DeepSeek claims 545% daily cost-profit ratio for its AI models
What's the story
Chinese artificial intelligence (AI) start-up, DeepSeek, has revealed financial details for its popular V3 and R1 models. The company claims a theoretical daily cost-profit ratio of as much as 545%. However, it has also cautioned that real earnings could be much lower. This is the first time DeepSeek has revealed details about its profit margins from less computationally intensive "inference" tasks.
Market influence
DeepSeek's AI models impact global market
DeepSeek's disclosure could rattle international AI stocks, which took a hit in January. This was because of a spike in the popularity of web and app chatbots powered by DeepSeek's R1 and V3 models. The company has previously claimed it spent less than $6 million on chips to train the model, a number much lower than US rivals like OpenAI.
Investment strategy
Cost-effective approach raises questions
DeepSeek trained on NVIDIA's H800 chips, which are less powerful than those accessible to US AI firms. This has raised doubts about US AI companies' promises to invest billions in advanced chips. DeepSeek's cost-effective approach has ignited a debate among US AI investors about the need for such high-end investments.
Financial breakdown
DeepSeek reveals daily inference cost and revenue
In a recent GitHub post, DeepSeek disclosed that the total daily inference cost for its V3 and R1 models is $87,072, assuming an hourly rental cost of $2 for one H800 chip. The company also revealed that these models could theoretically generate daily revenues of $562,027. This results in a cost-profit ratio of 545% and an annual revenue potential of over $200 million.