Chinese AI startup Deepseek recently declared its AI model to be highly profitable.
In X's post, Deepseek boasts that the “cost profit margin” for its online service is 545%. However, the margin is calculated based on “theoretical income.”
These numbers were discussed in more detail at the end of a longer GitHub post that outlines an approach to achieving “higher throughput and low latency.” The company writes that when considering using the V3 and R1 models during the 24-hour period, if all of that use was billed using the R1 price, DeepSeek already has $562,027 in daily revenue.
Meanwhile, the cost of leasing the required GPU (graphic processing unit) was only $87,072.
The company acknowledged that actual revenue was “significantly low” for a variety of reasons, including the nightly discount, the low price of the V3 and the fact that “only a subset of services are monetized.”
Of course, if the app or website is not free and other discounts are not available, then use is probably much lower. Therefore, these calculations appear to be very speculative. Currently it's more of a gesture to potential profit margins in the future than a real snapshot of Deepseek's bottom line profits.
However, the company shares these figures in a broader discussion of AI costs and potential profitability. Deepseek jumped into the spotlight in January, facing US trade restrictions that prevent Chinese companies from accessing the most powerful chips, with a new model that appears to match Openai's O1 in certain benchmarks, despite being developed at a much lower cost. Tech stocks fell, and analysts raised questions about AI spending.
Deepseek's technology didn't just rattle Wall Street. The app temporarily expelled Openai's ChatGpt, which is at the top of Apple's App Store, but has since dropped from the general rankings, and is now ranked 6th in productivity after ChatGpt, Grok and Google Gemini.