Many of us now use AI chatbots on a daily basis. ChatGPT from Open AI is still the most widely used tool, but its Chinese counterpart DeepSeek has long since taken the AI world by storm. This is particularly exciting in terms of maintenance costs, as DeepSeek is significantly cheaper than ChatGPT. Reuters, among others, found this out. How can that be?
Theoretical cost structure and profitability
According to DeepSeek, the daily operating costs for the inference of their V3 and R1 models amount to approximately 87,072 US dollars. Assuming full monetization, this could result in daily revenues of USD 562,027, which corresponds to a theoretical cost-benefit ratio of 545%. Extrapolated, this would mean annual revenues of over 200 million US dollars.
However, the company states that the actual revenue is significantly lower. The reasons for this include the lower usage costs of the V3 model, the partially free provision of services via the web and app and reduced fees for developers during off-peak times.
ChatGPT, on the other hand, is significantly more expensive. According to Futurism.com magazine, it costs OpenAI, the company behind the AI chatbot, around 700,000 US dollars a day. Even despite a premium subscription, which costs around 200 US dollars a month, OpenAI is said to continue making losses with ChatGPT, according to the website TechCrunch.
No wonder investors are wondering why they should continue to put their money into ChatGPT when the competition in the form of DeepSeek only costs a fraction of that.
Development and training costs
The training costs of the V3 model were stated by DeepSeek to be less than 6 million US dollars. However, this sum relates exclusively to the final training and does not take into account the previous research and development costs, experiments or personnel costs. Therefore, the actual total costs for the development of the model could be significantly higher.
Impact on the global AI industry
DeepSeek’s disclosures have caused a global stir and led to a drop in AI share prices outside China. Investors are questioning the high spending of Western AI companies, especially when Chinese models achieve comparable performance at a lower cost.
Nevertheless, experts emphasize that DeepSeek’s actual development costs may be higher than stated. In addition, factors such as data quality, infrastructure and long-term sustainability could influence cost efficiency.
Pricing and market positioning
DeepSeek offers its services at competitive prices. Use of the web browser version and the app is free and comparable in quality to ChatGPT 4.0, which costs €20 per month as a subscription. Access to the APIs is also significantly cheaper compared to other providers. The V3 model costs 1.4 US cents per million tokens input and 28 US cents per million tokens output. The more advanced R1 model costs 14 US cents per million tokens input and 2.19 US dollars output.
Conclusion
DeepSeek is positioning itself as a cost-effective alternative in the AI market. While the stated low development and operating costs appear impressive, caution is advised as not all cost factors have been transparently disclosed. The actual profitability and long-term impact on the global AI industry remains to be seen.
