China’s DeepSeek Claims Theoretical Cost-Profit Ratio of 545% Per Day
On Saturday, Chinese artificial intelligence startup DeepSeek disclosed remarkable data concerning its cost and revenue figures tied to its successful V3 and R1 models. The company claims these models boast a theoretical cost-profit ratio of up to 545% per day. However, DeepSeek has suggested that actual revenue is likely to be significantly lower than this optimistic projection.
This announcement is a milestone, representing the first instance where the Hangzhou-based firm has shared any insight into its profit margins from the less computationally intensive “inference” tasks. Inference occurs in the crucial stage post-training where trained AI models execute predictions or perform tasks, such as operating through chatbots.
The disclosure holds the potential to impact AI stocks outside China, which experienced a downturn in January. This downturn was triggered after web and app chatbots, powered by DeepSeek’s R1 and V3 models, saw a surge in global popularity. Concerns arose partly from DeepSeek’s assertion that it invested less than $6 million in the chips used for training the models, a stark contrast to the higher expenditures of US-based AI competitors.
The chips utilized by DeepSeek, which are Nvidia’s H800, are considerably less powerful compared to those available to major US AI companies. This has fueled investor skepticism regarding US firms’ commitments to spend billions on the latest high-performance chips. Such revelations spotlight the disparity in resource allocation and efficiency between Chinese and US AI startups.
DeepSeek provided further details in a GitHub post released on Saturday. They noted that with a rental assumption cost of $2 per hour for a single H800 chip, the aggregate daily inference expense for the V3 and R1 models amounted to $87,072. By stark comparison, the models’ potential daily revenue reached $562,027, culminating in a theoretical cost-profit ratio of 545%. If these figures were sustained over a year, it would indicate a potential revenue exceeding $200 million.
This revelation underscores the innovative cost-efficiency that DeepSeek has managed to achieve, raising questions about the financial models and resource strategies of other AI entities, particularly those in the United States. As AI continues to scale new heights, the industry’s focus on profitability and sustainability has never been more crucial.
DeepSeek’s data disclosure sets a precedent and provides valuable insights into the financial dynamics at play within the burgeoning field of artificial intelligence, fostering further discourse on cost-saving and revenue-maximizing strategies in the AI sector.
