popularity worldwide.
The sell-off in AI stocks was partly due to claims that DeepSeek spent less than $6 million on chips used to train its models, far less than what U.S. rivals like OpenAI spend.
The chips DeepSeek claims to have used, Nvidia's H800, are also far less powerful than what OpenAI and other US AI companies have access to, making investors even more skeptical about US AI companies' pledges to spend billions on advanced chips.
Assuming the cost of renting an H800 chip is $2 per hour, the total daily inference cost for its V3 and R1 models is $87,072, DeepSeek said in a GitHub post published Saturday. By contrast, the theoretical daily revenue generated by these models is $562,027, resulting in a 545% return on investment. Over a year, that would add up to more than $200 million in revenue.
However, the company added that its "actual revenue is significantly lower" because the cost of using its V3 model is lower than the R1 model, only some services are monetized because web and app access remains free, and developers pay less during off-peak hours.
Chinese officials have ordered the country’s top artificial intelligence (AI) researchers and entrepreneurs to avoid business trips to the United States, the Wall Street Journal reported. The newspaper said Chinese authorities are concerned that AI experts traveling abroad could reveal confidential information about the country’s technological development.
Chinese authorities are also concerned that executives traveling on business could be detained and used as bargaining chips in negotiations between the United States and China, recalling the arrest of a Huawei executive in Canada at Washington's request during US President Donald Trump's first term.
According to the source, executives of Chinese AI companies who go on business trips must report their plans to the authorities before leaving and upon returning, give a detailed account of what they did and who they met.
Source: https://www.baogiaothong.vn/deepseek-tiet-lo-loi-nhuan-khung-tu-cac-mo-hinh-ai-192250301220905413.htm
Comment (0)