OpenAI turned Wall Street upside down with its AI models. The alleged low cost of development, training, and maintenance of R1 and V3 caused the stocks of companies like NVIDIA to crash, losing up to $600 billion in value in 24 hours. Now, DeepSeek is boasting that it could earn daily profit margins of up to 545%. However, the figure is not as realistic as you might think.
DeepSeek arguably started a debate about profit margins in the artificial intelligence segment. The fall in NVIDIA stock occurred after DeepSeek’s claims that it invested only $6 million in training the R1 model. The latter boasts capabilities comparable to GPT-4o, which was much more expensive to train. Not surprisingly, the Chinese company is also participating in the debate with its own arguments.
DeepSeek’s theoretical “cost profit margin” is impressive, but highly optimistic
DeepSeek said that, according to its calculations, it has a “cost profit margin” of 545%. They are not talking about current profits but about theoretical revenue margins. The condition for achieving such margins is that current usage levels of its V3 and R1 models are billed at the R1 paid plan prices. This would represent about $562,027 per day for the company, according to the X/Twitter post.
The company calculates that, under this “ideal scenario,” the daily cost of renting the necessary GPUs would be about $87,072. In a later post, DeepSeek admits that current revenues are “substantially lower.” The firm cites various reasons, such as the lower cost of the V3 model’s paid plans and that “only a subset of services are monetized.”
DeepSeek’s estimates represent an overly optimistic ideal scenario. No AI service—or tech service in general—has managed to get all its users to subscribe to paid plans. Nor have they succeeded in persuading all users to choose the most costly plan available. So, the Chinese firm’s calculations seem more like a study of potential revenue than a realistic goal.
It’s noteworthy that the alleged low cost of the DeepSeek AI models has been called into question. Some estimate a real total investment of $1.6 billion to develop and train them.
