DeepSeek reveals theoretical margin on its AI models is 545%

By Saritha Rai

Chinese artificial intelligence phenomenon DeepSeek revealed some financial numbers on Saturday, saying its “theoretical” profit margin could be more than five times costs, peeling back a layer of the secrecy that shrouds business models in the AI industry.

The 20-month-old startup that rattled Silicon Valley with its innovative and inexpensive approach to building AI models, said on X its V3 and R1 models’ cost of inferencing to sales during a 24-hour-period on the last day of February put profit margins at 545 per cent.

Inferencing refers to the computing power, electricity, data storage and other resources needed to make AI models work in real time.

However, DeepSeek added a disclaimer in details it provided on GitHub, saying its actual revenues are substantially lower for various reasons, including the fact that only a small set of its services are monetized and it offers discounts during off-peak hours. Nor do the costs factor in all the R&D and training expenses for building its models.

While the eye-popping profit margins are therefore hypothetical, the reveal comes at a time when profitability of AI startups and their models is a hot topic among technology investors.

Companies from OpenAI Inc. to Anthropic PBC are experimenting with various revenue models, from subscription-based to charging for usage to collecting licensing fees, as they race to build ever more sophisticated AI products. But investors are questioning these business models and their return on investment, opening a debate on the feasibility of reaching profitability any day soon.

The Hangzhou-based startup said Saturday on X that its online service had a “cost profit margin of 545 per cent” and gave an overview of its operations including how it optimized computing power by balancing load — that is managing traffic so that work is evenly distributed between multiple servers and data centers. DeepSeek said it innovated to optimize the amount of data processed by the AI model in a given time period, and managed latency — the wait time between a user submitting a query and receiving the answer.

In a series of unusual steps beginning early this week, the startup, which has espoused open-source AI, surprised many in the industry by sharing some key innovations and data underpinning its models, in contrast to the proprietary approach of its biggest US rivals like OpenAI.