Chinese AI startup DeepSeek has released financial data suggesting that its AI models could achieve an impressive 545% theoretical cost-profit margin.

The company shared these figures, explaining that its V3 and R1 models could generate $562,027 in daily revenue if all usage were billed at the higher R1 pricing, while the cost of leasing GPUs would only amount to $87,072. However, DeepSeek quickly clarified that its actual revenue is significantly lower due to discounts, varied pricing structures, and the fact that not all of its services are monetized.

The disclosure is notable as it provides rare insight into the business economics of an emerging AI player, particularly one operating under U.S. chip export restrictions.

DeepSeek’s rise has already shaken the AI industry, with its cost-efficient models drawing attention for their ability to compete with leading U.S. firms like OpenAI despite significantly lower training expenses. The company claims to have spent less than $6 million on chips to develop its models, a stark contrast to the billions invested by American AI firms.

By revealing details of its revenue potential, DeepSeek is not only making a statement about its own business viability but also fueling ongoing discussions about the profitability of AI startups in general.

Why It Matters: DeepSeek’s disclosure intensifies the debate over AI profitability, highlighting how a low-cost approach can rival billion-dollar investments in AI infrastructure. With AI firms like OpenAI and Anthropic still searching for sustainable revenue models, their claims potentially challenge industry assumptions and could influence how investors assess AI spending.

• 545% Theoretical Profit Margin: DeepSeek estimates a massive cost-profit ratio based on projected revenue, but acknowledges real earnings are lower due to discounts, pricing structures, and free services.

• $562K Daily Revenue Projection: If all usage were billed at R1 pricing, annual revenue could exceed $200 million, though current monetization strategies limit actual earnings.

• Lower-Cost AI Model Training: The company developed its models using Nvidia H800 chips, significantly less advanced than those used by U.S. competitors like OpenAI, yet still achieved competitive results.

• AI Industry Profitability Questions: DeepSeek’s approach raises doubts about whether massive infrastructure investments are necessary, as its low-cost development strategy challenges prevailing industry norms.

Go Deeper -> DeepSeek Claims ‘Theoretical’ Profit Margins of 545% – TechCrunch

China’s DeepSeek Claims Theoretical Cost-Profit Ratio of 545% Per Day – Reuters

DeepSeek Reveals Theoretical Margin on Its AI Models Is 545% – Yahoo Finance