DeepSeek Claims Its AI Model Hits 545% Profit Per Day

- DeepSeek announces 545% profitability of its AI model - Using NVIDIA H800 GPUs, daily operational costs reach $87,072, with potential revenue of $562,027 [Unblock Media] Chinese AI startup DeepSeek has recently announced that the theoretical cost-to-revenue ratio for its AI model inference systems, V3 and R1, reaches 545%, drawing significant attention from the industry. This announcement stands out as a rare case of transparently disclosing AI model profitability, sparking discussions across the AI sector. DeepSeek calculated that by operating its V3 and R1 models for 24 hours, leasing NVIDIA H800 GPUs at $2 per hour resulted in daily operational costs of $87,072. Simultaneously, under the R1 model’s pricing scheme, the potential revenue from processed tokens was estimated at $562,027, leading to a theoretical cost-to-revenue ratio of 545%. However, these figures are based on theoretical calculations, meaning actual profitability may be lower due to factors such as free service offerings, discounted rates, and varying usage patterns. DeepSeek’s pricing strategy is notably aggressive compared to competitors. The premium R1 model is priced at $0.14 per million input tokens for cache hits, $0.55 for cache misses, and $2.19 per million output tokens. In contrast, OpenAI’s GPT-4o model costs $2.50 per input token and $10.00 per output token, while the GPT-4o mini model is priced at $0.15 per input token and $0.60 per output token. This comparison highlights the aggressive nature of DeepSeek’s pricing strategy. The use of the term "545% profitability" in DeepSeek’s announcement has sparked controversy within the tech community. Critics argue that the term "profitability" may be misleading, as it implies realized profit rather than a theoretical calculation. A more precise term would be "gross margin" or "contribution margin," which better reflects the difference between revenue and direct costs without accounting for other operational expenses. This debate underscores the need for standardized reporting metrics in the rapidly evolving AI industry. DeepSeek’s aggressive pricing strategy offers businesses opportunities to reduce costs. For instance, if a small or medium-sized enterprise (SME) were to adopt AI services, using OpenAI’s GPT-4o model would cost $2.50 per input token, whereas DeepSeek’s R1 model would provide the same service at just $0.14 per input token in a cache hit scenario. This significant price difference lowers the barriers to AI adoption, encouraging more companies to leverage AI technology. DeepSeek’s announcement has had a broad impact on the AI industry. AI-related stock markets have experienced heightened volatility, with NVIDIA losing approximately 17% of its value and $588.8 billion in market capitalization. Additionally, DeepSeek’s lower development costs (estimated at $5-6 million) challenge existing assumptions about the investment needed for AI infrastructure. The company’s models reportedly consume 10 to 40 times less energy than comparable American AI technologies, influencing energy demand forecasts for AI applications. These developments are prompting the industry to reassess AI strategies, emphasizing cost efficiency, energy consumption, and open collaboration. DeepSeek’s theoretical profitability announcement has ignited important discussions on AI industry profitability, pricing strategies, terminology, and real-world applications. These discussions are expected to play a crucial role in shaping the future direction of the AI sector and influencing corporate strategies.