Chinese company DeepSeek, the developer of the popular V3 and R1 IPS models, has shared some financial data regarding their operation. According to Reuters, the company could theoretically make huge profits, but the actual figures are much lower.
► Subscribe to the telegram channel “Minfina”: the main financial news
Hypothetical income exceeds expenses several times
DeepSeek estimates that, based on the $2 per hour cost of renting one NVIDIA H800 chip, the daily operating costs of the V3 and R1 models are $87,072. At the same time, the potential daily revenue generated by these models is estimated at $562,027. This means that the theoretical revenue exceeds the costs by 545%, and the annual revenue could exceed $200 million.
Real incomes are significantly lower
However, DeepSeek is quick to point out that the company's actual revenues are “significantly lower” than potential. This is because the V3 model costs less to use than the R1, and because some features, such as internet and app access, are provided to users for free.
Impact on the AI market
The release of these figures could have a significant impact on AI stocks outside of China. It’s worth recalling that DeepSeek’s R1 model made waves in January, as it demonstrated performance comparable to OpenAI’s latest models, but at a price 90-95% lower.
🕵️ We created a short survey to learn more about our readers.
💛💙 Your answers will help us become better, pay more attention to the topics that interest you.
🤗 We would be grateful if you could take a minute to answer our questions.
- Technologies