China-based artificial intelligence startup DeepSeek shared cost and revenue data for its V3 and R1 models. The company said it could theoretically achieve a cost-to-profit ratio of up to 545% per day, but emphasized that actual revenues were below this level.
DeepSeek V3 and R1 Models
DeepSeek's V3 and R1 models are the company's most advanced language and reasoning models in AI. V3 excels in natural language processing and chat applications, while the R1 model is used for tasks requiring more complex reasoning and logic.
Profit Margin and Cost Analysis
By using Nvidia's H800 chips, DeepSeek achieves lower costs for model training than its US competitors. The company calculates that at a rental cost of $2 per hour for the H800 chip, the daily inference cost is $87,072. Theoretically, the daily revenue from these models is $562,027, implying a profit margin of 545%.
Real Revenue and User Engagement
However, DeepSeek states that its actual revenues are lower due to several factors. These include the lower cost of using the V3 model, some services being offered for free, and developers paying less during off-peak hours.
Market Impacts
DeepSeek's announcements have caused volatility in global artificial intelligence stocks. The company's low-cost and high-performance models are causing US rivals to rethink their strategies.
Chip Usage and Competition
Nvidia CEO Jensen Huang has stated that DeepSeek's R1 model are reasoning models that require more computational power, leading to an increase in resource-intensive models.
Conclusion
DeepSeek's cost and revenue data illustrate the company's impact and strategies in the AI market. Its low-cost and efficient models are driving significant changes in the global AI industry.
0 Comments: