Chinese AI startup DeepSeek has made a bold claim about the profitability of its V3 and R1 models, revealing a theoretical cost-profit ratio of up to 545% per day. While the company cautioned that the actual revenue would likely be far lower, this disclosure marks the first time DeepSeek has revealed details about its cost and revenue structure for “inference” tasks.
The revelation has sent ripples through the AI industry, particularly among U.S.-based firms. DeepSeek’s models, which are behind a wave of popular web and app chatbots, have gained significant traction, causing concern in global AI markets. The company’s announcement comes on the heels of a sharp decline in AI stock values worldwide, especially after DeepSeek’s impressive cost management was revealed.
Unlike U.S. rivals like OpenAI, which rely on expensive, cutting-edge chips for training their models, DeepSeek claims it spent under $6 million on chips to train its models. This is notably lower than the billions spent by American firms, raising questions about their reliance on top-tier hardware. DeepSeek’s choice of Nvidia’s H800 chips, which are less powerful than those used by competitors, further intensifies these concerns.
In its GitHub post, DeepSeek outlined the cost and revenue breakdown, assuming the rental cost of an H800 chip is $2 per hour. According to the data, the daily cost of running its models amounts to $87,072, while potential daily revenue could reach $562,027, leading to the impressive theoretical profit margin. Over a year, this would result in more than $200 million in revenue.
However, DeepSeek also noted that actual revenue would be significantly lower. Factors such as lower costs for using its V3 model, off-peak pricing, and the fact that not all services are monetized—such as free web and app access—contribute to the disparity between theoretical and actual figures.
This revelation has the potential to reshape the global AI landscape, especially as investors and competitors in the U.S. begin to reassess their own models in light of DeepSeek’s cost-effective approach. To read more, visit the original source here.