Chinese artificial intelligence startup DeepSeek has once again jolted the global AI market by disclosing critical cost and revenue data about its popular V3 and R1 models, claiming a theoretical cost-profit ratio of up to 545% per day.
However, the Hangzhou-based company also cautioned that actual revenue is significantly lower due to several mitigating factors, including free services and variable pricing.
This is the first time DeepSeek has publicly shared any insight into its profit margins from “inference” tasks, a phase in AI deployment where trained models execute tasks such as predictions and chatbot interactions. The company revealed these figures through a GitHub post on Saturday, giving investors and analysts a closer look at the financial dynamics of its models, which have gained global popularity through web and app-based chatbots.
Register for Tekedia Mini-MBA edition 17 (June 9 – Sept 6, 2025) today for early bird discounts. Do annual for access to Blucera.com.
Tekedia AI in Business Masterclass opens registrations.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register to become a better CEO or Director with Tekedia CEO & Director Program.
DeepSeek’s Rise As A Cost-Efficient Alternative
DeepSeek’s rise in the AI industry has been nothing short of disruptive. The company first turned heads earlier this year when it revealed that it had spent less than $6 million on chips used to train its models. This is a stark contrast to the billions of dollars that U.S. rivals like OpenAI have invested in cutting-edge hardware. Moreover, DeepSeek relies on Nvidia’s H800 chips, which are significantly less powerful than the hardware deployed by American AI firms.
This development not only questioned the efficiency of U.S. AI firms’ spending strategies but also caused a sell-off in AI stocks outside China. Many investors began to rethink the sustainability of high-cost approaches, especially as DeepSeek’s models, despite using less advanced chips, managed to deliver competitive performance.
The company’s approach has exposed a potential vulnerability in the business models of Western AI firms, which are built on heavy investments in expensive technology. DeepSeek has proven that cost-efficiency can be a viable path to profitability, challenging the notion that only top-tier hardware can support successful AI deployments.
The Numbers Behind DeepSeek’s Model
DeepSeek’s financial snapshot offered a glimpse into its business model, with the rental cost of one Nvidia H800 chip estimated at $2 per hour. According to the company, the total daily inference cost for its V3 and R1 models is $87,072, while the theoretical daily revenue could reach $562,027. If this potential were fully realized, the models could generate just over $200 million in annual revenue, boasting a cost-profit ratio of 545%.
However, DeepSeek was quick to clarify that these numbers represent an ideal scenario. The real-world revenue is substantially lower due to several factors: the lower operational cost of the V3 model, the limited monetization of its services, and discounted pricing during off-peak hours. Furthermore, while some services generate income, many remain free on web and app platforms, limiting profitability.
Censorship Concerns: A Major Roadblock Outside China
Despite its impressive financial model and cost-efficiency, DeepSeek faces a significant barrier to international expansion—censorship. Unlike Western AI models, which are often built on open data sets and trained with a focus on free expression, DeepSeek’s models are required to adhere to strict Chinese censorship laws.
For instance, its chatbots and AI tools are programmed to filter out politically sensitive topics, avoiding discussions on issues like Tiananmen Square, Hong Kong protests, and Taiwan’s sovereignty. This built-in censorship has made the models less attractive to international developers and global enterprises that value unrestricted access to information.
In regions where freedom of speech and openness are crucial—such as the United States, Europe, and parts of Asia—DeepSeek’s censored outputs are seen as a liability, hindering its adoption. Industry analysts have pointed out that developers outside China might be reluctant to integrate DeepSeek’s models into their systems if it means compromising on data freedom.
This censorship issue could impact DeepSeek’s profitability, especially as international markets account for a significant share of AI companies’ revenue. However, DeepSeek has a potential safety net in the Chinese market, which is large and lucrative enough to support sustained growth.
China’s massive domestic market could serve as a buffer for DeepSeek as it navigates international challenges. The country’s booming AI ecosystem, combined with government support for local tech firms, provides a fertile ground for DeepSeek to thrive.
With the Chinese government actively encouraging the development of homegrown technologies, DeepSeek could focus on monetizing its services locally, potentially avoiding the pitfalls of global competition. Moreover, China’s tightly regulated internet space means that censorship compliance might not be as much of a hindrance domestically as it is abroad.