Chinese AI company DeepSeek recently highlighted the potential profitability of its models, though its claims come with significant caveats. In a social media post, the startup stated that its online services achieved a theoretical “cost profit margin” of 545%, calculated under idealized conditions. This figure was detailed in a broader technical discussion on GitHub, where the company outlined efforts to optimize performance. According to the analysis, if all usage of its V3 and R1 models over a 24-hour period were billed at R1 pricing, daily revenue could reach $562,027, while the cost of renting the required graphics processing units (GPUs) would total $87,072.
However, DeepSeek acknowledged its actual earnings are far lower due to factors like discounted nighttime rates, cheaper pricing for the V3 model, and free access to web and app services. If these free tiers and discounts were removed, user engagement would likely drop, making the projections more speculative than reflective of current financial realities. The company shared these metrics amid ongoing industry debates about the economic viability of AI technologies. This follows DeepSeek’s January launch of a model reportedly rivaling OpenAI’s GPT-4 on select benchmarks, developed despite U.S. export restrictions limiting access to advanced chips. The announcement coincided with market turbulence, as tech stocks dipped and analysts scrutinized the sustainability of AI-related investments.