DeepSeek Claims Theoretical 545% Daily Profit, Raising Eyebrows

Chinese AI startup DeepSeek has made a bold claim about its cost-profit ratio, reporting a theoretical 545% daily profit for its V3 and R1 models, Reuters reports. This comes as the company releases financial data related to its AI models for the first time, shedding light on a key area of investor concern.

DeepSeek's revelation is particularly noteworthy given the recent sell-off in US AI stocks. Investors were rattled by claims that DeepSeek spent significantly less on training chips than its US counterparts, raising questions about the cost competitiveness of OpenAI and other US AI firms.

The company's latest statement further underscores this cost disparity. DeepSeek claims that its V3 and R1 models, which power popular chatbots, have a theoretical daily inference cost of $87,072, based on a rental cost of $2 per hour for an Nvidia H800 chip. However, the models are estimated to generate $562,027 in daily revenue, resulting in a theoretical cost-profit ratio of 545%. This equates to just over $200 million in annual revenue.

However, DeepSeek cautions that its actual revenue is "substantially lower" due to several factors. The cost of using the V3 model is lower than the R1 model, only some services are currently monetized, and developers pay less during off-peak hours.

The technology sector has been closely watching DeepSeek's cost structure, particularly in light of the massive investments being poured into AI development, especially in the US. The company's claims of significantly lower training chip costs and its latest profit projections could further intensify scrutiny of US AI firms' spending strategies and their ability to compete on cost.