Chinese AI Startup DeepSeek Claims 545% Profit Margin on AI Models
DeepSeek, a Chinese artificial intelligence startup, has announced a potential 545% cost-profit margin for its AI models under specific conditions. The company revealed these figures in a recent GitHub post, highlighting the monetization of only a “subset” of its services.
The projected margin is based on the assumption that all users of DeepSeek’s largely free AI models would transition to a paid plan. According to the company’s calculations, its AI models, V3 and R1, have a daily running cost of $87,072, primarily due to Nvidia chip leasing expenses.
If billed at DeepSeek-R1’s pricing, the company estimates daily revenue could reach $562,027, potentially translating to annual revenue exceeding $200 million. However, industry experts caution that this scenario has not yet been achieved by competitors in the market.
DeepSeek’s operational strategy involves using Nvidia H800 chips for AI model inference during peak hours and reallocating them for research and training at night. The company also implements cost-saving mechanisms such as cross-node batch scaling, computation-communication overlap, and load balancing.
The startup’s financial disclosures have sparked interest in the tech industry, particularly following the launch of its cost-effective AI model. This announcement led to a tech stock sell-off in Silicon Valley due to claims of low-cost model training.
DeepSeek’s success with less powerful Nvidia H800 chips has raised questions about the necessity of extensive AI infrastructure investments. However, the company’s cost claims have been met with skepticism from tech giants and regulatory bodies.
Google DeepMind CEO Demis Hassabis criticized the claims as exaggerated, suggesting they only cover the final training phase. Industry experts argue that the broader investment in development and infrastructure may not be fully accounted for in DeepSeek’s cost projections.
As the AI industry continues to evolve rapidly, DeepSeek’s bold claims and operational strategies are likely to face further scrutiny and analysis from competitors and market observers alike.