DeepSeek revealed that its AI models V3 and R1 achieve a theoretical daily cost-profit ratio of up to 545% through profitability estimates. The company issued warnings about actual revenue being substantially lower than estimated, but DeepSeek provided its first disclosure of financial margins from AI inference operations.
The profitability estimates released by DeepSeek have created further market instability in global AI sectors following the January decline of U.S. AI stocks because of R1 and V3 model expansion. DeepSeek’s announcement that it spent under $6 million on model training chips caused investors to sell off shares because it demonstrated lower costs than U.S. competitors and OpenAI’s billions of dollars in high-end hardware investments.
DeepSeek’s models use Nvidia H800 chips as their processing unit which proves less powerful compared to the hardware employed by OpenAI and Western AI companies. DeepSeek’s recent disclosure reveals that daily revenue from H800 chip rentals amounts to $562,027 while costing only $87,072 per day, which indicates over $200 million in theoretical annual revenue.
Real earnings at DeepSeek fall short of initial projections because of multiple influencing elements. V3 model usage from the company comes at a lower price point, and most of their services have no cost, and their pricing structure adjusts according to market demand. The company’s claims oppose the typical notion that AI progress demands significant financial investments thus transforming current business models.
The market-disrupting, cost-efficient AI solutions from DeepSeek create rising pressure for U.S. firms to defend their expensive investments in advanced chips and infrastructure. The company’s strategy demonstrates China’s expanding role in AI advancement, which leads to heightened global market competition in this field.
