Summary of the Key News in Plain Language
DeepSeek, a leading domestic AI company that gained fame for developing a top-tier large model with just over 5 million US dollars, has officially announced the initiation of its listing process. It has chosen CITIC Securities as its sponsor and aims to list on the STAR Market of the Shanghai Stock Exchange. At the same time, the company is also pursuing a significant private fundraising round of 50 billion yuan. This development shatters the myth that large models can be created by cutting costs. DeepSeek, which is already among the world's top players, has entered a new phase where it needs to invest heavily to build long-term competitiveness. It aims to secure long-term funding to acquire computing power and build a strong core team, while avoiding being constrained by short-term capital. This move coincides with the Shanghai Stock Exchange’s recent introduction of a special listing pathway for large model companies, making DeepSeek the first domestic company to target the A-share market.
---
Detailed Analysis
1. Don’t Be Misled by the Claim of “5.57 Million US Dollars for a Large Model”
There was a widespread misconception that DeepSeek spent only 5.57 million US dollars to develop its V3 large model, leading many to think that developing large models was a highly profitable endeavor. However, this figure is significantly inflated. The money mentioned refers solely to the cost of GPU electricity and computing power rental for the final stage of training, which is akin to the registration fee for the last exam to get into Tsinghua University—just a few dozen yuan. This does not account for the years of trial and error, algorithm adjustments, data collection, and numerous small-scale experiments by the team, nor does it include ongoing expenses. DeepSeek’s model is now used by hundreds of thousands of developers and millions of users daily. Maintaining the servers and providing stable services requires a large engineering team. To develop the next generation of models, the company needs to purchase tens of thousands of H800 GPUs, each costing over a hundred thousand yuan. The total cost is in the tens of billions of yuan, and the previous funding was far from sufficient.
2. Raising Private Capital While Pursuing an IPO
DeepSeek’s goal is not just to acquire funds for GPUs but also to retain its talent. Many wonder why the company is raising another 50 billion yuan and going public when it just raised 7.4 billion yuan in June. The real purpose is to address the severe shortage of top AI engineers. In the AI industry, experienced engineers are in high demand, with annual salaries of up to 2 million yuan. For companies that were not listed, equity was an unvalued asset. However, after going public, the value of equity is clearly reflected in the stock price. Equity granted to key employees is understood by everyone, providing a tangible incentive to stay with the company, which is more important than simply buying more GPUs.
3. Selective Financing and Investor Requirements
DeepSeek’s new financing terms are quite stringent: investors must sign a five-year lock-up agreement and do not have voting rights or the ability to interfere with company operations. Despite this, capital from all sources is competing for shares. The reason for such strict conditions is that the R&D process in the AI industry is different from traditional businesses. Breakthroughs in cutting-edge technology may not generate revenue for 3-5 years. If shareholders of a listed company constantly pressure management for immediate profits, teams would be reluctant to invest in long-term, less profitable projects. DeepSeek wants to use its funds to maintain control over its R&D and avoid being distracted by short-term financial goals.
4. Choosing the STAR Market
DeepSeek has chosen the STAR Market because of a policy change in June 2026 that allows AI large model companies to list without requiring years of profitability or high annual revenue. This new rule opens up opportunities for companies that are still investing heavily in R&D. By listing on the STAR Market, DeepSeek can avoid the need to go public in other regions with more complex procedures.
5. Evaluating Future Value
When DeepSeek submits its prospectus, it will be touted as the “first AI large model stock.” Investors should focus on a crucial metric: how many users are actually paying for its services. While open-source large models are common and many people use them for free, the actual revenue generated by DeepSeek’s model is crucial. If the service fees do not cover the costs, even if the model performs exceptionally well globally, it will merely be a costly technology toy, not a sustainable business. Whether DeepSeek can make money through its technology will be revealed soon.
In summary, DeepSeek’s move towards listing is a reflection of its need to invest in long-term growth and retain its core team, rather than merely acquiring more hardware. The choice of the STAR Market is strategic, taking advantage of a regulatory change that facilitates the listing of AI large model companies. The true value of the company will be determined by its ability to generate revenue through its technology.