DeepSeek claims theoretical profit margins of 545%.

DeepSeek, a Chinese AI startup, recently announced that its AI models were very profitable — but with some asterisks.

DeepSeek boasted in a postthat its online services had a “cost-profit margin” of 545%. This margin is calculated on the basis of “theoretical earnings.”

The article discussed these numbers more in detail at its end. DeepSeek has a longer GitHub posting which outlines its approach to achieving a “higher throughput” and “lower latency.” If all usage was billed with R1 pricing during a 24 hour period, DeepSeek already had $562,027 of daily revenue.

The cost of leasing the GPUs (graphics processor units) required would have been $87,072.

According to the company, its actual revenue is “substantially” lower due to a number of factors, including nighttime discounts, lower pricing of V3, and that “only a small subset of services is monetized,” while web and app access remains free.

Ofcourse, if the website and app weren’t free and if there weren’t other discounts available, usage would be much lower. These calculations are highly speculative, and more of a hint at potential future profits than a snapshot of DeepSeek’s current bottom line.

The company is sharing the numbers amid broader debates on AI’s potential profitability and cost. DeepSeek shot to prominence in January with a model that supposedly matched OpenAI’s o1 in certain benchmarks. This was despite the fact that the model was developed at a lower cost and in spite of U.S. Trade restrictions that prevent Chinese firms from accessing the fastest chips. Analysts raised concerns about AI spending and tech stocks fell.

DeepSeek’s tech didn’t just shake Wall Street. Its app briefly replaced OpenAI’s ChatGPT as the top Apple App Store app — but it has since fallen off the general rankings, and is now ranked #6 for productivity, behind ChatGPT Grok and Google Gemini.

www.aiobserver.co