China’s DeepSeek Asserts Theoretical Daily Cost-Profit Ratio of 545% – One America News Network
By Reuters
March 1, 2025 – 4:16 AM PST
Advertisement
BEIJING (Reuters) – On Saturday, Chinese AI startup DeepSeek shared its expenses and revenue figures concerning its popular V3 and R1 models, asserting a potential cost-profit ratio of up to 545% per day, while also noting that actual revenue is expected to be much lower.
This is the first instance where the Hangzhou-based firm has disclosed any insights into its profit margins from less intensive “inference” operations, the phase following the training of AI models, where they execute predictions or tasks, such as via chatbots.
The announcement might further unsettle AI stocks outside of China, which faced a significant decrease in January following a worldwide surge in the use of chatbots powered by DeepSeek’s R1 and V3 models.
The drop was partly influenced by DeepSeek’s assertion that it invested less than $6 million on chips for training the model, contrasting sharply with the substantially higher expenditures of U.S. competitors like OpenAI.
The chips DeepSeek claims to have used, Nvidia’s H800, are also considerably less powerful than those accessible to OpenAI and other American AI companies, prompting investors to question the validity of U.S. AI firms’ commitments to invest billions in advanced chips.
In a GitHub update released on Saturday, DeepSeek stated that if the rental cost for one H800 chip is $2 per hour, then the overall daily inference expense for its V3 and R1 models amounts to $87,072. On the other hand, the theoretical daily revenue from these models is estimated at $562,027, yielding a cost-profit ratio of 545%. Over a year, this would equate to just above $200 million in revenue.
Nevertheless, the company clarified that its “actual revenue is significantly less” due to the lower costs associated with its V3 model compared to the R1, limited monetization of certain services since web and application access is free, and reduced payments from developers during non-peak times.
Reporting by Eduardo Baptista; Editing by Daren Butler
Advertisements below
