DeepSeek Challenges U.S. AI Firms on Spending.

DeepSeek AI has shared cost and revenue data for its DeepSeek-V3 and DeepSeek-R1 models, revealing a theoretical daily cost-profit ratio of 545%. The Hangzhou-based AI startup estimated that running inference tasks for its models costs $87,072 per day while generating $562,027 in revenue. However, it cautioned that actual revenue is significantly lower due to monetization limits and lower off-peak usage costs.

This is the first time DeepSeek has disclosed details about its profit margins from inference tasks, which involve using trained AI models to make predictions. The announcement comes after its chatbot-powered models gained popularity, causing AI stocks outside China to drop in January.

The company also revealed that it spent under $6 million on training chips, significantly less than what U.S. firms like OpenAI have invested. NVIDIA’s H800 chips, used by DeepSeek, are less powerful than those available to U.S. competitors. This raises questions about whether massive AI chip investments by Western firms are justified.

With growing AI competition, DeepSeek’s cost-efficient approach challenges traditional spending models. Investors remain watchful as the AI landscape evolves.