DeepSeek Claims Theoretical Cost-Profit Ratio Of 545% Per Day

Chinese AI startup DeepSeek disclosed some cost and revenue data on Saturday related to its popular V3 and R1 models, claiming a theoretical cost-profit ratio of up to 545% per day.

The chips DeepSeek says it used, Nvidia's H800, are also much less powerful than those available to OpenAI and other U.S. AI firms. I Photo: Trong Khiem Nguyen Flickr

However, the company cautioned that actual revenue would be significantly lower, Eduardo Baptista reported for Reuters.

This marks the first time the Hangzhou-based company has revealed any information about its profit margins from less computationally intensive "inference" tasks—the stage after training when AI models make predictions or perform tasks, such as through chatbots.

The revelation could further unsettle AI stocks outside China, which plunged in January after web and app chatbots powered by DeepSeek’s R1 and V3 models surged in popularity worldwide.

The sell-off was partly driven by DeepSeek’s claims that it spent less than $6 million on chips to train its model—far less than what U.S. rivals like OpenAI have spent.

The chips DeepSeek says it used, Nvidia's H800, are also much less powerful than those available to OpenAI and other U.S. AI firms, raising further questions among investors about the necessity of spending billions on cutting-edge chips.

In a GitHub post, DeepSeek stated that assuming the cost of renting one H800 chip is $2 per hour, the total daily inference cost for its V3 and R1 models is $87,072.

In contrast, the theoretical daily revenue generated by these models is $562,027, leading to a cost-profit ratio of 545%. Over the course of a year, this would amount to just over $200 million in revenue.

However, the company noted that its "actual revenue is substantially lower" due to several factors: the cost of using its V3 model is lower than that of the R1 model, only some services are monetized while web and app access remain free, and developers pay lower rates during off-peak hours.