DeepSeek shows power of V3, R1 models with theoretical 545% profit margin

Chinese tech start-up DeepSeek concluded its daily technical project in “Open Source Week” with a bold claim: its online inference services generated an extraordinary 545 per cent profit margin during a 24-hour run, thanks to advanced technological optimisations.

DeepSeek reported an average node occupancy of 226.75 across its V3 and R1 inference models from noon Beijing time on February 27, it said in a post on Saturday. Each node, comprising eight Nvidia H800 GPUs (graphics processing units) leased at a cost of US$2 per GPU per hour, resulted in a total operational cost of US$87,072.

Over the same time, the models processed a combined 608 billion input tokens and 168 billion output tokens, including user requests via web, mobile apps, and application programming interfaces (APIs). That would equal US$562,027 in revenue, if charged using DeepSeek R1’s pricing model, for a theoretical 545 per cent gain.

Nodes represent individual computational units handling tasks, while node occupancy shows their usage efficiency during inference requests. A token in artificial intelligence (AI) refers to a fundamental unit of data processed by the algorithm. DeepSeek bills users based on the total input and output tokens processed by its models.

DeepSeek, based in Hangzhou in eastern Zhejiang province, took the tech world by storm this year after unveiling its superior AI models built at a fraction of the costs incurred by its bigger US rivals. Its revelation helped wipe off billions from the market value of US tech stocks including Nvidia, and induced a bull run in Chinese tech stocks in Hong Kong.

01:20

China’s Alibaba releases new AI model, said to outperform competitors Deepseek and OpenAI’s GPT-4o

China’s Alibaba releases new AI model, said to outperform competitors Deepseek and OpenAI’s GPT-4o

In its posts on GitHub and Zhihu Chinese Q&A platform on Saturday, DeepSeek clarified that the cost and revenue were “purely theoretical.” Actual revenue could be lower because of its cheaper pricing for V3 model, substantial off-peak discounts on both models, and free-to-use chatbot services.