Deepseek claims “theoretical” profit margins of 545%

The Chinese Ki -Startup Deepseek recently explained that its AI models could be very profitable -with a few asterisks.

In A contribution to XDeepseek boasted that his online services have a “cost -profit margin” of 545%. However, this margin is calculated on the basis of the “theoretical income”.

In the end, these figures were discussed by more detail A longer GitHub contribution Output of the approach to achieve the “higher throughput and the lower latency”. The company wrote that it deals with the use of the V3 and R1 models during a period of 24 hours. If This use was all invoiced with R1 prices, Deepseek already had 562,027 US dollars a day.

In the meantime, the cost of leasing the required GPUs (graphics processing units) would only be $ 87,072.

The company admitted that the actual sales are “essentially lower” for various reasons, e.g. B. Night discounts, lower pricing for V3 and the fact that “only a subset of services is monetized”, whereby the web and app access the remaining free of charge.

If the app and the website were not free of charge and if other discounts were not available, the use would probably be much lower. So these calculations seem to be very speculative – more of a gesture for potential future profit margins than a real snapshot from Deepseek’s business result.

However, the company shares these figures in the middle of wider debates about the costs and the potential profitability of AI. Deepseek jumped into the spotlight In January, with a new model, the OpenAis O1 allegedly agreed on certain benchmarks, although they were developed too much lower costs, and in view of the US trade restrictions that prevent Chinese companies from accessing the most powerful chips. Tech shares fell and Analysts asked questions about AI editions.

Deepseek’s technology not only torn the Wall Street. Its app Briefly expelled openais chatt At the top of the Apple App Store – although it then fell from the general ranking and is currently in 6th place in productivity behind Chatgpt, Grok and Google Gemini.