Deepseek claims that the “theoretical” margins of a profit of 545%

The Chinese startup AI Deepseek has recently stated that its AI models can be very profitable – with some asterisks.

In PostDeepseek boasted that his online services have a “profit margin” of 545%. However, this margin is calculated on the basis of “theoretical income”.

He discussed these numbers more detailed at the end of Long -term Drawing his approach to achieve “higher bandwidth and lower latency”. The company wrote that when viewed the use of its V3 and R1 models in a 24-hour period, if This use has been charged with R1 prices, Deepseek will now have $ 562,027 daily revenue.

Meanwhile, the cost of leasing the necessary graphic processors (graphics units for processing) would be only $ 87,072.

The company acknowledged that its actual revenue was “significantly lower” for various reasons, such as discounts at night, lower prices for V3 and the fact that “only a subset of services are monetized”, with access to the network and applications remained free of charge.

Of course, if the app and the website were not free and if other discounts were not available, the use would probably be much smaller. So these calculations seem highly speculative – a sooner gesture to potential future profit margins than a real photo of the bottom line of Deepseek right now.

But the company shares these numbers against the backdrop of a wider debate about the price of AI and the potential profitability. Deepseek jumped into the spotlight In January, with a new model, which is supposedly corresponding to O1’s O1 of certain indicators, although it was developed at a much lower price and in front of trade restrictions in the United States, which prevents Chinese companies from accessing the most powerful chips. The technical stocks have broken up and Analysts raised questions about AI costsS

Deepseek technology not just rattled Wall Street. Its application Briefly displaced Openai chat At the top of Apple’s App Store – although it has subsequently fell from the common charts and is currently ranked No. 6 in performance after Chatgpt, Grok and Google Gemini.