DeepSeek's IPO: Two Realities and One Goal

The market never lacks IPO rumors of popular enterprises.

From SHEIN in the past to Unitree Robotics, the latest baton has now been passed to DeepSeek.

Recently, multiple media outlets have reported that DeepSeek has commissioned CITIC Securities to prepare for its IPO on the Sci-Tech Innovation Board, and CITIC Securities has already contacted DeepSeek and entered the due diligence phase, though the two sides have not yet signed a formal listing counseling agreement.

Three years ago, Liang Wenfeng, founder of DeepSeek, once said: "No financing, no commercialization, no roadshows". However, in April this year, DeepSeek launched its first round of external financing; in June, the first round of financing was completed, raising 500 billion yuan with a post-money valuation exceeding 3.5 trillion yuan; in July, the second round of financing was suspended; in August, the financing restarted, with a planned fundraising of about 500 billion yuan and a post-money valuation of 5 trillion yuan; in September, DeepSeek was exposed to be preparing for an IPO.

For a popular AI company like DeepSeek, its competitiveness in the AI field has gone far beyond the scope of technology itself. If it needs capital, a large number of institutions in the primary market are willing to invest, which can be said to offer as much capital as it wants. So why did it choose to go public via IPO?

If we look back at the speech Liang Wenfeng gave at the investor exchange meeting (hereinafter referred to as the "Exchange Meeting") in July this year, we may gain a new understanding of two realities and one goal of DeepSeek behind this IPO rumor.

The first realistic problem is how to maintain the stability of the organizational team, and an IPO can realize the monetization of equity incentives.

At the exchange meeting in July, Liang Wenfeng mentioned that the core interest of the company at present is to maintain the stability of the team, which is also a huge challenge it faces. With the recent completion of a new round of financing, employees hold a relatively large number of options with high value, which greatly eliminates this risk.

For a long time, DeepSeek has focused on technology R&D and has not built a market-oriented option incentive system, making it difficult to match the high-salary competition pattern of the industry. Its main feature used to be the "small but excellent" research model. However, with the great leap forward of the entire AI industry, as well as the emergence of more demands for improving its own model capabilities, product development and commercialization, system engineering capabilities have become more important in future competition, which also means that the original 300-person team faces further upgrading in terms of personnel scale and structure.

At the same time, the problem of core talent loss in DeepSeek has become increasingly prominent. From the second half of 2025 to the beginning of 2026, many key technical backbones of DeepSeek have successively moved to competitors with stronger capital strength. Meanwhile, the competition among large tech companies for top AI researchers has pushed talent costs to a new high.

This is also why we can clearly feel that DeepSeek has been increasing its recruitment efforts this year. For example, in June, DeepSeek stated that it is working to expand the scale of all departments by at least double, covering seven major sectors including full-stack development and algorithms, core AI system R&D, operation and maintenance, products, model data strategy, deep learning research and functional departments, opening a total of 33 job categories, with work locations including Beijing and Hangzhou. Recently, the company's Harness team has released about 150 new recruitment positions to the public, covering multiple core technical posts such as server-side development and Agent elastic computing R&D. This special targeted recruitment expansion is highly targeted, focusing on solving the system bottlenecks brought by the current rapid business growth.

Liang Wenfeng hopes to raise funds through the IPO to provide more competitive incentives for core employees and researchers, which will become an effective means to retain key talents and realize core interests.

The second realistic problem is how to grow stronger and bigger with the help of capital, so that DeepSeek can hit a trillion-yuan market value and become the "top player of large models" in China.

From a global perspective, there is still a large gap in valuation between the leading AI companies in China and the United States. For example, OpenAI's valuation has reached 1 trillion US dollars, and Anthropic's maximum valuation may reach 2 trillion US dollars. However, the overall valuations of China's listed players including Zhipu AI and MiniMax, as well as Moonshot AI which has just been rumored to submit its listing application, do not exceed 400 billion Hong Kong dollars. From this point of view, China is in great need of a landmark AI enterprise, and this role has obviously gradually fallen on DeepSeek.

For DeepSeek, it has now grown to a stage of "too large in scale and too high in capital investment", and can no longer be just an affiliated project of High-Flyer Quant. The Information disclosed that in the first seven months of this year, DeepSeek spent about 11 billion yuan on AI infrastructure-related expenditures, including renting servers equipped with AI chips, purchasing chips and other computing devices. This part of expenditure was about 1.2 billion yuan in 2025. If DeepSeek wants to continue to accelerate the expansion of its infrastructure, it must rely on the power of capital to lock in computing resources in advance.

In addition to obtaining more funds for development through the IPO, DeepSeek is also increasing its commercialization efforts to support future valuation with continuous revenue. According to The Information, DeepSeek's revenue in the first seven months of this year was about 475 million yuan, an increase of about ten times over the full year of 2025; during the same period, the gross profit margin of its API business reached 82.9%.

At the same time, DeepSeek has also taken the initiative to adjust its commercialization strategy. According to its official announcement, the company has adjusted the price of its V4 series APIs since August 16, and introduced a peak-valley pricing mechanism, where the price during off-peak hours is half of that during peak hours. Compared with the early strategy of expanding developer coverage through extremely low prices, this change means that DeepSeek has begun to find a new balance between call volume, computing power utilization and revenue.

Finally, let's look at DeepSeek's goal, which determines what business form and valuation logic the market will use to view this company.

Recently, discussions about AGI have also been extremely heated. On September 3, after OpenAI President Greg Brockman released the new model GPT-6 Astra, he dropped a sentence: "Welcome to the AGI era." Three days later, Jensen Huang spoke more directly on social platforms: "It only took four years from ChatGPT to o1, and then to GPT-6 Astra. AGI has arrived."

In Liang Wenfeng's view, the only main line of DeepSeek is to develop AGI, which is also the ultimate goal of the company. As Liang Wenfeng said before, in the large model industry, last year was a race to seize C-end traffic, and this year is a race to grab to-B revenue, every AI enterprise wants to participate in it. But for DeepSeek, these are just by-products, and it has not spent a lot of energy on these things, because its original intention is to pursue AGI.

This means that in the listing process, many people may care about specific figures such as revenue performance and user growth, but these are not what DeepSeek cares about. DeepSeek's goal is to seize the next-generation operating system-level platform with AGI and become the infrastructure of the national AI strategy. This narrative is grand, but its value is unique.

Up to now, DeepSeek has not officially announced its IPO plan, and the specific listing time, fundraising scale and final valuation have not been determined.

But in any case, from now on, we need to re-understand DeepSeek. Any listing dynamics of DeepSeek will dominate the headlines of public opinion in the future.

This article is from the WeChat Official Account "Business New Research Institute", author: Xiao Yi, published with authorization from 36Kr.